diff --git a/scripts/performance/README.md b/scripts/performance/README.md
index b5b2c6d854..8f7b2f28ec 100644
--- a/scripts/performance/README.md
+++ b/scripts/performance/README.md
@@ -239,11 +239,11 @@ Mounting cached files is not enough by itself. If `HF_HUB_OFFLINE` remains `0`,
 
 ##### Performance arguments
 
-- `-g/--gpu`: Target GPU type (`h100`, `b200`, `gb200`, `gb300`, `b300`).
+- `-g/--gpu`: Target GPU type (`h100`, `b200`, `gb200`, `gb300`, `b300`, `vrnvl72`). `vr200` is a deprecated alias for `vrnvl72`.
 - `-c/--compute_dtype`: Compute precision (`bf16`, `fp8_cs`, `fp8_mx`, `fp8_sc`, `nvfp4`). Default `bf16`.
 - `-vb/--enable_vboost`: Enable VBoost (tensor core power steering). Pass `true` or `false`. Disabled by default.
 - `-lgc/--lock_gpu_freq`: Lock GPU graphics clock to a fixed frequency in MHz (e.g. `1200`). Used for silicon simulation correlation studies. Disabled by default.
-- `-lmc/--peak_mem_clk`: Lock GPU memory clock to a fixed peak frequency in MHz (e.g. `2600`). Used for silicon simulation correlation studies. Defaults to `4752` MHz for VR200 and is disabled by default for other GPUs. Pass `-lmc -1` or `--peak_mem_clk -1` to disable the VR200 default.
+- `-lmc/--peak_mem_clk`: Lock GPU memory clock to a fixed peak frequency in MHz (e.g. `2600`). Used for silicon simulation correlation studies. Defaults to `4752` MHz for VRNVL72 and is disabled by default for other GPUs. Pass `-lmc -1` or `--peak_mem_clk -1` to disable the VRNVL72 default.
 - `-en/--enable_nsys`: Enable Nsight Systems profiling. Disabled by default.
 - `-pyp/--pytorch_profiler`: Enable PyTorch profiler. Pass `true` or `false`. Disabled by default.
 - `--profiling_start_step`: Defines start step for profiling. Default `10`.
diff --git a/scripts/performance/argument_parser.py b/scripts/performance/argument_parser.py
index 235fb43411..89f43a6d1d 100644
--- a/scripts/performance/argument_parser.py
+++ b/scripts/performance/argument_parser.py
@@ -33,10 +33,14 @@
     "b300": 8,
     "gb200": 4,
     "gb300": 4,
-    "vr200": 4,
+    "vrnvl72": 4,
     "r100": 1,
 }
 
+# Deprecated --gpu names and their replacements. When the deprecation window ends, drop an entry together with
+# its counterpart in LEGACY_HARDWARE_ALIASES (scripts/training/recipe_metadata.py).
+LEGACY_GPU_ALIASES = {"vr200": "vrnvl72"}
+
 
 def list_of_strings(arg):
     """Split a comma-separated string into a list of substrings."""
@@ -80,6 +84,16 @@ def lower_str(arg):
     return arg.lower()
 
 
+def canonical_gpu(arg: str) -> str:
+    """Map a deprecated GPU name to its replacement before argparse checks the choices."""
+    if arg in LEGACY_GPU_ALIASES:
+        # Every rank re-parses the forwarded argv, so only the launcher and global rank 0 warn.
+        if os.environ.get("RANK", os.environ.get("SLURM_PROCID", "0")) == "0":
+            logger.warning("--gpu %s is deprecated; use --gpu %s.", arg, LEGACY_GPU_ALIASES[arg])
+        return LEGACY_GPU_ALIASES[arg]
+    return arg
+
+
 def bool_arg(arg):
     """Convert a string CLI value to a boolean."""
     if arg.lower() in ["true", "1", "t", "yes", "y"]:
@@ -787,9 +801,9 @@ def parse_cli_args():
     performance_args.add_argument(
         "-g",
         "--gpu",
-        type=str,
+        type=canonical_gpu,
         choices=NUM_GPUS_PER_NODE_MAP.keys(),
-        help="Target gpu type.",
+        help="Target gpu type. vr200 is a deprecated alias for vrnvl72.",
         required=True,
     )
     performance_args.add_argument(
@@ -831,8 +845,8 @@ def parse_cli_args():
         "--peak_mem_clk",
         help="Lock GPU memory clock to the specified peak frequency in MHz via "
         "`sudo nvidia-smi -lmc <freq>,<freq>`. Runs once per node before training. "
-        "Defaults to 4752 MHz for VR200 and is disabled by default for other GPUs. "
-        "Pass `-lmc -1` or `--peak_mem_clk -1` to disable the VR200 default.",
+        "Defaults to 4752 MHz for VRNVL72 and is disabled by default for other GPUs. "
+        "Pass `-lmc -1` or `--peak_mem_clk -1` to disable the VRNVL72 default.",
         type=int,
         required=False,
         default=None,
diff --git a/scripts/performance/dump_perf_configs.py b/scripts/performance/dump_perf_configs.py
index 4c31f82cfc..f7a0295a89 100644
--- a/scripts/performance/dump_perf_configs.py
+++ b/scripts/performance/dump_perf_configs.py
@@ -65,10 +65,10 @@
     ("llama", "llama3_8b", "pretrain", 8, "gb200", "nvfp4"),
     ("llama", "llama3_8b", "pretrain", 32, "gb200", "bf16"),
     ("llama", "llama3_8b", "pretrain", 32, "gb200", "fp8_cs"),
-    ("llama", "llama3_8b", "pretrain", 8, "vr200", "bf16"),
-    ("llama", "llama3_8b", "pretrain", 8, "vr200", "fp8_cs"),
-    ("llama", "llama3_8b", "pretrain", 8, "vr200", "fp8_mx"),
-    ("llama", "llama3_8b", "pretrain", 8, "vr200", "nvfp4"),
+    ("llama", "llama3_8b", "pretrain", 8, "vrnvl72", "bf16"),
+    ("llama", "llama3_8b", "pretrain", 8, "vrnvl72", "fp8_cs"),
+    ("llama", "llama3_8b", "pretrain", 8, "vrnvl72", "fp8_mx"),
+    ("llama", "llama3_8b", "pretrain", 8, "vrnvl72", "nvfp4"),
     ("llama", "llama3_8b", "pretrain", 8, "b300", "bf16"),
     ("llama", "llama3_8b", "pretrain", 8, "b300", "fp8_cs"),
     ("llama", "llama3_8b", "pretrain", 8, "b300", "fp8_mx"),
@@ -172,10 +172,10 @@
     ("deepseek", "deepseek_v3", "pretrain", 256, "gb200", "fp8_cs"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "gb200", "fp8_mx"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "gb200", "nvfp4"),
-    ("deepseek", "deepseek_v3", "pretrain", 256, "vr200", "bf16"),
-    ("deepseek", "deepseek_v3", "pretrain", 256, "vr200", "fp8_cs"),
-    ("deepseek", "deepseek_v3", "pretrain", 256, "vr200", "fp8_mx"),
-    ("deepseek", "deepseek_v3", "pretrain", 256, "vr200", "nvfp4"),
+    ("deepseek", "deepseek_v3", "pretrain", 256, "vrnvl72", "bf16"),
+    ("deepseek", "deepseek_v3", "pretrain", 256, "vrnvl72", "fp8_cs"),
+    ("deepseek", "deepseek_v3", "pretrain", 256, "vrnvl72", "fp8_mx"),
+    ("deepseek", "deepseek_v3", "pretrain", 256, "vrnvl72", "nvfp4"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "b300", "bf16"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "b300", "fp8_cs"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "b300", "fp8_mx"),
@@ -255,8 +255,8 @@
     ("kimi", "kimi_k2", "pretrain", 256, "gb200", "bf16"),
     ("kimi", "kimi_k2", "pretrain", 256, "gb200", "fp8_cs"),
     ("kimi", "kimi_k2", "pretrain", 256, "gb200", "fp8_mx"),
-    ("kimi", "kimi_k2", "pretrain", 256, "vr200", "bf16"),
-    ("kimi", "kimi_k2", "pretrain", 256, "vr200", "fp8_mx"),
+    ("kimi", "kimi_k2", "pretrain", 256, "vrnvl72", "bf16"),
+    ("kimi", "kimi_k2", "pretrain", 256, "vrnvl72", "fp8_mx"),
     ("kimi", "kimi_k2", "pretrain", 256, "b300", "bf16"),
     ("kimi", "kimi_k2", "pretrain", 256, "b300", "fp8_cs"),
     ("kimi", "kimi_k2", "pretrain", 256, "b300", "fp8_mx"),
@@ -321,9 +321,9 @@
     ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "gb300", "nvfp4"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 72, "gb300", "nvfp4"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 512, "gb300", "fp8_mx"),
-    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vr200", "nvfp4"),
-    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vr200", "fp8_mx"),
-    ("gpt_oss", "gpt_oss_20b", "pretrain", 64, "vr200", "nvfp4"),
+    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vrnvl72", "nvfp4"),
+    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vrnvl72", "fp8_mx"),
+    ("gpt_oss", "gpt_oss_20b", "pretrain", 64, "vrnvl72", "nvfp4"),
     # GPT-OSS 120B
     ("gpt_oss", "gpt_oss_120b", "pretrain", 64, "gb300", "bf16"),
     ("gpt_oss", "gpt_oss_120b", "pretrain", 64, "gb300", "fp8_mx"),
@@ -408,10 +408,10 @@
     ("qwen_vl", "qwen35_vl_397b_a17b", "pretrain", 256, "h100", "bf16"),
     ("qwen_vl", "qwen35_vl_397b_a17b", "pretrain", 256, "h100", "fp8_cs"),
     # Additional flattened coverage
-    ("deepseek", "deepseek_v3", "pretrain", 128, "vr200", "bf16"),
-    ("deepseek", "deepseek_v3", "pretrain", 128, "vr200", "fp8_cs"),
-    ("deepseek", "deepseek_v3", "pretrain", 128, "vr200", "fp8_mx"),
-    ("deepseek", "deepseek_v3", "pretrain", 128, "vr200", "nvfp4"),
+    ("deepseek", "deepseek_v3", "pretrain", 128, "vrnvl72", "bf16"),
+    ("deepseek", "deepseek_v3", "pretrain", 128, "vrnvl72", "fp8_cs"),
+    ("deepseek", "deepseek_v3", "pretrain", 128, "vrnvl72", "fp8_mx"),
+    ("deepseek", "deepseek_v3", "pretrain", 128, "vrnvl72", "nvfp4"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "gb300", "fp8_mx", "large_scale"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "gb200", "fp8_mx", "large_scale"),
     ("deepseek", "deepseek_v3", "pretrain", 256, "b300", "fp8_mx", "large_scale"),
@@ -421,18 +421,18 @@
     ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "gb200", "nvfp4"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "b300", "fp8_mx"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "b300", "nvfp4"),
-    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vr200", "fp8_mx"),
-    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vr200", "nvfp4"),
+    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vrnvl72", "fp8_mx"),
+    ("gpt_oss", "gpt_oss_20b", "pretrain", 8, "vrnvl72", "nvfp4"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 64, "b300", "fp8_mx"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 64, "b300", "nvfp4"),
-    ("gpt_oss", "gpt_oss_20b", "pretrain", 64, "vr200", "nvfp4"),
+    ("gpt_oss", "gpt_oss_20b", "pretrain", 64, "vrnvl72", "nvfp4"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 72, "gb300", "nvfp4"),
     ("gpt_oss", "gpt_oss_20b", "pretrain", 72, "gb200", "nvfp4"),
-    ("gpt_oss", "gpt_oss_120b", "pretrain", 64, "vr200", "bf16"),
-    ("gpt_oss", "gpt_oss_120b", "pretrain", 64, "vr200", "fp8_mx"),
-    ("llama", "llama3_70b", "pretrain", 64, "vr200", "bf16"),
-    ("llama", "llama3_70b", "pretrain", 64, "vr200", "fp8_mx"),
-    ("llama", "llama3_70b", "pretrain", 64, "vr200", "nvfp4"),
+    ("gpt_oss", "gpt_oss_120b", "pretrain", 64, "vrnvl72", "bf16"),
+    ("gpt_oss", "gpt_oss_120b", "pretrain", 64, "vrnvl72", "fp8_mx"),
+    ("llama", "llama3_70b", "pretrain", 64, "vrnvl72", "bf16"),
+    ("llama", "llama3_70b", "pretrain", 64, "vrnvl72", "fp8_mx"),
+    ("llama", "llama3_70b", "pretrain", 64, "vrnvl72", "nvfp4"),
     ("llama", "llama31_405b", "pretrain", 128, "gb300", "bf16"),
     ("llama", "llama31_405b", "pretrain", 128, "gb300", "fp8_cs"),
     ("llama", "llama31_405b", "pretrain", 128, "gb300", "fp8_mx"),
@@ -447,19 +447,19 @@
     ("llama", "llama31_405b", "pretrain", 256, "b300", "nvfp4"),
     ("llama", "llama31_405b", "pretrain", 256, "b200", "fp8_mx"),
     ("llama", "llama31_405b", "pretrain", 256, "b200", "nvfp4"),
-    ("llama", "llama31_405b", "pretrain", 256, "vr200", "bf16"),
-    ("llama", "llama31_405b", "pretrain", 256, "vr200", "fp8_mx"),
-    ("llama", "llama31_405b", "pretrain", 256, "vr200", "nvfp4"),
-    ("nemotronh", "nemotron_3_nano", "pretrain", 8, "vr200", "bf16"),
-    ("nemotronh", "nemotron_3_nano", "pretrain", 8, "vr200", "fp8_mx"),
-    ("nemotronh", "nemotron_3_nano", "pretrain", 8, "vr200", "nvfp4"),
-    ("nemotronh", "nemotron_3_super", "pretrain", 64, "vr200", "bf16"),
-    ("nemotronh", "nemotron_3_super", "pretrain", 64, "vr200", "fp8_mx"),
-    ("nemotronh", "nemotron_3_super", "pretrain", 64, "vr200", "nvfp4"),
-    ("nemotronh", "nemotron_3_ultra", "pretrain", 256, "vr200", "fp8_mx"),
-    ("nemotronh", "nemotron_3_ultra", "pretrain", 256, "vr200", "nvfp4"),
-    ("qwen", "qwen3_30b_a3b", "pretrain", 8, "vr200", "bf16"),
-    ("qwen", "qwen3_30b_a3b", "pretrain", 8, "vr200", "fp8_mx"),
+    ("llama", "llama31_405b", "pretrain", 256, "vrnvl72", "bf16"),
+    ("llama", "llama31_405b", "pretrain", 256, "vrnvl72", "fp8_mx"),
+    ("llama", "llama31_405b", "pretrain", 256, "vrnvl72", "nvfp4"),
+    ("nemotronh", "nemotron_3_nano", "pretrain", 8, "vrnvl72", "bf16"),
+    ("nemotronh", "nemotron_3_nano", "pretrain", 8, "vrnvl72", "fp8_mx"),
+    ("nemotronh", "nemotron_3_nano", "pretrain", 8, "vrnvl72", "nvfp4"),
+    ("nemotronh", "nemotron_3_super", "pretrain", 64, "vrnvl72", "bf16"),
+    ("nemotronh", "nemotron_3_super", "pretrain", 64, "vrnvl72", "fp8_mx"),
+    ("nemotronh", "nemotron_3_super", "pretrain", 64, "vrnvl72", "nvfp4"),
+    ("nemotronh", "nemotron_3_ultra", "pretrain", 256, "vrnvl72", "fp8_mx"),
+    ("nemotronh", "nemotron_3_ultra", "pretrain", 256, "vrnvl72", "nvfp4"),
+    ("qwen", "qwen3_30b_a3b", "pretrain", 8, "vrnvl72", "bf16"),
+    ("qwen", "qwen3_30b_a3b", "pretrain", 8, "vrnvl72", "fp8_mx"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 64, "gb300", "bf16"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 64, "gb300", "fp8_cs"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 64, "gb300", "fp8_mx"),
@@ -476,9 +476,9 @@
     ("qwen", "qwen3_235b_a22b", "pretrain", 64, "b200", "fp8_cs"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 64, "b200", "fp8_mx"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 64, "b200", "nvfp4"),
-    ("qwen", "qwen3_235b_a22b", "pretrain", 256, "vr200", "bf16"),
-    ("qwen", "qwen3_235b_a22b", "pretrain", 256, "vr200", "fp8_mx"),
-    ("qwen", "qwen3_235b_a22b", "pretrain", 256, "vr200", "nvfp4"),
+    ("qwen", "qwen3_235b_a22b", "pretrain", 256, "vrnvl72", "bf16"),
+    ("qwen", "qwen3_235b_a22b", "pretrain", 256, "vrnvl72", "fp8_mx"),
+    ("qwen", "qwen3_235b_a22b", "pretrain", 256, "vrnvl72", "nvfp4"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 256, "gb300", "fp8_mx", "large_scale"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 256, "gb200", "fp8_mx", "large_scale"),
     ("qwen", "qwen3_235b_a22b", "pretrain", 256, "b300", "fp8_mx", "large_scale"),
diff --git a/scripts/performance/setup_experiment.py b/scripts/performance/setup_experiment.py
index 807a46e6b1..fd12119cf7 100755
--- a/scripts/performance/setup_experiment.py
+++ b/scripts/performance/setup_experiment.py
@@ -672,7 +672,7 @@ def main(
         ]
     )
 
-    if peak_mem_clk is None and gpu == "vr200":
+    if peak_mem_clk is None and gpu == "vrnvl72":
         peak_mem_clk = 4752
     if peak_mem_clk == -1:
         peak_mem_clk = None
diff --git a/scripts/performance/utils/executors.py b/scripts/performance/utils/executors.py
index 59b25d3428..f229eef2db 100644
--- a/scripts/performance/utils/executors.py
+++ b/scripts/performance/utils/executors.py
@@ -185,7 +185,7 @@ def slurm_executor(
     log_repo_status_cmd = "bash /opt/Megatron-Bridge/docker/common/print_sha.sh /nemo_run/configs/repo_status.json"
     custom_bash_cmds.append(log_repo_status_cmd)
 
-    numa_divisor = 2 if gpu.lower() in ["gb200", "gb300", "vr200"] else 4
+    numa_divisor = 2 if gpu.lower() in ["gb200", "gb300", "vrnvl72"] else 4
     numa_cmd = f"numactl --cpunodebind=$((SLURM_LOCALID/{numa_divisor})) --membind=$((SLURM_LOCALID/{numa_divisor}))"
     if gpu.lower() in ["b300"] and enable_pct_binding:
         numa_cmd += " -C $((SLURM_LOCALID * 16)),$((SLURM_LOCALID * 16 + 1))"
diff --git a/scripts/performance/utils/overrides.py b/scripts/performance/utils/overrides.py
index 1830bb19dc..17586c22ab 100644
--- a/scripts/performance/utils/overrides.py
+++ b/scripts/performance/utils/overrides.py
@@ -280,7 +280,7 @@ def _apply_flat_cli_environment_compatibility(
     if connection_override:
         moe_a2a_overlap = args.moe_a2a_overlap if args.moe_a2a_overlap is not None else base_moe_a2a_overlap
         max_connections = 32 if effective_dispatcher_backend in {"deepep", "hybridep", "ncclep"} else 8
-        if gpu in {"b200", "b300", "gb200", "gb300", "vr200"}:
+        if gpu in {"b200", "b300", "gb200", "gb300", "vrnvl72"}:
             max_connections = 32
         elif (tp_size > 1 or cp_size > 1) and not moe_a2a_overlap:
             max_connections = 1
@@ -292,7 +292,7 @@ def _apply_flat_cli_environment_compatibility(
         if gpu in {"h100", "b200", "b300"}:
             domain_size = 8
             ranks_per_domain = min(ep_size, domain_size)
-        elif gpu in {"gb200", "gb300", "vr200", "r100"}:
+        elif gpu in {"gb200", "gb300", "vrnvl72", "r100"}:
             domain_size = 72
             if ep_size > domain_size:
                 raise ValueError("HybridEP expert parallel size must not exceed the 72-rank NVLink domain.")
diff --git a/scripts/performance/utils/utils.py b/scripts/performance/utils/utils.py
index 54c6f4053c..26a8a90036 100644
--- a/scripts/performance/utils/utils.py
+++ b/scripts/performance/utils/utils.py
@@ -324,7 +324,7 @@ def apply_target_topology_environment(
             "NVLINK_DOMAIN_SIZE": 8,
             "USE_MNNVL": 0,
         }
-    elif normalized_gpu in {"gb200", "gb300", "vr200", "r100"}:
+    elif normalized_gpu in {"gb200", "gb300", "vrnvl72", "r100"}:
         if expert_model_parallel_size > 72:
             raise ValueError("HybridEP expert parallel size must not exceed the 72-rank NVLink domain.")
         topology = {
diff --git a/scripts/training/README.md b/scripts/training/README.md
index bd86810fcd..8d84139230 100644
--- a/scripts/training/README.md
+++ b/scripts/training/README.md
@@ -107,6 +107,9 @@ through configurable library recipes. Explicit benchmark dataset replacement rem
 Five legacy duplicate names resolve to the benchmark definition; their library workloads remain available through
 the corresponding generic recipe aliases. New recipe names should be unique across both packages.
 
