Skip to content

add tuned GEMM config for Qwen3.5 bf16 on MI308X#3077

Open
zovonoir wants to merge 3 commits intoROCm:mainfrom
zovonoir:perf/qwen35-bf16-tuned-gemm
Open

add tuned GEMM config for Qwen3.5 bf16 on MI308X#3077
zovonoir wants to merge 3 commits intoROCm:mainfrom
zovonoir:perf/qwen35-bf16-tuned-gemm

Conversation

@zovonoir
Copy link
Copy Markdown
Contributor

@zovonoir zovonoir commented May 8, 2026

Summary

  • Add hipblaslt tuned GEMM configuration for Qwen3.5-27B bf16 inference on MI308X (80 CUs, TP2)
  • Covers common M/N/K shapes for dense linear layers

Benchmark

MI308X, Qwen3.5-27B, ATOM SGLang plugin mode, TP2, ISL=60381, OSL=132:

Version TPOT (ms) TTFT (ms) Tput/GPU (tok/s)
ATOM baseline 18.9 17562 1510
+ layernorm + qwen3_next optimizations (ROCm/ATOM#708, #3051) 15.6 15300 1605
+ tuned GEMM (this patch) 14.8 12613 2078
  • TTFT reduced by 18% (15300 → 12613 ms)
  • Throughput improved by 29% (1605 → 2078 tok/s/GPU)

Files Changed

  • aiter/configs/model_configs/qwen35_bf16_tuned_gemm.csv — tuning config

Test plan

  • End-to-end benchmark on MI308X with Qwen3.5-27B
  • CI

🤖 Generated with Claude Code

hipblaslt tuned GEMM configuration for Qwen3.5-27B bf16 inference
on MI308X (80 CUs, TP2). Covers common M/N/K shapes for dense
linear layers.

Benchmark (MI308X, Qwen3.5-27B, TP2, ISL=60381, OSL=132):

| Version                         | TPOT(ms) | TTFT(ms) | Tput/GPU |
|---------------------------------|----------|----------|----------|
| ATOM baseline                   | 18.7     | 15266    | 1574     |
| + layernorm + qwen3_next optim  | 15.6     | 15300    | 1605     |
| + tuned GEMM (this patch)       | 15.0     | 11405    | 2088     |

TTFT reduced by 25% (15300 → 11405 ms), throughput +30% (1605 → 2088).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@zovonoir zovonoir requested review from a team and Copilot May 8, 2026 01:26
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review any files in this pull request.


💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented May 8, 2026

🏷️ CI Guide

Runs automatically on every PR:

  • ✅ Pre-checks (submodule verification, code formatting)
  • ✅ Aiter op tests (gfx942 + gfx950)
  • ✅ Triton tests on MI35X (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label Tests
ci:triton-300x Run an additional Triton test job on MI300X in PRs; main branch always runs both MI35X and MI300X
ci:sglang SGLang integration tests
ci:atom ATOM benchmark (DeepSeek-R1 + GPT-OSS)
ci:vllm vLLM benchmark
ci:all All of the above

Add labels via the sidebar or gh pr edit 3077 --add-label <label>

@zovonoir
Copy link
Copy Markdown
Contributor Author

zovonoir commented May 8, 2026

I will choose other libtypes and try again.

zovonoir and others added 2 commits May 8, 2026 19:32
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants