Skip to content

feat: add Moore Threads multi-GPU Suite E support - #62

Draft
Keke946 wants to merge 3 commits into
FreedomIntelligence:mainfrom
Keke946:feat/moorethreads-suite-e-tp
Draft

feat: add Moore Threads multi-GPU Suite E support#62
Keke946 wants to merge 3 commits into
FreedomIntelligence:mainfrom
Keke946:feat/moorethreads-suite-e-tp

Conversation

@Keke946

@Keke946 Keke946 commented Sep 8, 2026

Copy link
Copy Markdown

Summary

  • Add and validate multi-chip Suite E support for the Moore Threads vLLM-MUSA runner.
  • Submit MTT S4000 scaling results across 1, 2, 4, and 8 GPUs using tensor parallelism.
  • Use Meta-Llama-3-8B-Instruct with FP16 on vLLM-MUSA 0.4.2.

Results

GPUs Best output throughput Speedup vs. 1 GPU Scaling efficiency
1 409.71 tok/s 1.00x 100.0%
2 498.36 tok/s 1.22x 60.8%
4 570.06 tok/s 1.39x 34.8%
8 530.78 tok/s 1.30x 16.2%

The 4-GPU configuration gives the highest throughput. At 8 GPUs, communication and synchronization overhead outweigh the additional compute, reducing throughput below the 4-GPU result.

Validation

  • Suite E completed at all required chip counts: 1x, 2x, 4x, and 8x.
  • Three measured runs completed at client concurrency 8, 32, and 128 for every chip count.
  • Accuracy result was produced and marked valid by the benchmark (subset_score: 0.07; this old MUSA stack has a known batching-quality limitation).
  • python runners/validate_submission.py --dir results/community/mtt_s4000x8_suite_E_moorethreads_vllm_musa_e0266927_ca7cadca passes.
  • Validator warning: power telemetry is unavailable, so tokens/sec/watt is not reported.

Environment

  • Hardware: 8 × MTT S4000 48 GB
  • Runner: moorethreads_vllm_musa_e0266927
  • Framework: vLLM-MUSA 0.4.2
  • Run ID: ca7cadca

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.

1 participant