+Benchmark recipe names that use the deprecated `vr200` hardware name resolve to the renamed `vrnvl72` recipe with a
+warning.
+
 Text recipes default to `llm_step`; all recipes infer their modality-specific forward step from the same registry.
 Pass `--step-func NAME` to override that selection. Common training,
 sequence-length, parallelism, optimization, and checkpoint fields also have convenience flags such as
@@ -340,8 +343,8 @@ populate the executor's `additional_parameters` and are not forwarded to the
 training command or its srun step.
 
 Use `-lmc FREQ`, `--peak-mem-clk FREQ`, or `--peak_mem_clk FREQ` to lock the GPU memory clock to a fixed frequency in
-MHz once per node before training. VR200 benchmark recipes default to `4752`; other recipes leave the memory clock
-unlocked. Pass `-lmc -1` to disable the VR200 default explicitly.
+MHz once per node before training. VRNVL72 benchmark recipes default to `4752`; other recipes leave the memory clock
+unlocked. Pass `-lmc -1` to disable the VRNVL72 default explicitly.
 
 The compact launcher does not add rank-command prefixes or infer target-specific
 allocation and NUMA policies. Configure those explicitly for the target.
diff --git a/scripts/training/recipe_metadata.py b/scripts/training/recipe_metadata.py
index f29d948a5f..54bbccc450 100644
--- a/scripts/training/recipe_metadata.py
+++ b/scripts/training/recipe_metadata.py
@@ -19,11 +19,15 @@
 import argparse
 import ast
 import functools
+import logging
+import os
 import re
 from dataclasses import dataclass
 from pathlib import Path
 
 
+logger = logging.getLogger(__name__)
+
 BENCHMARK_RECIPE_FAMILY_PREFIXES = (
     ("qwen3_vl_", "qwen_vl"),
     ("qwen35_vl_", "qwen_vl"),
@@ -45,6 +49,10 @@
     r"(?:_[a-z0-9_]+)?_config$"
 )
 
+# Deprecated recipe hardware names and their replacements. When the deprecation window ends, drop an entry together
+# with its counterpart in LEGACY_GPU_ALIASES (scripts/performance/argument_parser.py).
+LEGACY_HARDWARE_ALIASES = {"vr200": "vrnvl72"}
+
 BENCHMARK_RECIPE_ROOT = Path(__file__).resolve().parents[2] / "src" / "megatron" / "bridge" / "perf_recipes"
 
 # These names are exported by both recipe packages today. Bare recipe lookup
@@ -181,8 +189,31 @@ def available_benchmark_recipe_metadata(recipe_name: str) -> BenchmarkRecipeMeta
     return metadata
 
 
+@functools.lru_cache(maxsize=None)
+def _canonical_recipe_name(recipe_name: str) -> str:
+    """Replace a deprecated hardware name in a benchmark recipe name when the renamed recipe is exported.
+
+    Cached so each process logs the deprecation once per requested name.
+    """
+    match = BENCHMARK_RECIPE_PATTERN.search(recipe_name)
+    if match is None or match.group("hardware") not in LEGACY_HARDWARE_ALIASES:
+        return recipe_name
+    start, end = match.span("hardware")
+    canonical_name = f"{recipe_name[:start]}{LEGACY_HARDWARE_ALIASES[match.group('hardware')]}{recipe_name[end:]}"
+    if available_benchmark_recipe_metadata(canonical_name) is None:
+        return recipe_name
+    # Every rank re-runs this lookup, so only the launcher and global rank 0 warn.
+    if os.environ.get("RANK", os.environ.get("SLURM_PROCID", "0")) == "0":
+        logger.warning("Recipe '%s' uses a deprecated hardware name; use '%s'.", recipe_name, canonical_name)
+    return canonical_name
+
+
 def resolved_benchmark_recipe_metadata(recipe_name: str) -> BenchmarkRecipeMetadata | None:
-    """Return benchmark metadata after applying unified lookup precedence."""
+    """Return benchmark metadata after applying legacy hardware aliases and unified lookup precedence.
+
+    A deprecated hardware name resolves to the renamed recipe, so callers load ``metadata.recipe_name``.
+    """
+    recipe_name = _canonical_recipe_name(recipe_name)
     if recipe_name in LIBRARY_RECIPE_PRECEDENCE_COLLISIONS:
         return None
     return available_benchmark_recipe_metadata(recipe_name)
diff --git a/scripts/training/recipe_runner.py b/scripts/training/recipe_runner.py
index a22cd3a731..07e90d0fb7 100644
--- a/scripts/training/recipe_runner.py
+++ b/scripts/training/recipe_runner.py
@@ -233,7 +233,10 @@ def load_recipe(
     seq_length: int | None = None,
 ) -> ConfigContainer:
     """Load an exact benchmark recipe when exported, otherwise a library recipe."""
-    if resolved_benchmark_recipe_metadata(recipe_name) is not None:
+    benchmark_metadata = resolved_benchmark_recipe_metadata(recipe_name)
+    if benchmark_metadata is not None:
+        # The resolved name replaces a deprecated hardware name in the requested one.
+        recipe_name = benchmark_metadata.recipe_name
         config_builder = find_benchmark_recipe(recipe_name)
         package_name = "megatron.bridge.perf_recipes"
         recipe_kind = "benchmark"
diff --git a/scripts/training/setup_experiment.py b/scripts/training/setup_experiment.py
index 9e76c66957..d801f0a2e1 100755
--- a/scripts/training/setup_experiment.py
+++ b/scripts/training/setup_experiment.py
@@ -126,7 +126,7 @@ def _build_parser() -> argparse.ArgumentParser:
         dest="peak_mem_clk",
         help=(
             "Lock the GPU memory clock to a fixed frequency in MHz once per node. "
-            "Defaults to 4752 for VR200 benchmark recipes and is disabled otherwise; pass -1 to disable the default."
+            "Defaults to 4752 for VRNVL72 benchmark recipes and is disabled otherwise; pass -1 to disable the default."
         ),
     )
     execution.add_argument("--experiment-name", help="NeMo-Run experiment name.")
@@ -218,12 +218,12 @@ def _resolve_peak_mem_clk(
     requested_peak_mem_clk: int | None,
     benchmark_metadata: BenchmarkRecipeMetadata | None,
 ) -> int | None:
-    """Resolve the explicit memory clock or the VR200 benchmark default."""
+    """Resolve the explicit memory clock or the VRNVL72 benchmark default."""
     if requested_peak_mem_clk == -1:
         return None
     if requested_peak_mem_clk is not None:
         return requested_peak_mem_clk
-    if benchmark_metadata is not None and benchmark_metadata.hardware == "vr200":
+    if benchmark_metadata is not None and benchmark_metadata.hardware == "vrnvl72":
         return 4752
     return None
 
diff --git a/src/megatron/bridge/perf_recipes/deepseek/__init__.py b/src/megatron/bridge/perf_recipes/deepseek/__init__.py
index ddcef4b632..5d9c852e22 100644
--- a/src/megatron/bridge/perf_recipes/deepseek/__init__.py
+++ b/src/megatron/bridge/perf_recipes/deepseek/__init__.py
@@ -52,18 +52,18 @@
     deepseek_v3_pretrain_1024gpu_h100_fp8sc_config,
     deepseek_v3_pretrain_1024gpu_h100_fp8sc_large_scale_config,
 )
-from megatron.bridge.perf_recipes.deepseek.vr200.deepseek_v3 import (
-    deepseek_v3_pretrain_64gpu_vr200_fp8mx_proxy_config,
-    deepseek_v3_pretrain_64gpu_vr200_nvfp4_proxy_config,
-    deepseek_v3_pretrain_128gpu_vr200_bf16_config,
-    deepseek_v3_pretrain_128gpu_vr200_fp8cs_config,
-    deepseek_v3_pretrain_128gpu_vr200_fp8mx_config,
-    deepseek_v3_pretrain_128gpu_vr200_nvfp4_config,
-    deepseek_v3_pretrain_256gpu_vr200_bf16_config,
-    deepseek_v3_pretrain_256gpu_vr200_fp8cs_config,
-    deepseek_v3_pretrain_256gpu_vr200_fp8mx_config,
-    deepseek_v3_pretrain_256gpu_vr200_nvfp4_config,
+from megatron.bridge.perf_recipes.deepseek.vrnvl72.deepseek_v3 import (
+    deepseek_v3_pretrain_64gpu_vrnvl72_fp8mx_proxy_config,
+    deepseek_v3_pretrain_64gpu_vrnvl72_nvfp4_proxy_config,
+    deepseek_v3_pretrain_128gpu_vrnvl72_bf16_config,
+    deepseek_v3_pretrain_128gpu_vrnvl72_fp8cs_config,
+    deepseek_v3_pretrain_128gpu_vrnvl72_fp8mx_config,
+    deepseek_v3_pretrain_128gpu_vrnvl72_nvfp4_config,
+    deepseek_v3_pretrain_256gpu_vrnvl72_bf16_config,
+    deepseek_v3_pretrain_256gpu_vrnvl72_fp8cs_config,
+    deepseek_v3_pretrain_256gpu_vrnvl72_fp8mx_config,
+    deepseek_v3_pretrain_256gpu_vrnvl72_nvfp4_config,
 )
-from megatron.bridge.perf_recipes.deepseek.vr200.deepseek_v4 import (
-    deepseek_v4_flash_pretrain_128gpu_vr200_fp8mx_config,
+from megatron.bridge.perf_recipes.deepseek.vrnvl72.deepseek_v4 import (
+    deepseek_v4_flash_pretrain_128gpu_vrnvl72_fp8mx_config,
 )
diff --git a/src/megatron/bridge/perf_recipes/deepseek/gb300/deepseek_v3.py b/src/megatron/bridge/perf_recipes/deepseek/gb300/deepseek_v3.py
index 6568d0c558..a5bf28ec0b 100644
--- a/src/megatron/bridge/perf_recipes/deepseek/gb300/deepseek_v3.py
+++ b/src/megatron/bridge/perf_recipes/deepseek/gb300/deepseek_v3.py
@@ -29,7 +29,7 @@
 
 
 def _build_deepseek_v3_gb300_bf16() -> ConfigContainer:
-    """Shared HybridEP BF16 base for the DeepSeek V3 256-GPU GB300 recipe and its VR200 alias."""
+    """Shared HybridEP BF16 base for the DeepSeek V3 256-GPU GB300 recipe and its VRNVL72 alias."""
     cfg = deepseek_v3_pretrain_config()
     cfg.mixed_precision = _perf_precision("bf16")
     _deepseek_v3_common(cfg)
@@ -136,7 +136,7 @@ def deepseek_v3_pretrain_256gpu_gb300_fp8cs_config() -> ConfigContainer:
 
 
 def _build_deepseek_v3_gb300_fp8mx() -> ConfigContainer:
-    """Shared HybridEP MXFP8 base for the DeepSeek V3 256-GPU GB300 recipe and its VR200 alias."""
+    """Shared HybridEP MXFP8 base for the DeepSeek V3 256-GPU GB300 recipe and its VRNVL72 alias."""
     cfg = deepseek_v3_pretrain_config()
     cfg.mixed_precision = _perf_precision("fp8_mx")
     _deepseek_v3_common(cfg)
@@ -198,7 +198,7 @@ def deepseek_v3_pretrain_256gpu_gb300_fp8mx_config() -> ConfigContainer:
 
 
 def _build_deepseek_v3_gb300_nvfp4() -> ConfigContainer:
-    """Shared HybridEP NVFP4 base for the DeepSeek V3 256-GPU GB300 recipe and its VR200 alias."""
+    """Shared HybridEP NVFP4 base for the DeepSeek V3 256-GPU GB300 recipe and its VRNVL72 alias."""
     cfg = deepseek_v3_pretrain_config()
     cfg.mixed_precision = _perf_precision("nvfp4")
     _deepseek_v3_common(cfg)
diff --git a/src/megatron/bridge/perf_recipes/deepseek/vr200/__init__.py b/src/megatron/bridge/perf_recipes/deepseek/vrnvl72/__init__.py
similarity index 100%
rename from src/megatron/bridge/perf_recipes/deepseek/vr200/__init__.py
rename to src/megatron/bridge/perf_recipes/deepseek/vrnvl72/__init__.py
diff --git a/src/megatron/bridge/perf_recipes/deepseek/vr200/deepseek_v3.py b/src/megatron/bridge/perf_recipes/deepseek/vrnvl72/deepseek_v3.py
similarity index 92%
rename from src/megatron/bridge/perf_recipes/deepseek/vr200/deepseek_v3.py
rename to src/megatron/bridge/perf_recipes/deepseek/vrnvl72/deepseek_v3.py
index 98454b93cc..d48159ac96 100644
--- a/src/megatron/bridge/perf_recipes/deepseek/vr200/deepseek_v3.py
+++ b/src/megatron/bridge/perf_recipes/deepseek/vrnvl72/deepseek_v3.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for DeepSeek V3."""
+"""VRNVL72 performance recipes for DeepSeek V3."""
 
 from megatron.bridge.perf_recipes.deepseek.common import (
     ConfigContainer,
@@ -31,8 +31,8 @@
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 
 
-def deepseek_v3_pretrain_128gpu_vr200_bf16_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 128× VR200, BF16."""
+def deepseek_v3_pretrain_128gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 128× VRNVL72, BF16."""
     cfg = deepseek_v3_pretrain_config()
     cfg.mixed_precision = _perf_precision("bf16")
     _deepseek_v3_common(cfg)
@@ -82,8 +82,8 @@ def deepseek_v3_pretrain_128gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_128gpu_vr200_fp8cs_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 128× VR200, FP8 current-scaling."""
+def deepseek_v3_pretrain_128gpu_vrnvl72_fp8cs_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 128× VRNVL72, FP8 current-scaling."""
     cfg = deepseek_v3_pretrain_config()
     cfg.mixed_precision = _perf_precision("fp8_cs")
     _deepseek_v3_common(cfg)
@@ -133,8 +133,8 @@ def deepseek_v3_pretrain_128gpu_vr200_fp8cs_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_128gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 128× VR200, MXFP8."""
+def deepseek_v3_pretrain_128gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 128× VRNVL72, MXFP8."""
     cfg = deepseek_v3_pretrain_config()
     cfg.mixed_precision = _perf_precision("fp8_mx")
     _deepseek_v3_common(cfg)
@@ -189,8 +189,8 @@ def deepseek_v3_pretrain_128gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_128gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 128× VR200, NVFP4."""
+def deepseek_v3_pretrain_128gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 128× VRNVL72, NVFP4."""
     cfg = deepseek_v3_pretrain_config()
     cfg.mixed_precision = _perf_precision("nvfp4")
     _deepseek_v3_common(cfg)
@@ -242,8 +242,8 @@ def deepseek_v3_pretrain_128gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 256× VR200, BF16 (alias of GB300)."""
+def deepseek_v3_pretrain_256gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 256× VRNVL72, BF16 (alias of GB300)."""
     cfg = _build_deepseek_v3_gb300_bf16()
     cfg.model.cuda_graph_scope = ["attn"]
 
@@ -272,8 +272,8 @@ def deepseek_v3_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_256gpu_vr200_fp8cs_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 256× VR200, FP8-CS (alias of GB300)."""
+def deepseek_v3_pretrain_256gpu_vrnvl72_fp8cs_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 256× VRNVL72, FP8-CS (alias of GB300)."""
     cfg = deepseek_v3_pretrain_256gpu_gb300_fp8cs_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -300,8 +300,8 @@ def deepseek_v3_pretrain_256gpu_vr200_fp8cs_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_256gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 256× VR200, FP8-MX (alias of GB300)."""
+def deepseek_v3_pretrain_256gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 256× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = _build_deepseek_v3_gb300_fp8mx()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -333,8 +333,8 @@ def deepseek_v3_pretrain_256gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_64gpu_vr200_fp8mx_proxy_config() -> ConfigContainer:
-    """DeepSeek V3 MXFP8 debugging proxy: 13 decoder layers on 64 VR200 GPUs.
+def deepseek_v3_pretrain_64gpu_vrnvl72_fp8mx_proxy_config() -> ConfigContainer:
+    """DeepSeek V3 MXFP8 debugging proxy: 13 decoder layers on 64 VRNVL72 GPUs.
 
     Keep GBS=4096, per-layer shapes, EP=32 and one MTP layer. PP=1 without
     virtual pipelining gives dense DP=64, expert DP=2 and 64 microbatches.
@@ -342,7 +342,7 @@ def deepseek_v3_pretrain_64gpu_vr200_fp8mx_proxy_config() -> ConfigContainer:
     overlap path. This reduced-depth model does not reproduce at-scale
     pipeline traffic, convergence or memory use; GPU validation is required.
     """
-    cfg = deepseek_v3_pretrain_256gpu_vr200_fp8mx_config()
+    cfg = deepseek_v3_pretrain_256gpu_vrnvl72_fp8mx_config()
     cfg.model.num_layers = 13
     # Retain DeepSeek V3's three leading dense layers; shorten its MoE pattern.
     cfg.model.moe_layer_freq = [0] * 3 + [1] * (cfg.model.num_layers - 3)
@@ -374,8 +374,8 @@ def deepseek_v3_pretrain_64gpu_vr200_fp8mx_proxy_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_256gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """DeepSeek V3 pretrain: 256× VR200, NVFP4 with full-iteration CUDA graph."""
+def deepseek_v3_pretrain_256gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """DeepSeek V3 pretrain: 256× VRNVL72, NVFP4 with full-iteration CUDA graph."""
     cfg = _build_deepseek_v3_gb300_nvfp4()
 
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
@@ -408,8 +408,8 @@ def deepseek_v3_pretrain_256gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def deepseek_v3_pretrain_64gpu_vr200_nvfp4_proxy_config() -> ConfigContainer:
-    """DeepSeek V3 NVFP4 debugging proxy: 13 decoder layers on 64 VR200 GPUs.
+def deepseek_v3_pretrain_64gpu_vrnvl72_nvfp4_proxy_config() -> ConfigContainer:
+    """DeepSeek V3 NVFP4 debugging proxy: 13 decoder layers on 64 VRNVL72 GPUs.
 
