NVIDIA / Fuser

A Fusion Code Generator for NVIDIA GPUs (commonly known as "nvFuser")
Other
271 stars 53 forks source link

[Do not merge] Overlap AG+GEMM benchmark #3367

Open samnordmann opened 2 weeks ago

samnordmann commented 2 weeks ago

I open this pr to potentially reuse the branch for other benchmarks and to gather experimental results. Whether we should merge it is open to discussion.

The experiment measures the performance of AG+Matmul with a collective-based pipeline to achieve overlap. Analysis: https://docs.google.com/document/d/1gLSYe6RmZXQQojgRL-ZZm3Ggm4xnzD5VSbeFPHCPkgs/edit?usp=sharing