rdna3

Zig INferenCe Engine — Local LLM inference on AMD GPUs and Apple Silicon

Windows-only version of ComfyUI which uses AMD's official ROCm and PyTorch libraries to get better performance with AMD GPUs. [auto-installation and popular performance enhancing packages like triton * sage-attention * flash-attention * bitsandbytes included ]

The intelligent OptiScaler installer Linux gamers needed. Automates FSR4, XeSS & DLSS configuration with GPU-optimized profiles for RDNA3/4, Arc & RTX cards.

Utilities for accessing AMD's Machine-Readable GPU ISA Specifications.

Installation script for an AI applications using ROCm on Linux.

TRELLIS (Microsoft's Image-to-3D generator) running on AMD GPUs with ROCm. Includes Gaussian splatting, mesh extraction, and GLB export. Tested on RX 7800 XT.

A turnkey, fully-local AI workstation engineered for the AMD Ryzen AI Max+ 395. LLM inference, voice, document parsing, browser automation, agents — all on-device.

A honest port of vLLM-ROCm for windows.

Reproducible RDNA reference for Meta Muse-Glimmer-30B — adapting MI-series ROCm recipes to Ryzen AI and Radeon with vLLM, llama.cpp, DFlash and auditable benchmarks.

Evaluation-backed AMD ROCm port of HunyuanOCR-1.5 with reproducible OmniDocBench v1.6 benchmarks across llama.cpp, vLLM, and Transformers on gfx1100.

Unlock fast, local LLM inference on AMD-powered mini PCs delivering 65-87 t/s for large models without cloud or subscription costs

PyTorch built from source for AMD RDNA 3.5 (gfx1150) — Radeon 890M/880M GPU acceleration

Docker infrastructure for AMD Strix Halo (RDNA 3.5 / gfx1151): PyTorch + ROCm base container and a separate Ollama LLM service. Two folders, two Compose files, one Strix Halo box.

High-performance FP32 GEMV for AMD RDNA 3 (gfx1100), with reproducible performance and numerical analysis.

FlashAttention-2 Triton kernel for AMD RDNA3 (gfx1100 / Radeon RX 7900 XT/XTX/GRE): forward, backward, and split-K decode

Working ROCm 7.2.x + PyTorch environment for RDNA3. Built after fighting pipeline rats and wondering, “Lisa Su, girl… what are they doing down there.”

FLUX.1-dev on AMD Radeon consumer GPUs — fast, low-VRAM, and shippable. Backport patches + benchmarks for torchao + diffusers group_offload on ROCm.

Multi-GPU tensor/context parallel diffusion on AMD ROCm — with the patch that makes it actually work.

🔥 Automated nightly & release builds of ROCmFPX, Ciru ROCmFPX (DualView Q7/Q8), and q38rocm with AMD ROCm 7 for Windows & Linux. Portable standalone binaries with bundled runtime libraries for Strix Halo (gfx1151), RDNA4, RDNA3, and Homebrew tap.

Local LLM benchmarks on AMD Strix Halo — 26+ models tested across RADV, AMDVLK, and ROCm with llama.cpp