     Keep GBS=4096, per-layer shapes, EP=32 and one MTP layer. PP=1 without
     virtual pipelining gives dense DP=64, expert DP=2 and 64 microbatches.
@@ -417,7 +417,7 @@ def deepseek_v3_pretrain_64gpu_vr200_nvfp4_proxy_config() -> ConfigContainer:
     overlap path. Use this reduced-depth model for feature and kernel
     performance experiments, not convergence or at-scale acceptance.
     """
-    cfg = deepseek_v3_pretrain_256gpu_vr200_nvfp4_config()
+    cfg = deepseek_v3_pretrain_256gpu_vrnvl72_nvfp4_config()
     cfg.model.num_layers = 13
     cfg.model.moe_layer_freq = [0] * 3 + [1] * (cfg.model.num_layers - 3)
     cfg.model.pipeline_model_parallel_size = 1
diff --git a/src/megatron/bridge/perf_recipes/deepseek/vr200/deepseek_v4.py b/src/megatron/bridge/perf_recipes/deepseek/vrnvl72/deepseek_v4.py
similarity index 87%
rename from src/megatron/bridge/perf_recipes/deepseek/vr200/deepseek_v4.py
rename to src/megatron/bridge/perf_recipes/deepseek/vrnvl72/deepseek_v4.py
index 775ce8cbcf..1ed0d3bf3a 100644
--- a/src/megatron/bridge/perf_recipes/deepseek/vr200/deepseek_v4.py
+++ b/src/megatron/bridge/perf_recipes/deepseek/vrnvl72/deepseek_v4.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for DeepSeek V4."""
+"""VRNVL72 performance recipes for DeepSeek V4."""
 
 from megatron.bridge.perf_recipes.deepseek.gb300.deepseek_v4 import (
     deepseek_v4_flash_pretrain_128gpu_gb300_fp8mx_config,
@@ -20,8 +20,8 @@
 from megatron.bridge.training.config import ConfigContainer
 
 
-def deepseek_v4_flash_pretrain_128gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """DeepSeek V4 Flash pretrain: 128× VR200, MXFP8 (alias of GB300)."""
+def deepseek_v4_flash_pretrain_128gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """DeepSeek V4 Flash pretrain: 128× VRNVL72, MXFP8 (alias of GB300)."""
     cfg = deepseek_v4_flash_pretrain_128gpu_gb300_fp8mx_config()
 
     # Rubin's grouped GLU kernel does not support DeepSeek V4's SwiGLU clamp parameters. This is a
@@ -30,7 +30,7 @@ def deepseek_v4_flash_pretrain_128gpu_vr200_fp8mx_config() -> ConfigContainer:
     cfg.model.use_transformer_engine_op_fuser = True
     cfg.model.moe_use_grouped_tensor = True
 
-    # Keep the VR200 launch environment explicit instead of inheriting it from GB300.
+    # Keep the VRNVL72 launch environment explicit instead of inheriting it from GB300.
     cfg.env_vars = {
         **COMMON_PERF_ENV_VARS,
         "CUDA_DEVICE_MAX_CONNECTIONS": 32,
diff --git a/src/megatron/bridge/perf_recipes/gpt_oss/__init__.py b/src/megatron/bridge/perf_recipes/gpt_oss/__init__.py
index 9b3352d7ab..f026e1b57d 100644
--- a/src/megatron/bridge/perf_recipes/gpt_oss/__init__.py
+++ b/src/megatron/bridge/perf_recipes/gpt_oss/__init__.py
@@ -28,10 +28,10 @@
     gpt_oss_120b_pretrain_64gpu_h100_bf16_config,
     gpt_oss_120b_pretrain_64gpu_h100_fp8cs_config,
 )
-from megatron.bridge.perf_recipes.gpt_oss.vr200.gpt_oss import (
-    gpt_oss_20b_pretrain_8gpu_vr200_fp8mx_config,
-    gpt_oss_20b_pretrain_8gpu_vr200_nvfp4_config,
-    gpt_oss_20b_pretrain_64gpu_vr200_nvfp4_config,
-    gpt_oss_120b_pretrain_64gpu_vr200_bf16_config,
-    gpt_oss_120b_pretrain_64gpu_vr200_fp8mx_config,
+from megatron.bridge.perf_recipes.gpt_oss.vrnvl72.gpt_oss import (
+    gpt_oss_20b_pretrain_8gpu_vrnvl72_fp8mx_config,
+    gpt_oss_20b_pretrain_8gpu_vrnvl72_nvfp4_config,
+    gpt_oss_20b_pretrain_64gpu_vrnvl72_nvfp4_config,
+    gpt_oss_120b_pretrain_64gpu_vrnvl72_bf16_config,
+    gpt_oss_120b_pretrain_64gpu_vrnvl72_fp8mx_config,
 )
diff --git a/src/megatron/bridge/perf_recipes/gpt_oss/gb300/gpt_oss.py b/src/megatron/bridge/perf_recipes/gpt_oss/gb300/gpt_oss.py
index 472db4e2ce..dfcdec9064 100644
--- a/src/megatron/bridge/perf_recipes/gpt_oss/gb300/gpt_oss.py
+++ b/src/megatron/bridge/perf_recipes/gpt_oss/gb300/gpt_oss.py
@@ -216,7 +216,7 @@ def gpt_oss_120b_pretrain_64gpu_gb300_bf16_config() -> ConfigContainer:
 
 
 def _build_gpt_oss_120b_gb300_fp8mx() -> ConfigContainer:
-    """Shared HybridEP MXFP8 base for the GPT-OSS 120B GB300 recipe and its VR200 alias."""
+    """Shared HybridEP MXFP8 base for the GPT-OSS 120B GB300 recipe and its VRNVL72 alias."""
     cfg = gpt_oss_120b_pretrain_config()
     cfg.mixed_precision = _perf_precision("fp8_mx")
     cfg.model.moe_router_fusion = True
diff --git a/src/megatron/bridge/perf_recipes/gpt_oss/vr200/__init__.py b/src/megatron/bridge/perf_recipes/gpt_oss/vrnvl72/__init__.py
similarity index 100%
rename from src/megatron/bridge/perf_recipes/gpt_oss/vr200/__init__.py
rename to src/megatron/bridge/perf_recipes/gpt_oss/vrnvl72/__init__.py
diff --git a/src/megatron/bridge/perf_recipes/gpt_oss/vr200/gpt_oss.py b/src/megatron/bridge/perf_recipes/gpt_oss/vrnvl72/gpt_oss.py
similarity index 91%
rename from src/megatron/bridge/perf_recipes/gpt_oss/vr200/gpt_oss.py
rename to src/megatron/bridge/perf_recipes/gpt_oss/vrnvl72/gpt_oss.py
index 7dc0e3985f..dbc5450e9c 100644
--- a/src/megatron/bridge/perf_recipes/gpt_oss/vr200/gpt_oss.py
+++ b/src/megatron/bridge/perf_recipes/gpt_oss/vrnvl72/gpt_oss.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for GPT-OSS."""
+"""VRNVL72 performance recipes for GPT-OSS."""
 
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 from megatron.bridge.perf_recipes.gpt_oss.common import (
@@ -30,8 +30,8 @@
 )
 
 
-def gpt_oss_20b_pretrain_8gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """GPT-OSS 20B pretrain: 8× VR200, NVFP4 (alias of GB300)."""
+def gpt_oss_20b_pretrain_8gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """GPT-OSS 20B pretrain: 8× VRNVL72, NVFP4 (alias of GB300)."""
     cfg = gpt_oss_20b_pretrain_8gpu_gb300_nvfp4_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -60,8 +60,8 @@ def gpt_oss_20b_pretrain_8gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def gpt_oss_20b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """GPT-OSS 20B pretrain: 8× VR200, MXFP8."""
+def gpt_oss_20b_pretrain_8gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """GPT-OSS 20B pretrain: 8× VRNVL72, MXFP8."""
     cfg = gpt_oss_20b_pretrain_config()
     cfg.mixed_precision = _gpt_oss_20b_fp8mx_precision()
 
@@ -108,8 +108,8 @@ def gpt_oss_20b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def gpt_oss_20b_pretrain_64gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """GPT-OSS 20B pretrain: 64× VR200, NVFP4."""
+def gpt_oss_20b_pretrain_64gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """GPT-OSS 20B pretrain: 64× VRNVL72, NVFP4."""
     cfg = gpt_oss_20b_pretrain_config()
     cfg.mixed_precision = _gpt_oss_20b_nvfp4_precision()
 
@@ -158,8 +158,8 @@ def gpt_oss_20b_pretrain_64gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def gpt_oss_120b_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
-    """GPT-OSS 120B pretrain: 64× VR200, BF16 (alias of GB300)."""
+def gpt_oss_120b_pretrain_64gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """GPT-OSS 120B pretrain: 64× VRNVL72, BF16 (alias of GB300)."""
     cfg = gpt_oss_120b_pretrain_64gpu_gb300_bf16_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -179,8 +179,8 @@ def gpt_oss_120b_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def gpt_oss_120b_pretrain_64gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """GPT-OSS 120B pretrain: 64× VR200, FP8-MX (alias of GB300)."""
+def gpt_oss_120b_pretrain_64gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """GPT-OSS 120B pretrain: 64× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = _build_gpt_oss_120b_gb300_fp8mx()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
diff --git a/src/megatron/bridge/perf_recipes/kimi/__init__.py b/src/megatron/bridge/perf_recipes/kimi/__init__.py
index 88e814b1ac..31f53470d3 100644
--- a/src/megatron/bridge/perf_recipes/kimi/__init__.py
+++ b/src/megatron/bridge/perf_recipes/kimi/__init__.py
@@ -24,7 +24,7 @@
     kimi_k2_pretrain_1024gpu_h100_fp8cs_config,
     kimi_k2_pretrain_1024gpu_h100_fp8sc_config,
 )
-from megatron.bridge.perf_recipes.kimi.vr200.kimi_k2 import (
-    kimi_k2_pretrain_256gpu_vr200_bf16_config,
-    kimi_k2_pretrain_256gpu_vr200_fp8mx_config,
+from megatron.bridge.perf_recipes.kimi.vrnvl72.kimi_k2 import (
+    kimi_k2_pretrain_256gpu_vrnvl72_bf16_config,
+    kimi_k2_pretrain_256gpu_vrnvl72_fp8mx_config,
 )
diff --git a/src/megatron/bridge/perf_recipes/kimi/gb300/kimi_k2.py b/src/megatron/bridge/perf_recipes/kimi/gb300/kimi_k2.py
index cfcbf9ac83..a41d5c916b 100644
--- a/src/megatron/bridge/perf_recipes/kimi/gb300/kimi_k2.py
+++ b/src/megatron/bridge/perf_recipes/kimi/gb300/kimi_k2.py
@@ -137,7 +137,7 @@ def kimi_k2_pretrain_256gpu_gb300_fp8cs_config() -> ConfigContainer:
 
 
 def _build_kimi_k2_gb300_fp8mx() -> ConfigContainer:
-    """Shared HybridEP MXFP8 base for the Kimi K2 GB300 recipe and its VR200 alias."""
+    """Shared HybridEP MXFP8 base for the Kimi K2 GB300 recipe and its VRNVL72 alias."""
     cfg = kimi_k2_pretrain_config()
     cfg.mixed_precision = _perf_precision("fp8_mx")
     cfg.mixed_precision.reuse_grad_buf_for_mxfp8_param_ag = False
diff --git a/src/megatron/bridge/perf_recipes/kimi/vr200/__init__.py b/src/megatron/bridge/perf_recipes/kimi/vrnvl72/__init__.py
similarity index 100%
rename from src/megatron/bridge/perf_recipes/kimi/vr200/__init__.py
rename to src/megatron/bridge/perf_recipes/kimi/vrnvl72/__init__.py
diff --git a/src/megatron/bridge/perf_recipes/kimi/vr200/kimi_k2.py b/src/megatron/bridge/perf_recipes/kimi/vrnvl72/kimi_k2.py
similarity index 90%
rename from src/megatron/bridge/perf_recipes/kimi/vr200/kimi_k2.py
rename to src/megatron/bridge/perf_recipes/kimi/vrnvl72/kimi_k2.py
index 6245b38773..208168dda5 100644
--- a/src/megatron/bridge/perf_recipes/kimi/vr200/kimi_k2.py
+++ b/src/megatron/bridge/perf_recipes/kimi/vrnvl72/kimi_k2.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for Kimi K2."""
+"""VRNVL72 performance recipes for Kimi K2."""
 
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 from megatron.bridge.perf_recipes.kimi.common import (
@@ -23,8 +23,8 @@
 )
 
 
-def kimi_k2_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
-    """Kimi K2 pretrain: 256× VR200, BF16 (alias of GB300)."""
+def kimi_k2_pretrain_256gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Kimi K2 pretrain: 256× VRNVL72, BF16 (alias of GB300)."""
     cfg = kimi_k2_pretrain_256gpu_gb300_bf16_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -49,8 +49,8 @@ def kimi_k2_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def kimi_k2_pretrain_256gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Kimi K2 pretrain: 256× VR200, MXFP8 (alias of GB300)."""
+def kimi_k2_pretrain_256gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Kimi K2 pretrain: 256× VRNVL72, MXFP8 (alias of GB300)."""
     cfg = _build_kimi_k2_gb300_fp8mx()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
diff --git a/src/megatron/bridge/perf_recipes/llama/__init__.py b/src/megatron/bridge/perf_recipes/llama/__init__.py
index 8e46a39def..1f0e28d86c 100644
--- a/src/megatron/bridge/perf_recipes/llama/__init__.py
+++ b/src/megatron/bridge/perf_recipes/llama/__init__.py
@@ -137,17 +137,17 @@
     llama3_8b_pretrain_8gpu_r100_fp8mx_config,
     llama3_8b_pretrain_8gpu_r100_nvfp4_config,
 )
-from megatron.bridge.perf_recipes.llama.vr200.llama3 import (
-    llama3_8b_pretrain_8gpu_vr200_bf16_config,
-    llama3_8b_pretrain_8gpu_vr200_fp8cs_config,
-    llama3_8b_pretrain_8gpu_vr200_fp8mx_config,
-    llama3_8b_pretrain_8gpu_vr200_nvfp4_config,
-    llama3_70b_pretrain_64gpu_vr200_bf16_config,
-    llama3_70b_pretrain_64gpu_vr200_fp8mx_config,
-    llama3_70b_pretrain_64gpu_vr200_nvfp4_config,
+from megatron.bridge.perf_recipes.llama.vrnvl72.llama3 import (
+    llama3_8b_pretrain_8gpu_vrnvl72_bf16_config,
+    llama3_8b_pretrain_8gpu_vrnvl72_fp8cs_config,
+    llama3_8b_pretrain_8gpu_vrnvl72_fp8mx_config,
+    llama3_8b_pretrain_8gpu_vrnvl72_nvfp4_config,
+    llama3_70b_pretrain_64gpu_vrnvl72_bf16_config,
+    llama3_70b_pretrain_64gpu_vrnvl72_fp8mx_config,
+    llama3_70b_pretrain_64gpu_vrnvl72_nvfp4_config,
 )
-from megatron.bridge.perf_recipes.llama.vr200.llama31 import (
-    llama31_405b_pretrain_256gpu_vr200_bf16_config,
-    llama31_405b_pretrain_256gpu_vr200_fp8mx_config,
-    llama31_405b_pretrain_256gpu_vr200_nvfp4_config,
+from megatron.bridge.perf_recipes.llama.vrnvl72.llama31 import (
+    llama31_405b_pretrain_256gpu_vrnvl72_bf16_config,
+    llama31_405b_pretrain_256gpu_vrnvl72_fp8mx_config,
+    llama31_405b_pretrain_256gpu_vrnvl72_nvfp4_config,
 )
diff --git a/src/megatron/bridge/perf_recipes/llama/vr200/__init__.py b/src/megatron/bridge/perf_recipes/llama/vrnvl72/__init__.py
similarity index 100%
rename from src/megatron/bridge/perf_recipes/llama/vr200/__init__.py
rename to src/megatron/bridge/perf_recipes/llama/vrnvl72/__init__.py
diff --git a/src/megatron/bridge/perf_recipes/llama/vr200/llama3.py b/src/megatron/bridge/perf_recipes/llama/vrnvl72/llama3.py
similarity index 87%
rename from src/megatron/bridge/perf_recipes/llama/vr200/llama3.py
rename to src/megatron/bridge/perf_recipes/llama/vrnvl72/llama3.py
index 4cea157b66..b18b26026b 100644
--- a/src/megatron/bridge/perf_recipes/llama/vr200/llama3.py
+++ b/src/megatron/bridge/perf_recipes/llama/vrnvl72/llama3.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for Llama 3."""
+"""VRNVL72 performance recipes for Llama 3."""
 
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 from megatron.bridge.perf_recipes.llama.common import (
@@ -28,8 +28,8 @@
 )
 
 
-def llama3_8b_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
-    """LLaMA 3 8B pretrain: 8× VR200, BF16 (alias of GB300)."""
+def llama3_8b_pretrain_8gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """LLaMA 3 8B pretrain: 8× VRNVL72, BF16 (alias of GB300)."""
     cfg = llama3_8b_pretrain_8gpu_gb300_bf16_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -49,8 +49,8 @@ def llama3_8b_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def llama3_8b_pretrain_8gpu_vr200_fp8cs_config() -> ConfigContainer:
-    """LLaMA 3 8B pretrain: 8× VR200, FP8-CS (alias of GB300)."""
+def llama3_8b_pretrain_8gpu_vrnvl72_fp8cs_config() -> ConfigContainer:
+    """LLaMA 3 8B pretrain: 8× VRNVL72, FP8-CS (alias of GB300)."""
     cfg = llama3_8b_pretrain_8gpu_gb300_fp8cs_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -70,8 +70,8 @@ def llama3_8b_pretrain_8gpu_vr200_fp8cs_config() -> ConfigContainer:
     return cfg
 
 
-def llama3_8b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """LLaMA 3 8B pretrain: 8× VR200, FP8-MX (alias of GB300)."""
+def llama3_8b_pretrain_8gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """LLaMA 3 8B pretrain: 8× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = llama3_8b_pretrain_8gpu_gb300_fp8mx_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -91,8 +91,8 @@ def llama3_8b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def llama3_8b_pretrain_8gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """LLaMA 3 8B pretrain: 8× VR200, NVFP4 (alias of GB300)."""
+def llama3_8b_pretrain_8gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """LLaMA 3 8B pretrain: 8× VRNVL72, NVFP4 (alias of GB300)."""
     cfg = llama3_8b_pretrain_8gpu_gb300_nvfp4_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -114,8 +114,8 @@ def llama3_8b_pretrain_8gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def llama3_70b_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
-    """LLaMA 3 70B pretrain: 64× VR200, BF16 (alias of GB300)."""
+def llama3_70b_pretrain_64gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """LLaMA 3 70B pretrain: 64× VRNVL72, BF16 (alias of GB300)."""
     cfg = llama3_70b_pretrain_64gpu_gb300_bf16_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -137,8 +137,8 @@ def llama3_70b_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def llama3_70b_pretrain_64gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """LLaMA 3 70B pretrain: 64× VR200, FP8-MX (alias of GB300)."""
+def llama3_70b_pretrain_64gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """LLaMA 3 70B pretrain: 64× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = llama3_70b_pretrain_64gpu_gb300_fp8mx_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -160,8 +160,8 @@ def llama3_70b_pretrain_64gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def llama3_70b_pretrain_64gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """LLaMA 3 70B pretrain: 64× VR200, NVFP4 (alias of GB300)."""
+def llama3_70b_pretrain_64gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """LLaMA 3 70B pretrain: 64× VRNVL72, NVFP4 (alias of GB300)."""
     cfg = llama3_70b_pretrain_64gpu_gb300_nvfp4_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
diff --git a/src/megatron/bridge/perf_recipes/llama/vr200/llama31.py b/src/megatron/bridge/perf_recipes/llama/vrnvl72/llama31.py
similarity index 87%
rename from src/megatron/bridge/perf_recipes/llama/vr200/llama31.py
rename to src/megatron/bridge/perf_recipes/llama/vrnvl72/llama31.py
index c82a136981..fb5a5f383b 100644
--- a/src/megatron/bridge/perf_recipes/llama/vr200/llama31.py
+++ b/src/megatron/bridge/perf_recipes/llama/vrnvl72/llama31.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for Llama 3.1."""
+"""VRNVL72 performance recipes for Llama 3.1."""
 
