Skip to content
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.

Sign up for GitHub

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

correct perf numbers and add details #414

Merged
merged 1 commit into from
Dec 27, 2024
Merged

correct perf numbers and add details #414

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 4 additions & 1 deletion docs/performance/hunyuanvideo.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,9 @@

xDiT is [HunyuanVideo](https://github.com/Tencent/HunyuanVideo?tab=readme-ov-file#-parallel-inference-on-multiple-gpus-by-xdit)'s official parallel inference engine. On H100 and H20 GPUs, xDiT reduces the generation time of 1028x720 videos from 31 minutes to 5 minutes, and 960x960 videos from 28 minutes to 6 minutes.

The H100 and H20 performance benchmarks are done with the official HunyuanVideo repository. The L20 performance benchmarks are done with the `diffusers` implementation.
The L20 performance benchmarks are measured using this [script](examples/hunyuan_video_usp_example.py), along with `flash-attn==2.7.2.post1` and CUDA 12.4.

### 1280x720 Resolution (129 frames, 50 steps) - Ulysses Latency (seconds)

<center>
Expand All @@ -22,6 +25,6 @@ xDiT is [HunyuanVideo](https://github.com/Tencent/HunyuanVideo?tab=readme-ov-fil
|----------|--------|---------|---------|---------|
| H100 | 1,735.01 | 934.09 | 645.45 | 367.02 |
| H20 | 6,621.46 | 3,400.55 | 2,310.48 | 1,214.67 |
| L20 | 6,039.08 | 3,260.62 | 2,070.96 | |
| L20 | 6,039.08 | 3,260.62 | 2,284.74 | |

</center>
Loading