Repository navigation
[PowerX] rate B300 HGX GPUs at 1,100 W in the power model port / [PowerX] 功耗模型移植版将 B300 HGX GPU 额定功耗改为 1,100 W - #1736
Merged
Merged
Conversation
中文:功耗模型移植版将 B300 HGX 的 GPU 额定功耗改为 1,100 W,与上游 InferenceX power_model 保持一致。
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The power model port pins InferenceX
power_modelat059bdc00e, whereB300HGXBoardaccepts only a 1,200 W GPU rating. The GPU registry (packages/constants/src/gpu-keys.ts) rates B300 at 1,100 W. So a B300 chart draws its Rated TDP line at 1,100 W while its modeled IT and PUE lines size fans for 1,200 W. For one B300 chassis at 100 W per GPU, idle, modeled IT power is 2,034.83 W at the old rating and 2,035.94 W at 1,100 W.This ports SemiAnalysisAI/InferenceX#4960, which moves the upstream B300 rating to 1,100 W, the air-cooled HGX B300 figure.
Changes:
ubb/b300_hgx.ts:gpu_tdp_wdefaults to and accepts only 1100, matching upstreamb300_hgx.py.hgx_fans.ts:HGX_FAN_BASELINElists B300 at 1100 W and moves to version 2, matching upstream.scripts/generate-power-model-parity.py:REVISIONpins the upstream commit that carries the change.power-model.manifest.jsonandpower-model.parity.json: regenerated by that script, not edited by hand.Testing:
vitest runoversrc/lib/power,src/components/inference,src/lib/charts,src/lib/views-api,src/lib/api,src/components/gpu-powerandsrc/lib/powerx: 150 files, 4,962 tests passed, includingpower-model-parity.test.tsagainst the regenerated fixture.Blocker:
REVISIONpoints at the head of the unmerged upstream branch (95a63a2). After SemiAnalysisAI/InferenceX#4960 merges, re-pinREVISIONto its merge commit and regenerate both files before merging this PR.Related: Companion of SemiAnalysisAI/InferenceX#4960.
PR checklist
inferencex-skillschanges/zhsiblingdocs/chinese-copy.mdNote
Low Risk
Localized power-model parity and B300 fan-sizing correction; no auth, API, or data-path changes beyond modeled watt estimates.
Overview
Aligns the TypeScript power model port with upstream InferenceX by rating B300 HGX at 1,100 W (air-cooled HGX spec) instead of 1,200 W, matching the app GPU registry and fixing the mismatch where charts drew Rated TDP at 1,100 W but modeled fans/IT power as if GPUs were 1,200 W.
B300HGXBoardnow defaults to and accepts onlygpu_tdp_w1,100, andHGX_FAN_BASELINEmoves to version 2 with B300 listed at 1,100 W so normalized fan sizing uses the same design heat. The parity generator’sREVISIONis re-pinned to InferenceX commit95a63a2(carrying upstream PR 4960);power-model.manifest.jsonandpower-model.parity.jsonwere regenerated from that revision—no UI or API surface changes.Reviewed by Cursor Bugbot for commit d7b4b3d. Bugbot is set up for automated code reviews on this repo. Configure here.