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 from megatron.bridge.perf_recipes.llama.common import (
@@ -24,8 +24,8 @@
 )
 
 
-def llama31_405b_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
-    """Llama3.1 405B pretrain: 256x VR200, BF16 (alias of GB300)."""
+def llama31_405b_pretrain_256gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Llama3.1 405B pretrain: 256x VRNVL72, BF16 (alias of GB300)."""
     cfg = llama31_405b_pretrain_256gpu_gb300_bf16_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -43,8 +43,8 @@ def llama31_405b_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def llama31_405b_pretrain_256gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Llama3.1 405B pretrain: 256x VR200, FP8-MX (alias of GB300)."""
+def llama31_405b_pretrain_256gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Llama3.1 405B pretrain: 256x VRNVL72, FP8-MX (alias of GB300)."""
     cfg = llama31_405b_pretrain_256gpu_gb300_fp8mx_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -66,8 +66,8 @@ def llama31_405b_pretrain_256gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def llama31_405b_pretrain_256gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """Llama3.1 405B pretrain: 256x VR200, NVFP4 (alias of GB300)."""
+def llama31_405b_pretrain_256gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """Llama3.1 405B pretrain: 256x VRNVL72, NVFP4 (alias of GB300)."""
     cfg = llama31_405b_pretrain_256gpu_gb300_nvfp4_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
diff --git a/src/megatron/bridge/perf_recipes/nemotronh/__init__.py b/src/megatron/bridge/perf_recipes/nemotronh/__init__.py
index 46c29d69bb..6ef1d112d1 100644
--- a/src/megatron/bridge/perf_recipes/nemotronh/__init__.py
+++ b/src/megatron/bridge/perf_recipes/nemotronh/__init__.py
@@ -55,15 +55,15 @@
     nemotron_3_nano_pretrain_16gpu_h100_fp8cs_config,
     nemotron_3_super_pretrain_64gpu_h100_bf16_config,
 )
-from megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh import (
-    nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config,
-    nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config,
-    nemotron_3_nano_pretrain_8gpu_vr200_bf16_config,
-    nemotron_3_nano_pretrain_8gpu_vr200_fp8mx_config,
-    nemotron_3_nano_pretrain_8gpu_vr200_nvfp4_config,
-    nemotron_3_super_pretrain_64gpu_vr200_bf16_config,
-    nemotron_3_super_pretrain_64gpu_vr200_fp8mx_config,
-    nemotron_3_super_pretrain_64gpu_vr200_nvfp4_config,
-    nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config,
-    nemotron_3_ultra_pretrain_256gpu_vr200_nvfp4_config,
+from megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh import (
+    nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config,
+    nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config,
+    nemotron_3_nano_pretrain_8gpu_vrnvl72_bf16_config,
+    nemotron_3_nano_pretrain_8gpu_vrnvl72_fp8mx_config,
+    nemotron_3_nano_pretrain_8gpu_vrnvl72_nvfp4_config,
+    nemotron_3_super_pretrain_64gpu_vrnvl72_bf16_config,
+    nemotron_3_super_pretrain_64gpu_vrnvl72_fp8mx_config,
+    nemotron_3_super_pretrain_64gpu_vrnvl72_nvfp4_config,
+    nemotron_3_ultra_pretrain_256gpu_vrnvl72_fp8mx_config,
+    nemotron_3_ultra_pretrain_256gpu_vrnvl72_nvfp4_config,
 )
diff --git a/src/megatron/bridge/perf_recipes/nemotronh/gb300/nemotronh.py b/src/megatron/bridge/perf_recipes/nemotronh/gb300/nemotronh.py
index 2a751806aa..481814c27f 100644
--- a/src/megatron/bridge/perf_recipes/nemotronh/gb300/nemotronh.py
+++ b/src/megatron/bridge/perf_recipes/nemotronh/gb300/nemotronh.py
@@ -42,7 +42,7 @@
 
 
 def _build_nemotron_3_super_gb300_bf16() -> ConfigContainer:
-    """Shared HybridEP BF16 base for the Nemotron 3 Super GB300 recipe and its VR200 alias."""
+    """Shared HybridEP BF16 base for the Nemotron 3 Super GB300 recipe and its VRNVL72 alias."""
     cfg = nemotron_3_super_pretrain_config()
     cfg.mixed_precision = _perf_precision("bf16")
 
@@ -149,7 +149,7 @@ def nemotron_3_super_pretrain_64gpu_gb300_fp8mx_config() -> ConfigContainer:
 
 
 def _build_nemotron_3_super_gb300_nvfp4() -> ConfigContainer:
-    """Shared HybridEP NVFP4 base for the Nemotron 3 Super GB300 recipe and its VR200 alias."""
+    """Shared HybridEP NVFP4 base for the Nemotron 3 Super GB300 recipe and its VRNVL72 alias."""
     cfg = nemotron_3_super_pretrain_config()
     cfg.mixed_precision = _nemotron_3_super_nvfp4_precision()
 
@@ -395,7 +395,7 @@ def nemotron_3_ultra_pretrain_256gpu_gb300_nvfp4_config(*, num_gpus: int = 256)
 
 
 def _build_nemotron_3_nano_gb300_bf16() -> ConfigContainer:
-    """Shared HybridEP BF16 base for the Nemotron 3 Nano GB300 recipe and its VR200 alias."""
+    """Shared HybridEP BF16 base for the Nemotron 3 Nano GB300 recipe and its VRNVL72 alias."""
     cfg = nemotron_3_nano_pretrain_config()
     _apply_nemotron_3_nano_perf_defaults(cfg)
     cfg.mixed_precision = _perf_precision("bf16")
@@ -568,7 +568,7 @@ def nemotron_3_nano_pretrain_8gpu_gb300_nvfp4_config() -> ConfigContainer:
 
 
 def _build_nemotron_3_5_lightning_gb300_bf16() -> ConfigContainer:
-    """Shared HybridEP BF16 base for the Nemotron 3.5 Lightning GB300 recipe and its VR200 alias."""
+    """Shared HybridEP BF16 base for the Nemotron 3.5 Lightning GB300 recipe and its VRNVL72 alias."""
     cfg = _build_nemotron_3_nano_gb300_bf16()
     cfg.model.mtp_num_layers = 2
     cfg.model.mtp_hybrid_override_pattern = "*E"
diff --git a/src/megatron/bridge/perf_recipes/nemotronh/vr200/__init__.py b/src/megatron/bridge/perf_recipes/nemotronh/vrnvl72/__init__.py
similarity index 100%
rename from src/megatron/bridge/perf_recipes/nemotronh/vr200/__init__.py
rename to src/megatron/bridge/perf_recipes/nemotronh/vrnvl72/__init__.py
diff --git a/src/megatron/bridge/perf_recipes/nemotronh/vr200/nemotronh.py b/src/megatron/bridge/perf_recipes/nemotronh/vrnvl72/nemotronh.py
similarity index 87%
rename from src/megatron/bridge/perf_recipes/nemotronh/vr200/nemotronh.py
rename to src/megatron/bridge/perf_recipes/nemotronh/vrnvl72/nemotronh.py
index ee957c0c3e..bfb03ff6bb 100644
--- a/src/megatron/bridge/perf_recipes/nemotronh/vr200/nemotronh.py
+++ b/src/megatron/bridge/perf_recipes/nemotronh/vrnvl72/nemotronh.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for Nemotron 3 and 3.5."""
+"""VRNVL72 performance recipes for Nemotron 3 and 3.5."""
 
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 from megatron.bridge.perf_recipes.nemotronh.common import (
@@ -32,8 +32,8 @@
 )
 
 
-def nemotron_3_nano_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
-    """Nemotron 3 Nano pretrain: 8× VR200, BF16 (alias of GB300)."""
+def nemotron_3_nano_pretrain_8gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Nemotron 3 Nano pretrain: 8× VRNVL72, BF16 (alias of GB300)."""
     cfg = _build_nemotron_3_nano_gb300_bf16()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -61,8 +61,8 @@ def nemotron_3_nano_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def nemotron_3_nano_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Nemotron 3 Nano pretrain: 8× VR200, FP8-MX (alias of GB300)."""
+def nemotron_3_nano_pretrain_8gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Nemotron 3 Nano pretrain: 8× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = _build_nemotron_3_nano_gb300_mxfp8()
     cfg.model.use_transformer_engine_op_fuser = True
     cfg.model.moe_mlp_glu_interleave_size = 32
@@ -95,8 +95,8 @@ def nemotron_3_nano_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def nemotron_3_nano_pretrain_8gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """Nemotron 3 Nano pretrain: 8× VR200, NVFP4 (alias of GB300)."""
+def nemotron_3_nano_pretrain_8gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """Nemotron 3 Nano pretrain: 8× VRNVL72, NVFP4 (alias of GB300)."""
     cfg = nemotron_3_nano_pretrain_8gpu_gb300_nvfp4_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -126,8 +126,8 @@ def nemotron_3_nano_pretrain_8gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def nemotron_3_super_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
-    """Nemotron 3 Super pretrain: 64× VR200, BF16 (alias of GB300)."""
+def nemotron_3_super_pretrain_64gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Nemotron 3 Super pretrain: 64× VRNVL72, BF16 (alias of GB300)."""
     cfg = _build_nemotron_3_super_gb300_bf16()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -152,8 +152,8 @@ def nemotron_3_super_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def nemotron_3_super_pretrain_64gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Nemotron 3 Super pretrain: 64× VR200, MXFP8 with full-iteration CUDA graph."""
+def nemotron_3_super_pretrain_64gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Nemotron 3 Super pretrain: 64× VRNVL72, MXFP8 with full-iteration CUDA graph."""
     cfg = _build_nemotron_3_super_gb300_mxfp8()
     cfg.mixed_precision.fp8_dot_product_attention = True
     cfg.model.expert_model_parallel_size = 8
@@ -183,8 +183,8 @@ def nemotron_3_super_pretrain_64gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def nemotron_3_super_pretrain_64gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """Nemotron 3 Super pretrain: 64× VR200, NVFP4 with full-iteration CUDA graph."""
+def nemotron_3_super_pretrain_64gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """Nemotron 3 Super pretrain: 64× VRNVL72, NVFP4 with full-iteration CUDA graph."""
     cfg = _build_nemotron_3_super_gb300_nvfp4()
     cfg.model.expert_model_parallel_size = 16
 
@@ -217,8 +217,8 @@ def nemotron_3_super_pretrain_64gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config(*, num_gpus: int = 256) -> ConfigContainer:
-    """Nemotron 3 Ultra pretrain: 256× VR200, MXFP8 (alias of GB300).
+def nemotron_3_ultra_pretrain_256gpu_vrnvl72_fp8mx_config(*, num_gpus: int = 256) -> ConfigContainer:
+    """Nemotron 3 Ultra pretrain: 256× VRNVL72, MXFP8 (alias of GB300).
 
     Args:
         num_gpus: Launch GPU count used to size HSDP optimizer instances.
@@ -248,7 +248,7 @@ def nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config(*, num_gpus: int = 256)
         # Transformer Engine overlap settings for this model.
         "NVTE_BWD_LAYERNORM_SM_MARGIN": 20,
         "NVTE_FWD_LAYERNORM_SM_MARGIN": 20,
-        # Use cuDNN normalization kernels for the VR200 performance candidate.
+        # Use cuDNN normalization kernels for the VRNVL72 performance candidate.
         "NVTE_NORM_BWD_USE_CUDNN": 1,
         "NVTE_NORM_FWD_USE_CUDNN": 1,
         # Required by fine-grained activation offloading to avoid offloading weights.
@@ -259,8 +259,8 @@ def nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config(*, num_gpus: int = 256)
     return cfg
 
 
-def nemotron_3_ultra_pretrain_256gpu_vr200_nvfp4_config(*, num_gpus: int = 256) -> ConfigContainer:
-    """Nemotron 3 Ultra pretrain: 256× VR200, NVFP4 (alias of GB300).
+def nemotron_3_ultra_pretrain_256gpu_vrnvl72_nvfp4_config(*, num_gpus: int = 256) -> ConfigContainer:
+    """Nemotron 3 Ultra pretrain: 256× VRNVL72, NVFP4 (alias of GB300).
 
     Args:
         num_gpus: Launch GPU count used to size HSDP optimizer instances.
@@ -286,7 +286,7 @@ def nemotron_3_ultra_pretrain_256gpu_vr200_nvfp4_config(*, num_gpus: int = 256)
         # Transformer Engine overlap settings for this model.
         "NVTE_BWD_LAYERNORM_SM_MARGIN": 20,
         "NVTE_FWD_LAYERNORM_SM_MARGIN": 20,
-        # Use cuDNN normalization kernels for the VR200 performance candidate.
+        # Use cuDNN normalization kernels for the VRNVL72 performance candidate.
         "NVTE_NORM_BWD_USE_CUDNN": 1,
         "NVTE_NORM_FWD_USE_CUDNN": 1,
         # Required by fine-grained activation offloading to avoid offloading weights.
@@ -299,8 +299,8 @@ def nemotron_3_ultra_pretrain_256gpu_vr200_nvfp4_config(*, num_gpus: int = 256)
     return cfg
 
 
-def nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
-    """Nemotron 3.5 Lightning pretrain: 8× VR200, BF16 (alias of GB300)."""
+def nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Nemotron 3.5 Lightning pretrain: 8× VRNVL72, BF16 (alias of GB300)."""
     cfg = _build_nemotron_3_5_lightning_gb300_bf16()
     # Offload MoE expert activations to host memory: the current Rubin base image ships a Transformer
     # Engine without TE PR #3273, whose fused cross entropy holds an extra fp32 gradient per loss head.
@@ -309,7 +309,7 @@ def nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
     cfg.model.fine_grained_activation_offloading = True
     cfg.model.offload_modules = ["expert_fc1", "moe_act"]
 
-    # Keep the VR200 launch environment explicit instead of inheriting it from GB300.
+    # Keep the VRNVL72 launch environment explicit instead of inheriting it from GB300.
     cfg.env_vars = {
         **COMMON_PERF_ENV_VARS,
         "CUDA_DEVICE_MAX_CONNECTIONS": 32,
@@ -330,15 +330,15 @@ def nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Nemotron 3.5 Lightning pretrain: 8× VR200, MXFP8 with full-iteration CUDA graph."""
+def nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Nemotron 3.5 Lightning pretrain: 8× VRNVL72, MXFP8 with full-iteration CUDA graph."""
     cfg = _build_nemotron_3_5_lightning_gb300_mxfp8()
     cfg.model.recompute_modules = []
     cfg.model.use_transformer_engine_op_fuser = True
     cfg.mixed_precision.fp8_dot_product_attention = True
     _enable_nemotron_3_5_lightning_full_iteration(cfg)
 
-    # Keep the VR200 launch environment explicit instead of inheriting it from GB300.
+    # Keep the VRNVL72 launch environment explicit instead of inheriting it from GB300.
     cfg.env_vars = {
         **COMMON_PERF_ENV_VARS,
         "CUDA_DEVICE_MAX_CONNECTIONS": 32,
diff --git a/src/megatron/bridge/perf_recipes/qwen/__init__.py b/src/megatron/bridge/perf_recipes/qwen/__init__.py
index 75eb5edc42..f6b86289ea 100644
--- a/src/megatron/bridge/perf_recipes/qwen/__init__.py
+++ b/src/megatron/bridge/perf_recipes/qwen/__init__.py
@@ -97,11 +97,11 @@
     qwen3_next_80b_a3b_pretrain_128gpu_h100_bf16_config,
     qwen3_next_80b_a3b_pretrain_128gpu_h100_fp8cs_config,
 )
-from megatron.bridge.perf_recipes.qwen.vr200.qwen3_moe import (
-    qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config,
-    qwen3_30b_a3b_pretrain_8gpu_vr200_fp8mx_config,
-    qwen3_30b_a3b_pretrain_8gpu_vr200_nvfp4_config,
-    qwen3_235b_a22b_pretrain_256gpu_vr200_bf16_config,
-    qwen3_235b_a22b_pretrain_256gpu_vr200_fp8mx_config,
-    qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config,
+from megatron.bridge.perf_recipes.qwen.vrnvl72.qwen3_moe import (
+    qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config,
+    qwen3_30b_a3b_pretrain_8gpu_vrnvl72_fp8mx_config,
+    qwen3_30b_a3b_pretrain_8gpu_vrnvl72_nvfp4_config,
+    qwen3_235b_a22b_pretrain_256gpu_vrnvl72_bf16_config,
+    qwen3_235b_a22b_pretrain_256gpu_vrnvl72_fp8mx_config,
+    qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config,
 )
diff --git a/src/megatron/bridge/perf_recipes/qwen/gb300/qwen3_moe.py b/src/megatron/bridge/perf_recipes/qwen/gb300/qwen3_moe.py
index d3842d77c5..d6c24f0464 100644
--- a/src/megatron/bridge/perf_recipes/qwen/gb300/qwen3_moe.py
+++ b/src/megatron/bridge/perf_recipes/qwen/gb300/qwen3_moe.py
@@ -192,7 +192,7 @@ def qwen3_235b_a22b_pretrain_64gpu_gb300_fp8mx_config() -> ConfigContainer:
 
 
 def _build_qwen3_30b_a3b_gb300_bf16() -> ConfigContainer:
-    """Shared HybridEP BF16 base for the Qwen3 30B-A3B 8-GPU GB300 recipe and its VR200 alias."""
+    """Shared HybridEP BF16 base for the Qwen3 30B-A3B 8-GPU GB300 recipe and its VRNVL72 alias."""
     cfg = qwen3_30b_a3b_pretrain_config()
     cfg.mixed_precision = _perf_precision("bf16")
     cfg.model.bias_activation_fusion = True
@@ -308,7 +308,7 @@ def qwen3_30b_a3b_pretrain_8gpu_gb300_fp8cs_config() -> ConfigContainer:
 
 
 def _build_qwen3_30b_a3b_gb300_fp8mx() -> ConfigContainer:
-    """Shared HybridEP MXFP8 base for the Qwen3 30B-A3B 8-GPU GB300 recipe and its VR200 alias."""
+    """Shared HybridEP MXFP8 base for the Qwen3 30B-A3B 8-GPU GB300 recipe and its VRNVL72 alias."""
     cfg = qwen3_30b_a3b_pretrain_config()
     cfg.mixed_precision = _perf_precision("fp8_mx")
     cfg.model.bias_activation_fusion = True
@@ -368,7 +368,7 @@ def qwen3_30b_a3b_pretrain_8gpu_gb300_fp8mx_config() -> ConfigContainer:
 
 
 def _build_qwen3_235b_a22b_gb300_bf16() -> ConfigContainer:
-    """Shared HybridEP BF16 base for the Qwen3 235B-A22B 256-GPU GB300 recipe and its VR200 alias."""
+    """Shared HybridEP BF16 base for the Qwen3 235B-A22B 256-GPU GB300 recipe and its VRNVL72 alias."""
     cfg = qwen3_235b_a22b_pretrain_config()
     cfg.mixed_precision = _perf_precision("bf16")
     cfg.model.bias_activation_fusion = True
@@ -488,7 +488,7 @@ def qwen3_235b_a22b_pretrain_256gpu_gb300_fp8cs_config() -> ConfigContainer:
 
 
 def _build_qwen3_235b_a22b_gb300_fp8mx() -> ConfigContainer:
-    """Shared HybridEP MXFP8 base for the Qwen3 235B-A22B 256-GPU GB300 recipe and its VR200 alias."""
+    """Shared HybridEP MXFP8 base for the Qwen3 235B-A22B 256-GPU GB300 recipe and its VRNVL72 alias."""
     cfg = qwen3_235b_a22b_pretrain_config()
     cfg.mixed_precision = _perf_precision("fp8_mx")
     cfg.model.bias_activation_fusion = True
@@ -629,7 +629,7 @@ def qwen3_235b_a22b_pretrain_64gpu_gb300_nvfp4_full_iteration_config() -> Config
 
 
 def _build_qwen3_235b_a22b_gb300_nvfp4() -> ConfigContainer:
-    """Shared HybridEP NVFP4 base (full-iteration CG, A2A overlap) for the GB300 recipe and its VR200 alias."""
+    """Shared HybridEP NVFP4 base (full-iteration CG, A2A overlap) for the GB300 recipe and its VRNVL72 alias."""
     cfg = qwen3_235b_a22b_pretrain_256gpu_gb300_fp8cs_config()
     cfg.mixed_precision = _perf_precision("nvfp4")
     cfg.comm_overlap.tp_comm_overlap = False
diff --git a/src/megatron/bridge/perf_recipes/qwen/vr200/__init__.py b/src/megatron/bridge/perf_recipes/qwen/vrnvl72/__init__.py
similarity index 100%
rename from src/megatron/bridge/perf_recipes/qwen/vr200/__init__.py
rename to src/megatron/bridge/perf_recipes/qwen/vrnvl72/__init__.py
diff --git a/src/megatron/bridge/perf_recipes/qwen/vr200/qwen3_moe.py b/src/megatron/bridge/perf_recipes/qwen/vrnvl72/qwen3_moe.py
similarity index 88%
rename from src/megatron/bridge/perf_recipes/qwen/vr200/qwen3_moe.py
rename to src/megatron/bridge/perf_recipes/qwen/vrnvl72/qwen3_moe.py
index a642b92110..ac548cb2fe 100644
--- a/src/megatron/bridge/perf_recipes/qwen/vr200/qwen3_moe.py
+++ b/src/megatron/bridge/perf_recipes/qwen/vrnvl72/qwen3_moe.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for Qwen3 MoE."""
+"""VRNVL72 performance recipes for Qwen3 MoE."""
 
