All pull requests

Pull requests

• Dockerfile - Add cuda13.4.dockerfile

  #858·WenqingLan1 opened Sep 30, 2026Contributor·

  ·
  15

• Dockerfile - Add ROCm7.2 dockerfile

  #855·Nguyen Vuong (normanvuong) opened Sep 23, 2026·2 tasks done·

  ·
  20

• Benchmarks: Decouple cuDNN workload and legacy execution setup

  #853·gusui-msft opened Sep 16, 2026Contributor·

  ·
  7

• Benchmarks: Fix cuDNN storage/compute types and FP16 initialization

  #852·gusui-msft opened Sep 10, 2026Contributor·

  ·
  3

• Dockerfile - Add cuda13.3.dockerfile on pytorch:26.07-py3

  #846·gusui-msft opened Aug 18, 2026Contributor·

  ·
  17

• Pin GitHub Actions to full-length commit SHAs

  #844·Dan Fiedler (danfiedler-msft) opened Aug 12, 2026·

  ·
  8

• Add ROCm 7.2 support

  #814·Zheyu Shen (arsdragonfly) opened May 5, 2026·

  ·
  10

• Add ROCm 7.0 support

  #813·Zheyu Shen (arsdragonfly) opened May 5, 2026·

  ·
  2

• Add ROCm 6.4 support

  #812·Zheyu Shen (arsdragonfly) opened May 5, 2026·

  ·
  6

• SysInfo - Fix NIC info collection in sys_info output

  #806·one (alephpiece) opened Apr 20, 2026Contributor·

  ·

• Hugging Face Model integration in Superbench

  #803·Aishwarya Tonpe (Aishwarya-Tonpe) opened Apr 13, 2026Contributor·

  ·
  182

• Runner: make MPI bind-to configurable

  #797·one (alephpiece) opened Mar 27, 2026Contributor·

  ·
  9

  

• Enhancement- Add SM80/90 architecture to CUTLASS builds for CUDA 12.8+ for A100

  enhancement
  #785·Hongtao Zhang (polarG) opened Mar 11, 2026Contributor·

  ·
  6

  

• Docs - Add documentation for --no-docker parameter requirements (#715)

  documentation
  #783·NJX (NJX-njx) opened Mar 3, 2026·

  ·
  1

  20

  

• Bug Fix - Fix NVBandwidth benchmark results parsing bug (#748)

  benchmarksbugmicro-benchmarks
  #782·NJX (NJX-njx) opened Mar 3, 2026·

  ·
  1

  10

  

• Bechmarks: Model Benchmark - Add support of megatron lm deepseek v2 lite model training for cuda gpus

  benchmarksmodel-benchmarks
  #765·Yuting Jiang (yukirora) opened Dec 2, 2025Contributor·

  ·
  1

  

• Benchmarks: Micro benchmark - collect per-snapshot per-GPU flops/temp in gpu burn

  benchmarksmicro-benchmarks
  #735·Yuting Jiang (yukirora) opened Sep 8, 2025Contributor·

  ·
  10