RVC-Project/Retrieval-based-Voice-Conversion-WebUI

Slower Multi-GPU training with 2x the number of GPUs and 4x the amount of VRAM

Open

#244 opened on May 7, 2023

View on GitHub
 (1 comment) (1 reaction) (0 assignees)Python (2,849 forks)batch import
help wantedquestion

Repository metrics

Stars
 (18,427 stars)
PR merge metrics
 (No merged PRs in 30d)

Description

I have two systems training on identical datasets

System A has 4 x NVIDIA RTX A5000 (24GB VRAM per GPU), and a batch size of 12 per GPU.

System B has 7 x NVIDIA RTX A6000 (48GB VRAM per GPU), and a batch size of 18 per GPU.

I would expect System B to train much faster. However...

  • System A (96GB total VRAM, batch size 12) takes 11 seconds per epoch.

  • System B (336GB total VRAM, batch size 18) takes 13 seconds per epoch.

I'm wondering if this is down to the overhead of multi-GPU training, or if there's something I'm missing here?

Thank you

Contributor guide

Research direction
Investigate the training loop and data loading for potential bottlenecks in multi GPU communication.
Tech stack
python
Domain
machine learning
Issue type
Performance
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
PythonPyTorchmulti GPU training
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40