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 from megatron.bridge.perf_recipes.qwen.common import ConfigContainer
@@ -25,8 +25,8 @@
 )
 
 
-def qwen3_235b_a22b_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
-    """Qwen3 235B A22B pretrain: 256× VR200, BF16 (alias of GB300)."""
+def qwen3_235b_a22b_pretrain_256gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Qwen3 235B A22B pretrain: 256× VRNVL72, BF16 (alias of GB300)."""
     cfg = _build_qwen3_235b_a22b_gb300_bf16()
     cfg.model.cuda_graph_scope = ["moe_router", "moe_preprocess"]
 
@@ -53,8 +53,8 @@ def qwen3_235b_a22b_pretrain_256gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def qwen3_235b_a22b_pretrain_256gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Qwen3 235B A22B pretrain: 256× VR200, FP8-MX (alias of GB300)."""
+def qwen3_235b_a22b_pretrain_256gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Qwen3 235B A22B pretrain: 256× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = _build_qwen3_235b_a22b_gb300_fp8mx()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -81,8 +81,8 @@ def qwen3_235b_a22b_pretrain_256gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """Qwen3 235B A22B pretrain: 256× VR200, NVFP4, full-iteration CG and A2A overlap."""
+def qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """Qwen3 235B A22B pretrain: 256× VRNVL72, NVFP4, full-iteration CG and A2A overlap."""
     cfg = _build_qwen3_235b_a22b_gb300_nvfp4()
 
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
@@ -112,8 +112,8 @@ def qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config() -> ConfigContainer:
     return cfg
 
 
-def qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
-    """Qwen3 30B-A3B pretrain: 8× VR200, BF16 (alias of GB300)."""
+def qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Qwen3 30B-A3B pretrain: 8× VRNVL72, BF16 (alias of GB300)."""
     cfg = _build_qwen3_30b_a3b_gb300_bf16()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -138,8 +138,8 @@ def qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def qwen3_30b_a3b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Qwen3 30B-A3B pretrain: 8× VR200, FP8-MX (alias of GB300)."""
+def qwen3_30b_a3b_pretrain_8gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Qwen3 30B-A3B pretrain: 8× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = _build_qwen3_30b_a3b_gb300_fp8mx()
 
     cfg.train.micro_batch_size = 4
@@ -173,8 +173,8 @@ def qwen3_30b_a3b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def qwen3_30b_a3b_pretrain_8gpu_vr200_nvfp4_config() -> ConfigContainer:
-    """Qwen3 30B-A3B pretrain: 8× VR200, NVFP4 (same layout as GB300).
+def qwen3_30b_a3b_pretrain_8gpu_vrnvl72_nvfp4_config() -> ConfigContainer:
+    """Qwen3 30B-A3B pretrain: 8× VRNVL72, NVFP4 (same layout as GB300).
 
     NVFP4's fp4_param_gather path is incompatible with TP comm overlap, so it
     is disabled here.
diff --git a/src/megatron/bridge/perf_recipes/qwen_vl/__init__.py b/src/megatron/bridge/perf_recipes/qwen_vl/__init__.py
index 1ac76fdec8..ddca7d8ea0 100644
--- a/src/megatron/bridge/perf_recipes/qwen_vl/__init__.py
+++ b/src/megatron/bridge/perf_recipes/qwen_vl/__init__.py
@@ -80,14 +80,14 @@
     qwen35_vl_397b_a17b_pretrain_256gpu_h100_bf16_config,
     qwen35_vl_397b_a17b_pretrain_256gpu_h100_fp8cs_config,
 )
-from megatron.bridge.perf_recipes.qwen_vl.vr200.qwen35_vl import (
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_bf16_config,
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8cs_config,
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8mx_config,
-    qwen35_vl_122b_a10b_pretrain_32gpu_vr200_bf16_config,
-    qwen35_vl_122b_a10b_pretrain_32gpu_vr200_fp8cs_config,
-    qwen35_vl_122b_a10b_pretrain_32gpu_vr200_fp8mx_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_bf16_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8cs_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8mx_config,
+from megatron.bridge.perf_recipes.qwen_vl.vrnvl72.qwen35_vl import (
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_bf16_config,
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8cs_config,
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8mx_config,
+    qwen35_vl_122b_a10b_pretrain_32gpu_vrnvl72_bf16_config,
+    qwen35_vl_122b_a10b_pretrain_32gpu_vrnvl72_fp8cs_config,
+    qwen35_vl_122b_a10b_pretrain_32gpu_vrnvl72_fp8mx_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_bf16_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8cs_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8mx_config,
 )
diff --git a/src/megatron/bridge/perf_recipes/qwen_vl/vr200/__init__.py b/src/megatron/bridge/perf_recipes/qwen_vl/vrnvl72/__init__.py
similarity index 100%
rename from src/megatron/bridge/perf_recipes/qwen_vl/vr200/__init__.py
rename to src/megatron/bridge/perf_recipes/qwen_vl/vrnvl72/__init__.py
diff --git a/src/megatron/bridge/perf_recipes/qwen_vl/vr200/qwen35_vl.py b/src/megatron/bridge/perf_recipes/qwen_vl/vrnvl72/qwen35_vl.py
similarity index 86%
rename from src/megatron/bridge/perf_recipes/qwen_vl/vr200/qwen35_vl.py
rename to src/megatron/bridge/perf_recipes/qwen_vl/vrnvl72/qwen35_vl.py
index b11ebf5dc3..732c88d29d 100644
--- a/src/megatron/bridge/perf_recipes/qwen_vl/vr200/qwen35_vl.py
+++ b/src/megatron/bridge/perf_recipes/qwen_vl/vrnvl72/qwen35_vl.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""VR200 performance recipes for Qwen3.5-VL."""
+"""VRNVL72 performance recipes for Qwen3.5-VL."""
 
 from megatron.bridge.perf_recipes.environment import COMMON_PERF_ENV_VARS
 from megatron.bridge.perf_recipes.qwen_vl.common import ConfigContainer, _select_gdn_kernel_backend
@@ -28,10 +28,10 @@
 )
 
 
-def qwen35_vl_35b_a3b_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
-    """Qwen3.5-VL 35B-A3B pretrain: 8× VR200, BF16 (alias of GB300)."""
+def qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Qwen3.5-VL 35B-A3B pretrain: 8× VRNVL72, BF16 (alias of GB300)."""
     cfg = _build_qwen35_vl_35b_a3b_gb300_bf16()
-    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VR200; keep FLA until it is.
+    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VRNVL72; keep FLA until it is.
     _select_gdn_kernel_backend(cfg, "fla")
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -56,10 +56,10 @@ def qwen35_vl_35b_a3b_pretrain_8gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8cs_config() -> ConfigContainer:
-    """Qwen3.5-VL 35B-A3B pretrain: 8× VR200, FP8-CS (alias of GB300)."""
+def qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8cs_config() -> ConfigContainer:
+    """Qwen3.5-VL 35B-A3B pretrain: 8× VRNVL72, FP8-CS (alias of GB300)."""
     cfg = qwen35_vl_35b_a3b_pretrain_8gpu_gb300_fp8cs_config()
-    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VR200; keep FLA until it is.
+    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VRNVL72; keep FLA until it is.
     _select_gdn_kernel_backend(cfg, "fla")
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -84,10 +84,10 @@ def qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8cs_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Qwen3.5-VL 35B-A3B pretrain: 8× VR200, FP8-MX (alias of GB300)."""
+def qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Qwen3.5-VL 35B-A3B pretrain: 8× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = qwen35_vl_35b_a3b_pretrain_8gpu_gb300_fp8mx_config()
-    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VR200; keep FLA until it is.
+    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VRNVL72; keep FLA until it is.
     _select_gdn_kernel_backend(cfg, "fla")
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -112,8 +112,8 @@ def qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_122b_a10b_pretrain_32gpu_vr200_bf16_config() -> ConfigContainer:
-    """Qwen3.5-VL 122B-A10B pretrain: 32× VR200, BF16 (alias of GB300)."""
+def qwen35_vl_122b_a10b_pretrain_32gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Qwen3.5-VL 122B-A10B pretrain: 32× VRNVL72, BF16 (alias of GB300)."""
     cfg = qwen35_vl_122b_a10b_pretrain_32gpu_gb300_bf16_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -138,8 +138,8 @@ def qwen35_vl_122b_a10b_pretrain_32gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_122b_a10b_pretrain_32gpu_vr200_fp8cs_config() -> ConfigContainer:
-    """Qwen3.5-VL 122B-A10B pretrain: 32× VR200, FP8-CS (alias of GB300)."""
+def qwen35_vl_122b_a10b_pretrain_32gpu_vrnvl72_fp8cs_config() -> ConfigContainer:
+    """Qwen3.5-VL 122B-A10B pretrain: 32× VRNVL72, FP8-CS (alias of GB300)."""
     cfg = qwen35_vl_122b_a10b_pretrain_32gpu_gb300_fp8cs_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -164,8 +164,8 @@ def qwen35_vl_122b_a10b_pretrain_32gpu_vr200_fp8cs_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_122b_a10b_pretrain_32gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Qwen3.5-VL 122B-A10B pretrain: 32× VR200, FP8-MX (alias of GB300)."""
+def qwen35_vl_122b_a10b_pretrain_32gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Qwen3.5-VL 122B-A10B pretrain: 32× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = qwen35_vl_122b_a10b_pretrain_32gpu_gb300_fp8mx_config()
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -190,10 +190,10 @@ def qwen35_vl_122b_a10b_pretrain_32gpu_vr200_fp8mx_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_397b_a17b_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
-    """Qwen3.5-VL 397B-A17B pretrain: 64× VR200, BF16 (alias of GB300)."""
+def qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_bf16_config() -> ConfigContainer:
+    """Qwen3.5-VL 397B-A17B pretrain: 64× VRNVL72, BF16 (alias of GB300)."""
     cfg = qwen35_vl_397b_a17b_pretrain_64gpu_gb300_bf16_config()
-    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VR200; keep FLA until it is.
+    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VRNVL72; keep FLA until it is.
     _select_gdn_kernel_backend(cfg, "fla")
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -218,10 +218,10 @@ def qwen35_vl_397b_a17b_pretrain_64gpu_vr200_bf16_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8cs_config() -> ConfigContainer:
-    """Qwen3.5-VL 397B-A17B pretrain: 64× VR200, FP8-CS (alias of GB300)."""
+def qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8cs_config() -> ConfigContainer:
+    """Qwen3.5-VL 397B-A17B pretrain: 64× VRNVL72, FP8-CS (alias of GB300)."""
     cfg = qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8cs_config()
-    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VR200; keep FLA until it is.
+    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VRNVL72; keep FLA until it is.
     _select_gdn_kernel_backend(cfg, "fla")
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
@@ -246,10 +246,10 @@ def qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8cs_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8mx_config() -> ConfigContainer:
-    """Qwen3.5-VL 397B-A17B pretrain: 64× VR200, FP8-MX (alias of GB300)."""
+def qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8mx_config() -> ConfigContainer:
+    """Qwen3.5-VL 397B-A17B pretrain: 64× VRNVL72, FP8-MX (alias of GB300)."""
     cfg = qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8mx_config()
-    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VR200; keep FLA until it is.
+    # The GB300 base selects the cuDNN GDN kernel, which has not been measured on VRNVL72; keep FLA until it is.
     _select_gdn_kernel_backend(cfg, "fla")
     # Keep process settings next to the recipe so users can see the exact benchmark environment.
     cfg.env_vars = {
diff --git a/tests/unit_tests/recipes/nemotronh/test_nemotron_3_ultra.py b/tests/unit_tests/recipes/nemotronh/test_nemotron_3_ultra.py
index 28d250f86b..efe243ebdb 100644
--- a/tests/unit_tests/recipes/nemotronh/test_nemotron_3_ultra.py
+++ b/tests/unit_tests/recipes/nemotronh/test_nemotron_3_ultra.py
@@ -19,8 +19,8 @@
 
 import megatron.bridge.recipes as recipes
 from megatron.bridge.perf_recipes.nemotronh import (
-    nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config,
-    nemotron_3_ultra_pretrain_256gpu_vr200_nvfp4_config,
+    nemotron_3_ultra_pretrain_256gpu_vrnvl72_fp8mx_config,
+    nemotron_3_ultra_pretrain_256gpu_vrnvl72_nvfp4_config,
 )
 from megatron.bridge.recipes.nemotronh.gb200.nemotron_3_ultra import (
     nemotron_3_ultra_pretrain_256gpu_gb200_bf16_ep16_config,
@@ -166,9 +166,9 @@ def test_gb200_ep16_pretrain_preserves_functional_policy() -> None:
 
 
 @pytest.mark.unit
-def test_vr200_perf_recipe_uses_nvl72_ultra_topology() -> None:
-    """VR200 Ultra preserves the GB300 execution layout with explicit NVL72 settings."""
-    cfg = nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config()
+def test_vrnvl72_perf_recipe_uses_nvl72_ultra_topology() -> None:
+    """VRNVL72 Ultra preserves the GB300 execution layout with explicit NVL72 settings."""
+    cfg = nemotron_3_ultra_pretrain_256gpu_vrnvl72_fp8mx_config()
 
     assert cfg.model.tensor_model_parallel_size == 1
     assert cfg.model.pipeline_model_parallel_size == 1
@@ -191,9 +191,9 @@ def test_vr200_perf_recipe_uses_nvl72_ultra_topology() -> None:
 
 
 @pytest.mark.unit
-def test_vr200_nvfp4_perf_recipe_uses_nvl72_ultra_topology() -> None:
-    """VR200 Ultra NVFP4 preserves the GB300 NVFP4 execution layout with explicit NVL72 settings."""
-    cfg = nemotron_3_ultra_pretrain_256gpu_vr200_nvfp4_config()
+def test_vrnvl72_nvfp4_perf_recipe_uses_nvl72_ultra_topology() -> None:
+    """VRNVL72 Ultra NVFP4 preserves the GB300 NVFP4 execution layout with explicit NVL72 settings."""
+    cfg = nemotron_3_ultra_pretrain_256gpu_vrnvl72_nvfp4_config()
 
     assert cfg.model.tensor_model_parallel_size == 1
     assert cfg.model.pipeline_model_parallel_size == 1
diff --git a/tests/unit_tests/recipes/test_deepseek_v3_proxy_recipes.py b/tests/unit_tests/recipes/test_deepseek_v3_proxy_recipes.py
index c22e1cb393..0e39115e16 100644
--- a/tests/unit_tests/recipes/test_deepseek_v3_proxy_recipes.py
+++ b/tests/unit_tests/recipes/test_deepseek_v3_proxy_recipes.py
@@ -38,10 +38,10 @@ def _offline_recipes(monkeypatch: pytest.MonkeyPatch) -> None:
 @pytest.mark.parametrize("precision", ["fp8mx", "nvfp4"])
 def test_proxy_preserves_parent_except_depth_and_layout(precision: str) -> None:
     parent_fn: Callable[[], ConfigContainer] = getattr(
-        deepseek, f"deepseek_v3_pretrain_256gpu_vr200_{precision}_config"
+        deepseek, f"deepseek_v3_pretrain_256gpu_vrnvl72_{precision}_config"
     )
     proxy_fn: Callable[[], ConfigContainer] = getattr(
-        deepseek, f"deepseek_v3_pretrain_64gpu_vr200_{precision}_proxy_config"
+        deepseek, f"deepseek_v3_pretrain_64gpu_vrnvl72_{precision}_proxy_config"
     )
     parent = parent_fn()
     proxy = proxy_fn()
@@ -111,7 +111,7 @@ def test_proxy_preserves_parent_except_depth_and_layout(precision: str) -> None:
 
 @pytest.mark.parametrize("precision", ["fp8mx", "nvfp4"])
 def test_proxy_overlap_setup_without_pipeline_parallelism(monkeypatch: pytest.MonkeyPatch, precision: str) -> None:
-    cfg = getattr(deepseek, f"deepseek_v3_pretrain_64gpu_vr200_{precision}_proxy_config")()
+    cfg = getattr(deepseek, f"deepseek_v3_pretrain_64gpu_vrnvl72_{precision}_proxy_config")()
     # Validate the topology without requiring the GPU libraries in this unit test.
     monkeypatch.setattr("megatron.bridge.training.comm_overlap.is_te_min_version", lambda *_: True)
     monkeypatch.setattr("megatron.bridge.training.comm_overlap.is_torch_min_version", lambda *_: True)
@@ -137,7 +137,7 @@ def test_proxy_is_available_through_performance_selector(monkeypatch: pytest.Mon
     monkeypatch.syspath_prepend(str(Path(__file__).resolve().parents[3] / "scripts" / "performance"))
     from utils.utils import get_perf_recipe_by_name, list_available_config_variants
 
-    cfg = get_perf_recipe_by_name("deepseek_v3", "pretrain", 64, "vr200", precision, config_variant="proxy")
+    cfg = get_perf_recipe_by_name("deepseek_v3", "pretrain", 64, "vrnvl72", precision, config_variant="proxy")
     assert cfg.train.global_batch_size == 4096
     assert cfg.model.num_layers == 13
     assert cfg.model.pipeline_model_parallel_size == 1
@@ -147,7 +147,7 @@ def test_proxy_is_available_through_performance_selector(monkeypatch: pytest.Mon
     assert "proxy" in list_available_config_variants(
         model_family_name="deepseek",
         model_recipe_name="deepseek_v3",
-        gpu="vr200",
+        gpu="vrnvl72",
         compute_dtype=precision,
         task="pretrain",
     )
diff --git a/tests/unit_tests/recipes/test_deepseek_v4_flash_perf_recipes.py b/tests/unit_tests/recipes/test_deepseek_v4_flash_perf_recipes.py
index 7f8d5ae988..3d14c89302 100644
--- a/tests/unit_tests/recipes/test_deepseek_v4_flash_perf_recipes.py
+++ b/tests/unit_tests/recipes/test_deepseek_v4_flash_perf_recipes.py
@@ -24,7 +24,7 @@
     deepseek_v4_flash_pretrain_128gpu_b300_fp8mx_config,
     deepseek_v4_flash_pretrain_128gpu_gb200_fp8mx_config,
     deepseek_v4_flash_pretrain_128gpu_gb300_fp8mx_config,
-    deepseek_v4_flash_pretrain_128gpu_vr200_fp8mx_config,
+    deepseek_v4_flash_pretrain_128gpu_vrnvl72_fp8mx_config,
 )
 from megatron.bridge.utils.cuda_graph import cuda_graph_module_names, is_full_iteration_cuda_graph
 from tests.unit_tests.recipes.recipe_test_utils import patch_recipe_construction_dependencies
@@ -136,24 +136,24 @@ def test_deepseek_v4_flash_128gpu_gb300_fp8mx_config() -> None:
     assert cfg.env_vars["NUM_OF_HYBRID_EP_RANKS_PER_NVLINK_DOMAIN"] == 32
 
 
-def test_deepseek_v4_flash_128gpu_vr200_fp8mx_disables_clamp_for_cutedsl() -> None:
-    vr200_cfg = deepseek_v4_flash_pretrain_128gpu_vr200_fp8mx_config()
+def test_deepseek_v4_flash_128gpu_vrnvl72_fp8mx_disables_clamp_for_cutedsl() -> None:
+    vrnvl72_cfg = deepseek_v4_flash_pretrain_128gpu_vrnvl72_fp8mx_config()
     gb300_cfg = deepseek_v4_flash_pretrain_128gpu_gb300_fp8mx_config()
 
