modelscope / modelscope/DiffSynth-Studio

VACE 1.3B inference speed does not benefit from multiple GPUs

Open
#860 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13.1k
Forks
1.3k
Avg merge
13h 12m
Merged PRs (30d)
45

Description

Hi,

I'm trying to run VACE 1.3B model using the example script (after changing it to using the 1.3B model). However, I notice that the multi-GPU set-up does not show a speed-up compared to the single-GPU setting. Both takes around 5 minutes to generate a 480p 81-frame video over 50 timesteps. (I'm using 1xA100 vs. 8xA100).

In contrast, the inference script provided by the original VACE codebase does show significant speedup with --ring_size 8 using 8 GPUs (around 5 minutes on 1 GPU and around 1 minute GPU).

I tried to dig a bit on why this happens but I couldn't locate the reason or fix it. Can someone take a look? Thanks!

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with examples/wanvideo/acceleration/unified_sequence_parallel.py and compare its multi-GPU configuration with the original VACE model-inference setup using --ring_size 8. Reproduce the reported 1-GPU and 8-GPU timings for the VACE 1.3B model at 480p and 81 frames, then trace why the configurations differ. Done means the DiffSynth-Studio example shows a measurable multi-GPU speedup comparable to the reference setup.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
distributed-systems, machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.