Skip to content

Update README.md #16

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 4, 2024
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,7 @@ PYTHONPATH=./python:$PYTHONPATH python3 test/test_gemm_only.py 4096 12288 6144 -
## Performance
We measured the examples from the above demo on both A800s and H800s. Each machine has 8 GPUs, with a TP size set to 8. The table below shows the performance comparison between flux and torch+nccl. It can be observed that by overlapping fine-grained computation and communication, Flux is able to effectively hide a significant portion of the communication time

| | M | K | N | Torch Gemm | Torch NCCL | Torch Total | Flux Gemm | Flux NCCL | Flux Total |
| | M | K | N | Torch Gemm | Torch NCCL | Torch Total | Flux Gemm | Flux Comm | Flux Total |
|----------|----------|----------|----------|----------|----------|----------|----------|----------|-----------|
| AG+Gemm(A800) | 4096 | 12288 | 49152 | 2.438ms | 0.662ms | 3.099ms | 2.378ms | 0.091ms | 2.469ms |
| Gemm+RS(A800) | 4096 | 49152 | 12288 | 2.453ms | 0.646ms | 3.100ms | 2.429ms | 0.080ms | 2.508ms |
Expand Down