-    assert vr200_cfg.model.activation_func_clamp_value is None
-    assert vr200_cfg.model.use_transformer_engine_op_fuser is True
-    assert vr200_cfg.model.moe_use_grouped_tensor is True
-    assert vr200_cfg.env_vars["NVTE_CUTEDSL_FUSED_GROUPED_MLP"] == 1
-    assert vr200_cfg.env_vars["NVTE_ALLOW_NONDETERMINISTIC_ALGO"] == 0
-    assert is_full_iteration_cuda_graph(vr200_cfg.model)
+    assert vrnvl72_cfg.model.activation_func_clamp_value is None
+    assert vrnvl72_cfg.model.use_transformer_engine_op_fuser is True
+    assert vrnvl72_cfg.model.moe_use_grouped_tensor is True
+    assert vrnvl72_cfg.env_vars["NVTE_CUTEDSL_FUSED_GROUPED_MLP"] == 1
+    assert vrnvl72_cfg.env_vars["NVTE_ALLOW_NONDETERMINISTIC_ALGO"] == 0
+    assert is_full_iteration_cuda_graph(vrnvl72_cfg.model)
 
     gb300_cfg.model.activation_func_clamp_value = None
     gb300_cfg.model.use_transformer_engine_op_fuser = True
     gb300_cfg.model.moe_use_grouped_tensor = True
-    assert vars(vr200_cfg.model) == vars(gb300_cfg.model)
-    for field in fields(vr200_cfg):
+    assert vars(vrnvl72_cfg.model) == vars(gb300_cfg.model)
+    for field in fields(vrnvl72_cfg):
         if field.name != "model":
-            assert getattr(vr200_cfg, field.name) == getattr(gb300_cfg, field.name), field.name
+            assert getattr(vrnvl72_cfg, field.name) == getattr(gb300_cfg, field.name), field.name
 
 
 def test_deepseek_v4_flash_128gpu_b300_fp8mx_config() -> None:
diff --git a/tests/unit_tests/recipes/test_deepseek_vr200_perf_recipes.py b/tests/unit_tests/recipes/test_deepseek_vrnvl72_perf_recipes.py
similarity index 77%
rename from tests/unit_tests/recipes/test_deepseek_vr200_perf_recipes.py
rename to tests/unit_tests/recipes/test_deepseek_vrnvl72_perf_recipes.py
index 0ebe10db5e..7b25607bc2 100644
--- a/tests/unit_tests/recipes/test_deepseek_vr200_perf_recipes.py
+++ b/tests/unit_tests/recipes/test_deepseek_vrnvl72_perf_recipes.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""Tests for DeepSeek V3 VR200 performance recipe topology."""
+"""Tests for DeepSeek V3 VRNVL72 performance recipe topology."""
 
 import importlib.util
 import sys
@@ -26,7 +26,7 @@
 pytestmark = pytest.mark.unit
 
 
-def _load_vr200_recipe_module(monkeypatch: pytest.MonkeyPatch) -> ModuleType:
+def _load_vrnvl72_recipe_module(monkeypatch: pytest.MonkeyPatch) -> ModuleType:
     common = ModuleType("megatron.bridge.perf_recipes.deepseek.common")
     common.ConfigContainer = SimpleNamespace
     common._benchmark_common = lambda _cfg: None
@@ -45,7 +45,7 @@ def _load_vr200_recipe_module(monkeypatch: pytest.MonkeyPatch) -> ModuleType:
     gb300 = ModuleType("megatron.bridge.perf_recipes.deepseek.gb300.deepseek_v3")
     for precision in ("bf16", "fp8cs", "fp8mx", "nvfp4"):
         setattr(gb300, f"deepseek_v3_pretrain_256gpu_gb300_{precision}_config", lambda: SimpleNamespace())
-        # HybridEP base builders shared with the VR200 aliases (the GB300 configs add NCCL EP on top).
+        # HybridEP base builders shared with the VRNVL72 aliases (the GB300 configs add NCCL EP on top).
         setattr(gb300, f"_build_deepseek_v3_gb300_{precision}", lambda: SimpleNamespace())
     monkeypatch.setitem(sys.modules, gb300.__name__, gb300)
 
@@ -53,8 +53,8 @@ def _load_vr200_recipe_module(monkeypatch: pytest.MonkeyPatch) -> ModuleType:
     environment.COMMON_PERF_ENV_VARS = {}
     monkeypatch.setitem(sys.modules, environment.__name__, environment)
 
-    path = Path(__file__).resolve().parents[3] / "src/megatron/bridge/perf_recipes/deepseek/vr200/deepseek_v3.py"
-    spec = importlib.util.spec_from_file_location("_deepseek_vr200_recipe_under_test", path)
+    path = Path(__file__).resolve().parents[3] / "src/megatron/bridge/perf_recipes/deepseek/vrnvl72/deepseek_v3.py"
+    spec = importlib.util.spec_from_file_location("_deepseek_vrnvl72_recipe_under_test", path)
     assert spec is not None and spec.loader is not None
     module = importlib.util.module_from_spec(spec)
     spec.loader.exec_module(module)
@@ -64,14 +64,16 @@ def _load_vr200_recipe_module(monkeypatch: pytest.MonkeyPatch) -> ModuleType:
 @pytest.mark.parametrize(
     "recipe_factory_name",
     (
-        "deepseek_v3_pretrain_128gpu_vr200_bf16_config",
-        "deepseek_v3_pretrain_128gpu_vr200_fp8cs_config",
-        "deepseek_v3_pretrain_128gpu_vr200_fp8mx_config",
-        "deepseek_v3_pretrain_128gpu_vr200_nvfp4_config",
+        "deepseek_v3_pretrain_128gpu_vrnvl72_bf16_config",
+        "deepseek_v3_pretrain_128gpu_vrnvl72_fp8cs_config",
+        "deepseek_v3_pretrain_128gpu_vrnvl72_fp8mx_config",
+        "deepseek_v3_pretrain_128gpu_vrnvl72_nvfp4_config",
     ),
 )
-def test_deepseek_v3_128gpu_vr200_topology_is_valid(recipe_factory_name: str, monkeypatch: pytest.MonkeyPatch) -> None:
-    module = _load_vr200_recipe_module(monkeypatch)
+def test_deepseek_v3_128gpu_vrnvl72_topology_is_valid(
+    recipe_factory_name: str, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    module = _load_vrnvl72_recipe_module(monkeypatch)
     recipe_factory = getattr(module, recipe_factory_name)
     cfg = recipe_factory()
 
diff --git a/tests/unit_tests/recipes/test_gb300_ncclep_perf_recipes.py b/tests/unit_tests/recipes/test_gb300_ncclep_perf_recipes.py
index 4f3b164988..ce024cb80e 100644
--- a/tests/unit_tests/recipes/test_gb300_ncclep_perf_recipes.py
+++ b/tests/unit_tests/recipes/test_gb300_ncclep_perf_recipes.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""Every GB300 MoE perf recipe switched to NCCL EP carries the full NCCL EP stack, and the VR200
+"""Every GB300 MoE perf recipe switched to NCCL EP carries the full NCCL EP stack, and the VRNVL72
 aliases built on the shared HybridEP bases stay on HybridEP."""
 
 import importlib
@@ -53,27 +53,27 @@
     ("qwen_vl.gb300.qwen35_vl", "qwen35_vl_35b_a3b_pretrain_8gpu_gb300_bf16_config"),
 )
 
-# VR200 aliases of the same recipes: they reuse the shared HybridEP builders and must not pick up NCCL EP.
-_VR200_ALIASES = (
-    ("deepseek.vr200.deepseek_v3", "deepseek_v3_pretrain_256gpu_vr200_bf16_config"),
-    ("deepseek.vr200.deepseek_v3", "deepseek_v3_pretrain_256gpu_vr200_fp8mx_config"),
-    ("deepseek.vr200.deepseek_v3", "deepseek_v3_pretrain_256gpu_vr200_nvfp4_config"),
-    ("gpt_oss.vr200.gpt_oss", "gpt_oss_120b_pretrain_64gpu_vr200_fp8mx_config"),
-    ("kimi.vr200.kimi_k2", "kimi_k2_pretrain_256gpu_vr200_fp8mx_config"),
-    ("nemotronh.vr200.nemotronh", "nemotron_3_nano_pretrain_8gpu_vr200_bf16_config"),
-    ("nemotronh.vr200.nemotronh", "nemotron_3_super_pretrain_64gpu_vr200_bf16_config"),
-    ("nemotronh.vr200.nemotronh", "nemotron_3_super_pretrain_64gpu_vr200_nvfp4_config"),
-    ("nemotronh.vr200.nemotronh", "nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config"),
-    ("nemotronh.vr200.nemotronh", "nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config"),
-    ("nemotronh.vr200.nemotronh", "nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config"),
-    ("deepseek.vr200.deepseek_v3", "deepseek_v3_pretrain_64gpu_vr200_fp8mx_proxy_config"),
-    ("deepseek.vr200.deepseek_v3", "deepseek_v3_pretrain_64gpu_vr200_nvfp4_proxy_config"),
-    ("qwen.vr200.qwen3_moe", "qwen3_235b_a22b_pretrain_256gpu_vr200_bf16_config"),
-    ("qwen.vr200.qwen3_moe", "qwen3_235b_a22b_pretrain_256gpu_vr200_fp8mx_config"),
-    ("qwen.vr200.qwen3_moe", "qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config"),
-    ("qwen.vr200.qwen3_moe", "qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config"),
-    ("qwen.vr200.qwen3_moe", "qwen3_30b_a3b_pretrain_8gpu_vr200_fp8mx_config"),
-    ("qwen_vl.vr200.qwen35_vl", "qwen35_vl_35b_a3b_pretrain_8gpu_vr200_bf16_config"),
+# VRNVL72 aliases of the same recipes: they reuse the shared HybridEP builders and must not pick up NCCL EP.
+_VRNVL72_ALIASES = (
+    ("deepseek.vrnvl72.deepseek_v3", "deepseek_v3_pretrain_256gpu_vrnvl72_bf16_config"),
+    ("deepseek.vrnvl72.deepseek_v3", "deepseek_v3_pretrain_256gpu_vrnvl72_fp8mx_config"),
+    ("deepseek.vrnvl72.deepseek_v3", "deepseek_v3_pretrain_256gpu_vrnvl72_nvfp4_config"),
+    ("gpt_oss.vrnvl72.gpt_oss", "gpt_oss_120b_pretrain_64gpu_vrnvl72_fp8mx_config"),
+    ("kimi.vrnvl72.kimi_k2", "kimi_k2_pretrain_256gpu_vrnvl72_fp8mx_config"),
+    ("nemotronh.vrnvl72.nemotronh", "nemotron_3_nano_pretrain_8gpu_vrnvl72_bf16_config"),
+    ("nemotronh.vrnvl72.nemotronh", "nemotron_3_super_pretrain_64gpu_vrnvl72_bf16_config"),
+    ("nemotronh.vrnvl72.nemotronh", "nemotron_3_super_pretrain_64gpu_vrnvl72_nvfp4_config"),
+    ("nemotronh.vrnvl72.nemotronh", "nemotron_3_ultra_pretrain_256gpu_vrnvl72_fp8mx_config"),
+    ("nemotronh.vrnvl72.nemotronh", "nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config"),
+    ("nemotronh.vrnvl72.nemotronh", "nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config"),
+    ("deepseek.vrnvl72.deepseek_v3", "deepseek_v3_pretrain_64gpu_vrnvl72_fp8mx_proxy_config"),
+    ("deepseek.vrnvl72.deepseek_v3", "deepseek_v3_pretrain_64gpu_vrnvl72_nvfp4_proxy_config"),
+    ("qwen.vrnvl72.qwen3_moe", "qwen3_235b_a22b_pretrain_256gpu_vrnvl72_bf16_config"),
+    ("qwen.vrnvl72.qwen3_moe", "qwen3_235b_a22b_pretrain_256gpu_vrnvl72_fp8mx_config"),
+    ("qwen.vrnvl72.qwen3_moe", "qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config"),
+    ("qwen.vrnvl72.qwen3_moe", "qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config"),
+    ("qwen.vrnvl72.qwen3_moe", "qwen3_30b_a3b_pretrain_8gpu_vrnvl72_fp8mx_config"),
+    ("qwen_vl.vrnvl72.qwen35_vl", "qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_bf16_config"),
 )
 
 
@@ -81,7 +81,7 @@
 # every recipe module that is already loaded, so a lazy import inside a test would escape the patch.
 _RECIPE_MODULES = {
     module_name: importlib.import_module(f"megatron.bridge.perf_recipes.{module_name}")
-    for module_name in sorted({name for name, _ in (*_GB300_NCCLEP_RECIPES, *_VR200_ALIASES)})
+    for module_name in sorted({name for name, _ in (*_GB300_NCCLEP_RECIPES, *_VRNVL72_ALIASES)})
 }
 
 
@@ -118,8 +118,8 @@ def test_gb300_deepseek_hsdp_uses_hybridep() -> None:
     assert cfg.env_vars.keys().isdisjoint(_NCCL_EP_ENV)
 
 
-@pytest.mark.parametrize(("module_name", "factory_name"), _VR200_ALIASES, ids=lambda value: value)
-def test_vr200_aliases_do_not_inherit_nccl_ep(module_name: str, factory_name: str) -> None:
+@pytest.mark.parametrize(("module_name", "factory_name"), _VRNVL72_ALIASES, ids=lambda value: value)
+def test_vrnvl72_aliases_do_not_inherit_nccl_ep(module_name: str, factory_name: str) -> None:
     cfg = _build(module_name, factory_name)
 
     # The aliases keep the dispatcher their HybridEP base had upstream (flex/hybridep, or alltoall for the
diff --git a/tests/unit_tests/recipes/test_moe_full_iteration_recipes.py b/tests/unit_tests/recipes/test_moe_full_iteration_recipes.py
index 23904cf812..3f34f5d418 100644
--- a/tests/unit_tests/recipes/test_moe_full_iteration_recipes.py
+++ b/tests/unit_tests/recipes/test_moe_full_iteration_recipes.py
@@ -18,16 +18,16 @@
 
 import pytest
 
-from megatron.bridge.perf_recipes.deepseek import deepseek_v3_pretrain_256gpu_vr200_nvfp4_config
+from megatron.bridge.perf_recipes.deepseek import deepseek_v3_pretrain_256gpu_vrnvl72_nvfp4_config
 from megatron.bridge.perf_recipes.nemotronh import (
     nemotron_3_super_pretrain_64gpu_gb200_fp8mx_config,
     nemotron_3_super_pretrain_64gpu_gb200_nvfp4_config,
     nemotron_3_super_pretrain_64gpu_gb300_fp8mx_config,
     nemotron_3_super_pretrain_64gpu_gb300_nvfp4_config,
-    nemotron_3_super_pretrain_64gpu_vr200_fp8mx_config,
-    nemotron_3_super_pretrain_64gpu_vr200_nvfp4_config,
+    nemotron_3_super_pretrain_64gpu_vrnvl72_fp8mx_config,
+    nemotron_3_super_pretrain_64gpu_vrnvl72_nvfp4_config,
 )
-from megatron.bridge.perf_recipes.qwen import qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config
+from megatron.bridge.perf_recipes.qwen import qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config
 from megatron.bridge.training.config import ConfigContainer
 from tests.unit_tests.recipes.recipe_test_utils import patch_recipe_construction_dependencies
 
@@ -72,8 +72,8 @@ def _assert_full_iteration_hybridep(cfg: ConfigContainer) -> None:
     assert "graph_capture_record_stream_reuse:True" in cfg.env_vars["PYTORCH_CUDA_ALLOC_CONF"]
 
 
-def test_qwen3_235b_vr200_nvfp4_uses_full_iteration_stack() -> None:
-    cfg = qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config()
+def test_qwen3_235b_vrnvl72_nvfp4_uses_full_iteration_stack() -> None:
+    cfg = qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config()
 
     _assert_full_iteration_hybridep(cfg)
     assert cfg.mixed_precision.fp4 == "e2m1"
@@ -94,8 +94,8 @@ def test_qwen3_235b_vr200_nvfp4_uses_full_iteration_stack() -> None:
     ) == (1, 4, 12, 32)
 
 
-def test_deepseek_v3_256gpu_vr200_nvfp4_uses_full_iteration_stack() -> None:
-    cfg = deepseek_v3_pretrain_256gpu_vr200_nvfp4_config()
+def test_deepseek_v3_256gpu_vrnvl72_nvfp4_uses_full_iteration_stack() -> None:
+    cfg = deepseek_v3_pretrain_256gpu_vrnvl72_nvfp4_config()
 
     _assert_full_iteration_hybridep(cfg)
     assert cfg.mixed_precision.fp4 == "e2m1"
@@ -124,7 +124,7 @@ def test_deepseek_v3_256gpu_vr200_nvfp4_uses_full_iteration_stack() -> None:
     [
         (nemotron_3_super_pretrain_64gpu_gb200_fp8mx_config, 2, 16),
         (nemotron_3_super_pretrain_64gpu_gb300_fp8mx_config, 1, 8),
-        (nemotron_3_super_pretrain_64gpu_vr200_fp8mx_config, 1, 8),
+        (nemotron_3_super_pretrain_64gpu_vrnvl72_fp8mx_config, 1, 8),
     ],
 )
 def test_nemotron_3_super_mxfp8_uses_full_iteration_stack(
@@ -162,7 +162,7 @@ def test_nemotron_3_super_mxfp8_uses_full_iteration_stack(
     [
         (nemotron_3_super_pretrain_64gpu_gb200_nvfp4_config, 2, 16),
         (nemotron_3_super_pretrain_64gpu_gb300_nvfp4_config, 1, 16),
-        (nemotron_3_super_pretrain_64gpu_vr200_nvfp4_config, 1, 16),
+        (nemotron_3_super_pretrain_64gpu_vrnvl72_nvfp4_config, 1, 16),
     ],
 )
 def test_nemotron_3_super_nvfp4_uses_full_iteration_stack(
diff --git a/tests/unit_tests/recipes/test_nemotron_3_5_lightning_perf_recipes.py b/tests/unit_tests/recipes/test_nemotron_3_5_lightning_perf_recipes.py
index fd84cbf23a..c973fc56f9 100644
--- a/tests/unit_tests/recipes/test_nemotron_3_5_lightning_perf_recipes.py
+++ b/tests/unit_tests/recipes/test_nemotron_3_5_lightning_perf_recipes.py
@@ -37,8 +37,8 @@
     nemotron_3_5_lightning_pretrain_8gpu_gb300_fp8mx_config,
     nemotron_3_5_lightning_pretrain_8gpu_gb300_fp8mx_fsdp_config,
     nemotron_3_5_lightning_pretrain_8gpu_gb300_nvfp4_config,
-    nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config,
-    nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config,
+    nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config,
+    nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config,
     nemotron_3_5_lightning_pretrain_16gpu_h100_bf16_config,
     nemotron_3_5_lightning_pretrain_16gpu_h100_fp8cs_config,
     nemotron_3_nano_pretrain_8gpu_b200_bf16_config,
@@ -110,11 +110,11 @@
 )
 _FULL_ITERATION_MXFP8_RECIPES = (
     nemotron_3_5_lightning_pretrain_8gpu_gb300_fp8mx_config,
-    nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config,
+    nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config,
 )
-_VR200_RECIPES = (
-    nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config,
-    nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config,
+_VRNVL72_RECIPES = (
+    nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config,
+    nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config,
 )
 _NEMOTRON_3_RECIPES = (
     nemotron_3_nano_pretrain_16gpu_h100_bf16_config,
@@ -205,9 +205,9 @@
     ("megatron.bridge.perf_recipes.nemotronh.gb300.nemotronh", "nemotron_3_nano_pretrain_8gpu_gb300_bf16_config"),
     ("megatron.bridge.perf_recipes.nemotronh.gb300.nemotronh", "nemotron_3_nano_pretrain_8gpu_gb300_fp8mx_config"),
     ("megatron.bridge.perf_recipes.nemotronh.gb300.nemotronh", "nemotron_3_nano_pretrain_8gpu_gb300_nvfp4_config"),
-    ("megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh", "nemotron_3_nano_pretrain_8gpu_vr200_bf16_config"),
-    ("megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh", "nemotron_3_nano_pretrain_8gpu_vr200_fp8mx_config"),
-    ("megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh", "nemotron_3_nano_pretrain_8gpu_vr200_nvfp4_config"),
+    ("megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh", "nemotron_3_nano_pretrain_8gpu_vrnvl72_bf16_config"),
+    ("megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh", "nemotron_3_nano_pretrain_8gpu_vrnvl72_fp8mx_config"),
+    ("megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh", "nemotron_3_nano_pretrain_8gpu_vrnvl72_nvfp4_config"),
     (
         "megatron.bridge.perf_recipes.nemotronh.h100.nemotronh",
         "nemotron_3_5_lightning_pretrain_16gpu_h100_bf16_config",
@@ -273,12 +273,12 @@
         "nemotron_3_5_lightning_pretrain_8gpu_gb300_nvfp4_config",
     ),
     (
-        "megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh",
-        "nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config",
+        "megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh",
+        "nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config",
     ),
     (
-        "megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh",
-        "nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config",
+        "megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh",
+        "nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config",
     ),
 )
 
@@ -312,7 +312,7 @@ def test_standard_perf_recipes_do_not_expose_mtp_flag(recipe_factory: Callable[[
         *_GB200_RECIPES,
         *_GB300_RECIPES,
         *_GB_FSDP_RECIPES,
-        *_VR200_RECIPES,
+        *_VRNVL72_RECIPES,
     ),
     ids=lambda recipe: recipe.__name__,
 )
@@ -621,46 +621,46 @@ def test_gb300_perf_recipe_topology(recipe_factory: Callable[[], ConfigContainer
 
 
 @pytest.mark.parametrize(
-    ("vr200_factory", "gb300_factory"),
+    ("vrnvl72_factory", "gb300_factory"),
     (
         (
-            nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config,
+            nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config,
             nemotron_3_5_lightning_pretrain_8gpu_gb300_bf16_config,
         ),
         (
-            nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config,
+            nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config,
             nemotron_3_5_lightning_pretrain_8gpu_gb300_fp8mx_config,
         ),
     ),
     ids=lambda value: value.__name__,
 )
-def test_vr200_perf_recipes_match_gb300_configs(
-    vr200_factory: Callable[[], ConfigContainer],
+def test_vrnvl72_perf_recipes_match_gb300_configs(
+    vrnvl72_factory: Callable[[], ConfigContainer],
     gb300_factory: Callable[[], ConfigContainer],
 ) -> None:
-    """VR200 Nemotron 3.5 recipes match their GB300 baselines up to the GB300 NCCL EP overlay.
+    """VRNVL72 Nemotron 3.5 recipes match their GB300 baselines up to the GB300 NCCL EP overlay.
 
