Skip to content

Run Example 7 (GEMM All-Scatter) and report benchmark results - #29

Draft
JoseSantosAMD with Copilot wants to merge 1 commit into
JoseSantosAMD/intellikitfrom
copilot/sub-pr-24-one-more-time
Draft

Run Example 7 (GEMM All-Scatter) and report benchmark results#29
JoseSantosAMD with Copilot wants to merge 1 commit into
JoseSantosAMD/intellikitfrom
copilot/sub-pr-24-one-more-time

Conversation

Copilot AI commented Mar 4, 2026

Copy link
Copy Markdown

Ran Example 7 (07_gemm_all_scatter) on 2 AMD Radeon GPU ranks to validate correctness and measure performance of the fused persistent GEMM + all-scatter kernel.

Results

Default dimensions — M=8192, N=4608, K=36864, fp16, 2 ranks (304 CUs each):

  • Validation: ✅ PASSED
  • Latency: 4.519 ms
  • Throughput: 615.94 TFLOPs
  • GEMM kernel avg: 4.408 ms (126 experiments)

Smoke test — M=4096, N=4608, K=4096, fp16:

  • Validation: ✅ PASSED
  • Latency: 0.685 ms
  • Throughput: 225.87 TFLOPs

The kernel fuses matrix multiply with cross-GPU scatter in a single persistent kernel — each rank computes its N-shard and scatters results to peers' global buffers in-flight, with no separate communication phase.


✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

Copilot AI mentioned this pull request Mar 4, 2026
Copilot AI changed the title [WIP] Add intellikit to project Run Example 7 (GEMM All-Scatter) and report benchmark results Mar 4, 2026
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