     The GB300 recipes default to NCCL EP (dispatcher backend, device-side expert counts and the
-    NCCL EP process setting); the VR200 aliases keep the shared HybridEP base. The VR200 BF16
+    NCCL EP process setting); the VRNVL72 aliases keep the shared HybridEP base. The VRNVL72 BF16
     recipe also offloads expert activations. Everything else must be identical.
     """
-    vr200_cfg = vr200_factory()
+    vrnvl72_cfg = vrnvl72_factory()
     gb300_cfg = gb300_factory()
-    assert vr200_cfg.model.moe_flex_dispatcher_backend == "hybridep"
+    assert vrnvl72_cfg.model.moe_flex_dispatcher_backend == "hybridep"
     assert gb300_cfg.model.moe_flex_dispatcher_backend == "ncclep"
 
-    if vr200_factory is nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config:
-        assert vr200_cfg.model.fine_grained_activation_offloading is True
-        assert vr200_cfg.model.offload_modules == ["expert_fc1", "moe_act"]
-        assert vr200_cfg.env_vars["NVTE_CPU_OFFLOAD_V1"] == 1
-        assert "graph_capture_record_stream_reuse:True" in vr200_cfg.env_vars["PYTORCH_CUDA_ALLOC_CONF"]
-        vr200_cfg.model.fine_grained_activation_offloading = gb300_cfg.model.fine_grained_activation_offloading
-        vr200_cfg.model.offload_modules = gb300_cfg.model.offload_modules
-
-    _enable_ncclep(vr200_cfg)
-    vr200_cfg.model.moe_use_grouped_tensor = True
-    vr200_cfg.env_vars = gb300_cfg.env_vars
-    assert vr200_cfg == gb300_cfg
+    if vrnvl72_factory is nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config:
+        assert vrnvl72_cfg.model.fine_grained_activation_offloading is True
+        assert vrnvl72_cfg.model.offload_modules == ["expert_fc1", "moe_act"]
+        assert vrnvl72_cfg.env_vars["NVTE_CPU_OFFLOAD_V1"] == 1
+        assert "graph_capture_record_stream_reuse:True" in vrnvl72_cfg.env_vars["PYTORCH_CUDA_ALLOC_CONF"]
+        vrnvl72_cfg.model.fine_grained_activation_offloading = gb300_cfg.model.fine_grained_activation_offloading
+        vrnvl72_cfg.model.offload_modules = gb300_cfg.model.offload_modules
+
+    _enable_ncclep(vrnvl72_cfg)
+    vrnvl72_cfg.model.moe_use_grouped_tensor = True
+    vrnvl72_cfg.env_vars = gb300_cfg.env_vars
+    assert vrnvl72_cfg == gb300_cfg
 
 
 @pytest.mark.parametrize("recipe_factory", _GB_FSDP_RECIPES, ids=lambda recipe: recipe.__name__)
diff --git a/tests/unit_tests/recipes/test_nemotronh_recipes.py b/tests/unit_tests/recipes/test_nemotronh_recipes.py
index f1c62da2e3..a4ccb58481 100644
--- a/tests/unit_tests/recipes/test_nemotronh_recipes.py
+++ b/tests/unit_tests/recipes/test_nemotronh_recipes.py
@@ -201,11 +201,11 @@ def test_nemotron_3_nano_gb200_defers_vocab_size_to_training_tokenizer():
             id="nano-gb300",
         ),
         pytest.param(
-            "megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh",
-            "nemotron_3_nano_pretrain_8gpu_vr200_fp8mx_config",
+            "megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh",
+            "nemotron_3_nano_pretrain_8gpu_vrnvl72_fp8mx_config",
             True,
             108,
-            id="nano-vr200",
+            id="nano-vrnvl72",
         ),
         pytest.param(
             "megatron.bridge.perf_recipes.nemotronh.gb200.nemotronh",
@@ -222,11 +222,11 @@ def test_nemotron_3_nano_gb200_defers_vocab_size_to_training_tokenizer():
             id="super-gb300",
         ),
         pytest.param(
-            "megatron.bridge.perf_recipes.nemotronh.vr200.nemotronh",
-            "nemotron_3_super_pretrain_64gpu_vr200_fp8mx_config",
+            "megatron.bridge.perf_recipes.nemotronh.vrnvl72.nemotronh",
+            "nemotron_3_super_pretrain_64gpu_vrnvl72_fp8mx_config",
             False,
             32,
-            id="super-vr200",
+            id="super-vrnvl72",
         ),
     ],
 )
diff --git a/tests/unit_tests/recipes/test_perf_recipe_environment.py b/tests/unit_tests/recipes/test_perf_recipe_environment.py
index 91a2832067..ccf490fe5e 100644
--- a/tests/unit_tests/recipes/test_perf_recipe_environment.py
+++ b/tests/unit_tests/recipes/test_perf_recipe_environment.py
@@ -46,23 +46,23 @@
     "QUANTIZATION_TYPE_DEBUG",
     "TORCHINDUCTOR_WORKER_START",
 }
-_VR200_CUDNN_LAYERNORM_RECIPES = {
-    "deepseek_v3_pretrain_64gpu_vr200_fp8mx_proxy_config",
-    "deepseek_v3_pretrain_128gpu_vr200_fp8mx_config",
-    "deepseek_v3_pretrain_256gpu_vr200_fp8mx_config",
-    "deepseek_v4_flash_pretrain_128gpu_vr200_fp8mx_config",
-    "gpt_oss_20b_pretrain_8gpu_vr200_fp8mx_config",
-    "gpt_oss_20b_pretrain_8gpu_vr200_nvfp4_config",
-    "gpt_oss_20b_pretrain_64gpu_vr200_nvfp4_config",
-    "kimi_k2_pretrain_256gpu_vr200_fp8mx_config",
-    "llama3_70b_pretrain_64gpu_vr200_bf16_config",
-    "nemotron_3_nano_pretrain_8gpu_vr200_bf16_config",
-    "nemotron_3_nano_pretrain_8gpu_vr200_fp8mx_config",
-    "nemotron_3_nano_pretrain_8gpu_vr200_nvfp4_config",
-    "nemotron_3_5_lightning_pretrain_8gpu_vr200_bf16_config",
-    "nemotron_3_5_lightning_pretrain_8gpu_vr200_fp8mx_config",
-    "nemotron_3_ultra_pretrain_256gpu_vr200_fp8mx_config",
-    "nemotron_3_ultra_pretrain_256gpu_vr200_nvfp4_config",
+_VRNVL72_CUDNN_LAYERNORM_RECIPES = {
+    "deepseek_v3_pretrain_64gpu_vrnvl72_fp8mx_proxy_config",
+    "deepseek_v3_pretrain_128gpu_vrnvl72_fp8mx_config",
+    "deepseek_v3_pretrain_256gpu_vrnvl72_fp8mx_config",
+    "deepseek_v4_flash_pretrain_128gpu_vrnvl72_fp8mx_config",
+    "gpt_oss_20b_pretrain_8gpu_vrnvl72_fp8mx_config",
+    "gpt_oss_20b_pretrain_8gpu_vrnvl72_nvfp4_config",
+    "gpt_oss_20b_pretrain_64gpu_vrnvl72_nvfp4_config",
+    "kimi_k2_pretrain_256gpu_vrnvl72_fp8mx_config",
+    "llama3_70b_pretrain_64gpu_vrnvl72_bf16_config",
+    "nemotron_3_nano_pretrain_8gpu_vrnvl72_bf16_config",
+    "nemotron_3_nano_pretrain_8gpu_vrnvl72_fp8mx_config",
+    "nemotron_3_nano_pretrain_8gpu_vrnvl72_nvfp4_config",
+    "nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_bf16_config",
+    "nemotron_3_5_lightning_pretrain_8gpu_vrnvl72_fp8mx_config",
+    "nemotron_3_ultra_pretrain_256gpu_vrnvl72_fp8mx_config",
+    "nemotron_3_ultra_pretrain_256gpu_vrnvl72_nvfp4_config",
 }
 
 
@@ -195,9 +195,9 @@ def test_explicit_environment_invariants_across_all_flat_recipes():
 
         cudnn_names = {"NVTE_NORM_BWD_USE_CUDNN", "NVTE_NORM_FWD_USE_CUDNN"}
         assert environment.keys().isdisjoint(cudnn_names) or environment.keys() >= cudnn_names
-        if path.parent.name == "vr200":
+        if path.parent.name == "vrnvl72":
             assert environment["CUDA_DEVICE_MAX_CONNECTIONS"] == 32
-            uses_cudnn_layernorm = function_name in _VR200_CUDNN_LAYERNORM_RECIPES
+            uses_cudnn_layernorm = function_name in _VRNVL72_CUDNN_LAYERNORM_RECIPES
             assert (environment.keys() >= cudnn_names) == uses_cudnn_layernorm
             if uses_cudnn_layernorm:
                 assert all(environment[name] == 1 for name in cudnn_names)
@@ -206,7 +206,7 @@ def test_explicit_environment_invariants_across_all_flat_recipes():
         assert not hybrid_ep_names or hybrid_ep_names == _HYBRID_EP_ENV_NAMES
         if hybrid_ep_names:
             gpu = path.parent.name
-            nvlink_domain_size = 72 if gpu in {"gb200", "gb300", "vr200"} else 8
+            nvlink_domain_size = 72 if gpu in {"gb200", "gb300", "vrnvl72"} else 8
             assert environment["NVLINK_DOMAIN_SIZE"] == nvlink_domain_size
             assert environment["USE_MNNVL"] == int(nvlink_domain_size == 72)
             assert environment["NUM_OF_HYBRID_EP_RANKS_PER_NVLINK_DOMAIN"] <= nvlink_domain_size
diff --git a/tests/unit_tests/recipes/test_qwen_perf_recipes.py b/tests/unit_tests/recipes/test_qwen_perf_recipes.py
index 039bdfe541..73e13b317c 100644
--- a/tests/unit_tests/recipes/test_qwen_perf_recipes.py
+++ b/tests/unit_tests/recipes/test_qwen_perf_recipes.py
@@ -31,7 +31,7 @@
     qwen3_30b_a3b_pretrain_8gpu_gb200_nvfp4_config,
     qwen3_30b_a3b_pretrain_8gpu_gb300_fp8cs_config,
     qwen3_30b_a3b_pretrain_8gpu_gb300_nvfp4_config,
-    qwen3_30b_a3b_pretrain_8gpu_vr200_nvfp4_config,
+    qwen3_30b_a3b_pretrain_8gpu_vrnvl72_nvfp4_config,
     qwen3_235b_a22b_pretrain_64gpu_b200_fp8cs_config,
     qwen3_235b_a22b_pretrain_64gpu_b200_nvfp4_config,
     qwen3_235b_a22b_pretrain_64gpu_b300_fp8cs_config,
@@ -52,7 +52,7 @@
     qwen3_235b_a22b_pretrain_256gpu_gb200_nvfp4_config,
     qwen3_235b_a22b_pretrain_256gpu_gb300_fp8cs_config,
     qwen3_235b_a22b_pretrain_256gpu_gb300_nvfp4_config,
-    qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config,
+    qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config,
 )
 
 
@@ -85,7 +85,7 @@ def test_qwen3_235b_64gpu_blackwell_topology_is_valid(
         qwen3_30b_a3b_pretrain_8gpu_gb300_nvfp4_config,
         qwen3_30b_a3b_pretrain_8gpu_b200_nvfp4_config,
         qwen3_30b_a3b_pretrain_8gpu_b300_nvfp4_config,
-        qwen3_30b_a3b_pretrain_8gpu_vr200_nvfp4_config,
+        qwen3_30b_a3b_pretrain_8gpu_vrnvl72_nvfp4_config,
         qwen3_235b_a22b_pretrain_64gpu_gb200_nvfp4_config,
         qwen3_235b_a22b_pretrain_256gpu_gb200_nvfp4_config,
         qwen3_235b_a22b_pretrain_64gpu_gb300_nvfp4_config,
@@ -94,7 +94,7 @@ def test_qwen3_235b_64gpu_blackwell_topology_is_valid(
         qwen3_235b_a22b_pretrain_256gpu_b200_nvfp4_config,
         qwen3_235b_a22b_pretrain_64gpu_b300_nvfp4_config,
         qwen3_235b_a22b_pretrain_256gpu_b300_nvfp4_config,
-        qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config,
+        qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config,
     ],
 )
 def test_nvfp4_disables_tp_comm_overlap(nvfp4_config_fn):
diff --git a/tests/unit_tests/scripts/performance/test_argument_parser.py b/tests/unit_tests/scripts/performance/test_argument_parser.py
new file mode 100644
index 0000000000..c78a916d82
--- /dev/null
+++ b/tests/unit_tests/scripts/performance/test_argument_parser.py
@@ -0,0 +1,89 @@
+# Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for the performance launcher's GPU argument."""
+
+import importlib.util
+import logging
+import sys
+import types
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+
+pytestmark = pytest.mark.unit
+
+_ARGUMENT_PARSER = Path(__file__).resolve().parents[4] / "scripts" / "performance" / "argument_parser.py"
+_REQUIRED_ARGS = ["--model_family_name", "llama", "--model_recipe_name", "llama3_8b", "--num_gpus", "8"]
+
+
+def _load_argument_parser():
+    """Load argument_parser.py with a stub nemo_run, which it uses only for a help-text default."""
+    nemo_run = types.ModuleType("nemo_run")
+    nemo_run_config = types.ModuleType("nemo_run.config")
+    nemo_run_config.get_nemorun_home = lambda: "/nemo_run"
+    nemo_run.config = nemo_run_config
+    spec = importlib.util.spec_from_file_location("test_performance_argument_parser", _ARGUMENT_PARSER)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    with patch.dict(sys.modules, {"nemo_run": nemo_run, "nemo_run.config": nemo_run_config}):
+        spec.loader.exec_module(module)
+    return module
+
+
+def test_deprecated_gpu_name_maps_to_vrnvl72_with_warning(caplog, monkeypatch):
+    argument_parser = _load_argument_parser()
+    monkeypatch.delenv("RANK", raising=False)
+    monkeypatch.delenv("SLURM_PROCID", raising=False)
+
+    with caplog.at_level(logging.WARNING):
+        args = argument_parser.parse_cli_args().parse_args([*_REQUIRED_ARGS, "--gpu", "vr200"])
+
+    assert args.gpu == "vrnvl72"
+    assert "--gpu vr200 is deprecated; use --gpu vrnvl72." in caplog.text
+
+
+@pytest.mark.parametrize("rank_variable", ["RANK", "SLURM_PROCID"])
+def test_deprecated_gpu_name_maps_without_warning_on_nonzero_ranks(caplog, monkeypatch, rank_variable):
+    argument_parser = _load_argument_parser()
+    monkeypatch.delenv("RANK", raising=False)
+    monkeypatch.delenv("SLURM_PROCID", raising=False)
+    monkeypatch.setenv(rank_variable, "3")
+
+    with caplog.at_level(logging.WARNING):
+        args = argument_parser.parse_cli_args().parse_args([*_REQUIRED_ARGS, "--gpu", "vr200"])
+
+    assert args.gpu == "vrnvl72"
+    assert "deprecated" not in caplog.text
+
+
+def test_current_gpu_name_is_unchanged_without_warning(caplog):
+    argument_parser = _load_argument_parser()
+
+    with caplog.at_level(logging.WARNING):
+        args = argument_parser.parse_cli_args().parse_args([*_REQUIRED_ARGS, "--gpu", "vrnvl72"])
+
+    assert args.gpu == "vrnvl72"
+    assert "deprecated" not in caplog.text
+    assert argument_parser.NUM_GPUS_PER_NODE_MAP[args.gpu] == 4
+
+
+def test_only_current_gpu_names_are_choices():
+    argument_parser = _load_argument_parser()
+
+    assert "vr200" not in argument_parser.NUM_GPUS_PER_NODE_MAP
+    with pytest.raises(SystemExit):
+        argument_parser.parse_cli_args().parse_args([*_REQUIRED_ARGS, "--gpu", "vr2000"])
diff --git a/tests/unit_tests/scripts/performance/test_executors.py b/tests/unit_tests/scripts/performance/test_executors.py
index da1af7342b..036b4d470e 100644
--- a/tests/unit_tests/scripts/performance/test_executors.py
+++ b/tests/unit_tests/scripts/performance/test_executors.py
@@ -199,9 +199,9 @@ def test_recipe_env_vars_are_exported_and_forced_into_container(tmp_path):
 
 
 @pytest.mark.skipif(not HAS_NEMO_RUN, reason="nemo_run not installed")
-def test_vr200_slurm_executor_uses_two_gpus_per_numa_node(tmp_path):
+def test_vrnvl72_slurm_executor_uses_two_gpus_per_numa_node(tmp_path):
     executor = slurm_executor(
-        gpu="vr200",
+        gpu="vrnvl72",
         account="test",
         partition="test",
         log_dir=str(tmp_path),
diff --git a/tests/unit_tests/scripts/performance/test_qwen35_vl_workload_base_configs.py b/tests/unit_tests/scripts/performance/test_qwen35_vl_workload_base_configs.py
index 63e80b51c0..ab0d37c5ee 100644
--- a/tests/unit_tests/scripts/performance/test_qwen35_vl_workload_base_configs.py
+++ b/tests/unit_tests/scripts/performance/test_qwen35_vl_workload_base_configs.py
@@ -46,13 +46,13 @@
     qwen35_vl_122b_a10b_pretrain_128gpu_h100_bf16_config,
     qwen35_vl_122b_a10b_pretrain_128gpu_h100_fp8cs_config,
 )
-from megatron.bridge.perf_recipes.qwen_vl.vr200.qwen35_vl import (
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_bf16_config,
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8cs_config,
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8mx_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_bf16_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8cs_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8mx_config,
+from megatron.bridge.perf_recipes.qwen_vl.vrnvl72.qwen35_vl import (
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_bf16_config,
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8cs_config,
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8mx_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_bf16_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8cs_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8mx_config,
 )
 from megatron.bridge.utils.cuda_graph import cuda_graph_module_names
 from tests.unit_tests.recipes.recipe_test_utils import (
@@ -364,15 +364,15 @@ def test_qwen35_vl_122b_gb300_keeps_cuda_graphs_disabled(monkeypatch: pytest.Mon
         qwen35_vl_397b_a17b_pretrain_64gpu_gb300_bf16_config,
         qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8cs_config,
         qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8mx_config,
-        qwen35_vl_397b_a17b_pretrain_64gpu_vr200_bf16_config,
-        qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8cs_config,
-        qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8mx_config,
+        qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_bf16_config,
+        qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8cs_config,
+        qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8mx_config,
     ],
 )
 def test_qwen35_vl_397b_hybridep_domain_matches_expert_parallel_size(
     recipe_fn: Callable, monkeypatch: pytest.MonkeyPatch
 ) -> None:
-    """The VR200 aliases inherit EP from GB300 but restate the HybridEP env inline.
+    """The VRNVL72 aliases inherit EP from GB300 but restate the HybridEP env inline.
 
     HybridEP asserts that the EP group size is divisible by
     ``NUM_OF_HYBRID_EP_RANKS_PER_NVLINK_DOMAIN``, and the launcher only re-derives
@@ -395,14 +395,14 @@ def test_qwen35_vl_397b_hybridep_domain_matches_expert_parallel_size(
     qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8cs_config,
     qwen35_vl_397b_a17b_pretrain_64gpu_gb300_fp8mx_config,
 )
-# VR200 aliases built from those recipes; they keep FLA until cuDNN GDN is measured on VR200.
-_VR200_CUDNN_GDN_ALIASES = (
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_bf16_config,
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8cs_config,
-    qwen35_vl_35b_a3b_pretrain_8gpu_vr200_fp8mx_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_bf16_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8cs_config,
-    qwen35_vl_397b_a17b_pretrain_64gpu_vr200_fp8mx_config,
+# VRNVL72 aliases built from those recipes; they keep FLA until cuDNN GDN is measured on VRNVL72.
+_VRNVL72_CUDNN_GDN_ALIASES = (
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_bf16_config,
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8cs_config,
+    qwen35_vl_35b_a3b_pretrain_8gpu_vrnvl72_fp8mx_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_bf16_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8cs_config,
+    qwen35_vl_397b_a17b_pretrain_64gpu_vrnvl72_fp8mx_config,
 )
 _QWEN35_VL_PERF_RECIPES = tuple(
     factory
@@ -439,7 +439,7 @@ def test_qwen35_vl_cudnn_gdn_is_limited_to_gb300_35b_397b(
 ) -> None:
     """Only the measured GB300 35B/397B recipes leave FLA.
 
-    122B, B200, B300, GB200 and VR200 were not measured (the VR200 aliases inherit the GB300 model config and reset
+    122B, B200, B300, GB200 and VRNVL72 were not measured (the VRNVL72 aliases inherit the GB300 model config and reset
     to FLA), and H100 is SM90, outside cuDNN's fast GDN engine.
     """
     patch_recipe_construction_dependencies(monkeypatch)
@@ -450,7 +450,7 @@ def test_qwen35_vl_cudnn_gdn_is_limited_to_gb300_35b_397b(
     assert config.model.gdn_kernel_backend == expected
 
 
-@pytest.mark.parametrize("recipe_fn", _GB300_CUDNN_GDN_RECIPES + _VR200_CUDNN_GDN_ALIASES, ids=recipe_factory_id)
+@pytest.mark.parametrize("recipe_fn", _GB300_CUDNN_GDN_RECIPES + _VRNVL72_CUDNN_GDN_ALIASES, ids=recipe_factory_id)
 def test_qwen35_vl_gdn_kernel_backend_not_invented_on_older_core(
     recipe_fn: Callable, monkeypatch: pytest.MonkeyPatch
 ) -> None:
@@ -522,7 +522,7 @@ def test_megatron_core_gdn_kernel_backend_contract() -> None:
 def test_qwen35_vl_perf_recipe_discovery_covers_the_cudnn_gdn_recipes() -> None:
     """The recipe matrix above must keep finding every recipe the cuDNN GDN tests reason about."""
     discovered = set(_QWEN35_VL_PERF_RECIPES)
-    assert set(_GB300_CUDNN_GDN_RECIPES + _VR200_CUDNN_GDN_ALIASES) <= discovered
+    assert set(_GB300_CUDNN_GDN_RECIPES + _VRNVL72_CUDNN_GDN_ALIASES) <= discovered
     platforms = {factory.__module__.split(".")[-2] for factory in discovered}
-    assert {"b200", "b300", "gb200", "gb300", "h100", "vr200"} <= platforms
+    assert {"b200", "b300", "gb200", "gb300", "h100", "vrnvl72"} <= platforms
     assert any("122b" in factory.__name__ for factory in discovered)
diff --git a/tests/unit_tests/scripts/performance/test_recipe_environment.py b/tests/unit_tests/scripts/performance/test_recipe_environment.py
index 2e757fd68a..87b0610f03 100644
--- a/tests/unit_tests/scripts/performance/test_recipe_environment.py
+++ b/tests/unit_tests/scripts/performance/test_recipe_environment.py
@@ -74,7 +74,7 @@ def _install_lightweight_environment_utils() -> None:
 
 @pytest.mark.parametrize(
     ("gpu", "ep_size", "expected_ranks", "expected_domain", "expected_mnnvl"),
-    [("h100", 32, 8, 8, 0), ("gb200", 32, 32, 72, 1), ("vr200", 64, 64, 72, 1)],
+    [("h100", 32, 8, 8, 0), ("gb200", 32, 32, 72, 1), ("vrnvl72", 64, 64, 72, 1)],
 )
 def test_target_topology_adapts_only_hybridep(gpu, ep_size, expected_ranks, expected_domain, expected_mnnvl):
     config = SimpleNamespace(
@@ -278,7 +278,7 @@ def test_flat_environment_preparation_defaults_missing_recipe_environment(model_
         comm_overlap=None,
     )
     args = SimpleNamespace(
-        gpu="vr200",
+        gpu="vrnvl72",
         moe_a2a_overlap=None,
         tensor_model_parallel_size=None,
         pipeline_model_parallel_size=None,
@@ -371,7 +371,7 @@ def test_exact_gpu_recipe_takes_precedence_over_canonical_fallback(monkeypatch):
 
 
 @pytest.mark.unit
-@pytest.mark.parametrize("gpu", ["gb200", "gb300", "vr200"])
+@pytest.mark.parametrize("gpu", ["gb200", "gb300", "vrnvl72"])
 @pytest.mark.parametrize("precision", ["fp8_mx", "nvfp4"])
 @pytest.mark.parametrize("override", [None, 4, 8])
 def test_ultra_weak_scaling_resolves_hsdp_before_user_overrides(monkeypatch, gpu, precision, override):
@@ -599,7 +599,7 @@ def test_flat_hydra_ep_override_updates_hybridep_topology_environment(monkeypatc
         comm_overlap=None,
     )
     args = SimpleNamespace(
-        gpu="vr200",
+        gpu="vrnvl72",
         moe_a2a_overlap=None,
         tensor_model_parallel_size=None,
         pipeline_model_parallel_size=None,
@@ -918,7 +918,7 @@ def test_flat_non_ncclep_backends_keep_hybridep_environment(dispatcher_backend):
     assert recipe.env_vars == original_env
 
 
-def test_vr200_argparse_overrides_keep_sm100_cuda_connection_count():
+def test_vrnvl72_argparse_overrides_keep_sm100_cuda_connection_count():
     from utils.overrides import _apply_flat_cli_environment_compatibility
 
     recipe = SimpleNamespace(
@@ -931,7 +931,7 @@ def test_vr200_argparse_overrides_keep_sm100_cuda_connection_count():
         ),
     )
     args = SimpleNamespace(
-        gpu="vr200",
+        gpu="vrnvl72",
         moe_a2a_overlap=None,
         tensor_model_parallel_size=2,
         pipeline_model_parallel_size=None,
diff --git a/tests/unit_tests/scripts/performance/test_setup_experiment.py b/tests/unit_tests/scripts/performance/test_setup_experiment.py
index 8cae2d1c57..56f3cebd6c 100644
--- a/tests/unit_tests/scripts/performance/test_setup_experiment.py
+++ b/tests/unit_tests/scripts/performance/test_setup_experiment.py
@@ -42,10 +42,10 @@ def test_default_experiment_name_uses_only_cli_metadata() -> None:
             task="pretrain",
             compute_dtype="fp8_mx",
             num_gpus=256,
-            gpu="vr200",
+            gpu="vrnvl72",
             config_variant="large_scale",
         )
-        == "pretrain_deepseek_v3_fp8_mx_gpus256_vr200_large_scale"
+        == "pretrain_deepseek_v3_fp8_mx_gpus256_vrnvl72_large_scale"
     )
     assert (
         setup_experiment._default_experiment_name(
@@ -88,8 +88,18 @@ def test_recipe_arguments_are_forwarded_unchanged() -> None:
     ]
 
 
+@pytest.mark.parametrize(
+    ("gpu", "gpus_per_node", "expected_gpu"),
+    [
+        ("h100", "8", "h100"),
+        # The deprecated name must reach main() as vrnvl72, which applies the 4752 MHz memory clock default.
+        ("vr200", "4", "vrnvl72"),
+    ],
+)
 @pytest.mark.parametrize("is_long_convergence_run", [False, True])
-def test_submission_dry_run_does_not_import_bridge_or_mcore(tmp_path: Path, is_long_convergence_run: bool) -> None:
+def test_submission_dry_run_does_not_import_bridge_or_mcore(
+    tmp_path: Path, is_long_convergence_run: bool, gpu: str, gpus_per_node: str, expected_gpu: str
+) -> None:
     blocker_dir = tmp_path / "login_node"
     blocker_dir.mkdir()
     (blocker_dir / "sitecustomize.py").write_text(
@@ -112,6 +122,9 @@ def find_spec(self, fullname, path=None, target=None):
 """
     )
     environment = os.environ.copy()
+    # Non-zero ranks skip the deprecated --gpu warning that this test expects from the launcher.
+    environment.pop("RANK", None)
+    environment.pop("SLURM_PROCID", None)
     environment["NEMORUN_HOME"] = str(tmp_path / "nemorun")
     environment["PYTHONPATH"] = os.pathsep.join(
         value for value in (str(blocker_dir), environment.get("PYTHONPATH")) if value
@@ -126,9 +139,9 @@ def find_spec(self, fullname, path=None, target=None):
         "--num_gpus",
         "8",
         "--gpus_per_node",
-        "8",
+        gpus_per_node,
         "--gpu",
-        "h100",
+        gpu,
         "--account",
         "test-account",
         "--partition",
@@ -155,7 +168,10 @@ def find_spec(self, fullname, path=None, target=None):
     output = result.stdout + result.stderr
     assert result.returncode == 0, output
     assert "bootstrap.py" in output
-    assert "pretrain_llama3_8b_bf16_gpus8_h100" in output
+    assert f"pretrain_llama3_8b_bf16_gpus8_{expected_gpu}" in output
+    assert ("--gpu vr200 is deprecated; use --gpu vrnvl72." in output) == (gpu == "vr200")
+    # The dry-run sbatch dump wraps at spaces, so match the clock value as one token.
+    assert ("4752,4752" in output) == (expected_gpu == "vrnvl72")
     assert "WANDB_RUN_ID" in output
     assert "WANDB_RESUME" in output
     assert "login-node import forbidden" not in output
diff --git a/tests/unit_tests/scripts/training/test_recipe_metadata.py b/tests/unit_tests/scripts/training/test_recipe_metadata.py
index c7c3cd74f7..018c41a5cb 100644
--- a/tests/unit_tests/scripts/training/test_recipe_metadata.py
+++ b/tests/unit_tests/scripts/training/test_recipe_metadata.py
@@ -18,6 +18,7 @@
 
 import ast
 import importlib.util
+import logging
 import re
 import sys
 from pathlib import Path
@@ -98,6 +99,61 @@ def test_metadata_does_not_impose_a_hardware_node_shape():
     assert metadata.hardware == "h100"
 
 
+def test_deprecated_hardware_name_resolves_to_renamed_recipe(caplog, monkeypatch):
+    module = _load_module()
+    legacy_name = "qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config"
+    monkeypatch.delenv("RANK", raising=False)
+    monkeypatch.delenv("SLURM_PROCID", raising=False)
+
+    with caplog.at_level(logging.WARNING):
+        metadata = module.selected_benchmark_recipe(["--recipe", legacy_name])
+
+    assert metadata is not None
+    assert metadata.recipe_name == "qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config"
+    assert metadata.hardware == "vrnvl72"
+    assert f"Recipe '{legacy_name}' uses a deprecated hardware name" in caplog.text
+    assert module.available_benchmark_recipe_metadata(legacy_name) is None
+
+
+@pytest.mark.parametrize("rank_variable", ["RANK", "SLURM_PROCID"])
+def test_deprecated_hardware_name_resolves_without_warning_on_nonzero_ranks(caplog, monkeypatch, rank_variable):
+    module = _load_module()
+    monkeypatch.delenv("RANK", raising=False)
+    monkeypatch.delenv("SLURM_PROCID", raising=False)
+    monkeypatch.setenv(rank_variable, "3")
+
+    with caplog.at_level(logging.WARNING):
+        metadata = module.resolved_benchmark_recipe_metadata("qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config")
+
+    assert metadata is not None
+    assert metadata.recipe_name == "qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config"
+    assert "deprecated" not in caplog.text
+
+
+def test_deprecated_hardware_name_without_renamed_recipe_is_not_rewritten(caplog, monkeypatch):
+    module = _load_module()
+    monkeypatch.delenv("RANK", raising=False)
+    monkeypatch.delenv("SLURM_PROCID", raising=False)
+
+    with caplog.at_level(logging.WARNING):
+        # qwen3_30b_a3b has no VRNVL72 fp8cs benchmark to redirect to.
+        metadata = module.resolved_benchmark_recipe_metadata("qwen3_30b_a3b_pretrain_8gpu_vr200_fp8cs_config")
+
+    assert metadata is None
+    assert "deprecated" not in caplog.text
+
+
+def test_current_hardware_name_resolves_without_deprecation_warning(caplog):
+    module = _load_module()
+
+    with caplog.at_level(logging.WARNING):
+        metadata = module.resolved_benchmark_recipe_metadata("qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config")
+
+    assert metadata is not None
+    assert metadata.hardware == "vrnvl72"
+    assert "deprecated" not in caplog.text
+
+
 @pytest.mark.parametrize(
     ("recipe_name", "task", "step_name"),
     [
diff --git a/tests/unit_tests/scripts/training/test_recipe_runner.py b/tests/unit_tests/scripts/training/test_recipe_runner.py
index ae70650fdf..6204aa74d4 100644
--- a/tests/unit_tests/scripts/training/test_recipe_runner.py
+++ b/tests/unit_tests/scripts/training/test_recipe_runner.py
@@ -94,7 +94,9 @@ def benchmark_recipe() -> object:
 
     library_finder = Mock()
     benchmark_finder = Mock(return_value=benchmark_recipe)
-    monkeypatch.setattr(recipe_runner, "resolved_benchmark_recipe_metadata", lambda _name: object())
+    monkeypatch.setattr(
+        recipe_runner, "resolved_benchmark_recipe_metadata", lambda name: SimpleNamespace(recipe_name=name)
+    )
     monkeypatch.setattr(recipe_runner, "find_library_recipe", library_finder)
     monkeypatch.setattr(recipe_runner, "find_benchmark_recipe", benchmark_finder)
 
@@ -103,6 +105,20 @@ def benchmark_recipe() -> object:
     library_finder.assert_not_called()
 
 
+def test_load_recipe_loads_renamed_benchmark_for_deprecated_hardware_name(
+    recipe_runner: ModuleType, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    config = object()
+    benchmark_finder = Mock(return_value=lambda: config)
+    library_finder = Mock()
+    monkeypatch.setattr(recipe_runner, "find_benchmark_recipe", benchmark_finder)
+    monkeypatch.setattr(recipe_runner, "find_library_recipe", library_finder)
+
+    assert recipe_runner.load_recipe("qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config") is config
+    benchmark_finder.assert_called_once_with("qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config")
+    library_finder.assert_not_called()
+
+
 def test_load_recipe_reports_both_packages_when_missing(
     recipe_runner: ModuleType, monkeypatch: pytest.MonkeyPatch
 ) -> None:
@@ -119,7 +135,9 @@ def test_duplicate_name_warns_and_selects_benchmark(
     recipe_name = "qwen3_235b_a22b_pretrain_256gpu_h100_bf16_config"
     config = object()
 
-    monkeypatch.setattr(recipe_runner, "resolved_benchmark_recipe_metadata", lambda _name: object())
+    monkeypatch.setattr(
+        recipe_runner, "resolved_benchmark_recipe_metadata", lambda name: SimpleNamespace(recipe_name=name)
+    )
     benchmark_finder = Mock(return_value=lambda: config)
     library_finder = Mock()
     monkeypatch.setattr(recipe_runner, "find_benchmark_recipe", benchmark_finder)
@@ -146,7 +164,9 @@ def test_finetuning_collision_selects_benchmark(
 ) -> None:
     config = object()
 
-    monkeypatch.setattr(recipe_runner, "resolved_benchmark_recipe_metadata", lambda _name: object())
+    monkeypatch.setattr(
+        recipe_runner, "resolved_benchmark_recipe_metadata", lambda name: SimpleNamespace(recipe_name=name)
+    )
     library_finder = Mock()
     benchmark_finder = Mock(return_value=lambda: config)
     monkeypatch.setattr(recipe_runner, "find_library_recipe", library_finder)
diff --git a/tests/unit_tests/scripts/training/test_run_recipe.py b/tests/unit_tests/scripts/training/test_run_recipe.py
index b3429646d9..b408a41ecd 100644
--- a/tests/unit_tests/scripts/training/test_run_recipe.py
+++ b/tests/unit_tests/scripts/training/test_run_recipe.py
@@ -616,7 +616,7 @@ def apply_overrides(recipe, overrides):
     module.main(
         [
             "--recipe",
-            "qwen3_235b_a22b_pretrain_256gpu_vr200_nvfp4_config",
+            "qwen3_235b_a22b_pretrain_256gpu_vrnvl72_nvfp4_config",
             "--mode",
             "pretrain",
             "--dry-run",
diff --git a/tests/unit_tests/scripts/training/test_setup_experiment.py b/tests/unit_tests/scripts/training/test_setup_experiment.py
index a7a7ef05e3..9d82a0c844 100644
--- a/tests/unit_tests/scripts/training/test_setup_experiment.py
+++ b/tests/unit_tests/scripts/training/test_setup_experiment.py
@@ -263,10 +263,10 @@ def test_parser_consumes_peak_mem_clk(option):
 @pytest.mark.parametrize(
     ("requested_peak_mem_clk", "hardware", "expected"),
     [
-        (None, "vr200", 4752),
+        (None, "vrnvl72", 4752),
         (None, "h100", None),
-        (2600, "vr200", 2600),
-        (-1, "vr200", None),
+        (2600, "vrnvl72", 2600),
+        (-1, "vrnvl72", None),
     ],
 )
 def test_peak_mem_clk_resolution(requested_peak_mem_clk, hardware, expected):
@@ -299,7 +299,15 @@ def test_peak_mem_clk_is_applied_once_per_slurm_node():
     assert "sudo nvidia-smi -lmc 4752,4752" in executor.setup_lines
 
 
-def test_main_applies_vr200_peak_mem_clk_default(monkeypatch):
+@pytest.mark.parametrize(
+    "recipe_name",
+    [
+        "qwen3_30b_a3b_pretrain_8gpu_vrnvl72_bf16_config",
+        # The deprecated hardware name resolves to the same benchmark recipe.
+        "qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config",
+    ],
+)
+def test_main_applies_vrnvl72_peak_mem_clk_default(monkeypatch, recipe_name):
     module = _load_setup_experiment_module()
     configured_clocks = []
     sentinel_executor = object()
@@ -348,7 +356,7 @@ def dryrun(self):
             "image.sqsh",
             "--submission-dry-run",
             "--recipe",
-            "qwen3_30b_a3b_pretrain_8gpu_vr200_bf16_config",
+            recipe_name,
         ]
     )