Why is yolov5 training slow?

[daigang896]

Search before asking

• [X] I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

Why is yolov5 training slow? Use the yolov5m6 pretraining model. Does anyone have the same problem?

Additional

No response

[glenn-jocher]

👋 Hello! Thanks for asking about training speed issues. YOLOv5 🚀 can be trained on CPU (slowest), single-GPU, or multi-GPU (fastest). If you would like to increase your training speed some options are:

• Increase --batch-size
• Reduce --img-size
• Reduce model size, i.e. from YOLOv5x -> YOLOv5l -> YOLOv5m -> YOLOv5s
• Train with multi-GPU DDP at larger --batch-size
• Train on cached data: python train.py --cache (RAM caching) or --cache disk (disk caching)
• Train on faster GPUs, i.e.: P100 -> V100 -> A100
• Train on free GPU backends with up to 16GB of CUDA memory:

Good luck 🍀 and let us know if you have any other questions!

[daigang896]

Hello. Train on a NVIDIA RTX A6000 48G card. When the batchsize is increased, the speed of each iteration becomes slow. But the GPU memory is sufficient, but the speed can not be improved. What is the bottleneck?

[Laughing-q]

@daigang896 It seems the bottleneck is data-loading as you've increased batch-size, maybe using more workers will help you. https://github.com/ultralytics/yolov5/blob/7398d2d77cbac9f66259926d49c26bfa3c257a9b/train.py#L457

[glenn-jocher]

@daigang896 also try --cache ram or --cache disk to reduce dataloading bottlenecks.

[daigang896]

Thanks, I'll try it.

[daigang896]

@glenn-jocher @Laughing-q Hello， More --workers values did not work. The set --workers==--batchsize=16 did not find that the training speed of each iteration was faster. The CPU utilization is low. I don't know what the problem is.

[daigang896]

@glenn-jocher Try -- cache ram found insufficient memory, try -- cache disk found no improvement in training speed.

[David-19940718]

@glenn-jocher Try -- cache ram found insufficient memory, try -- cache disk found no improvement in training speed.

For the most situation, follow the instructions by the author advise will be tackled.

In your case, I think that it may be caused by your machine, you can try another machine and repeate once time if supported.

Note that, it is very important to load the data into the memory, so, don't forget to add this line -- cache ram.

[daigang896]

Hello. At present, the training speed has been significantly improved and the problem has been solved by updating the card driver, cuda, cudnn and pytorch, and using yolov5 6.2 code.

[glenn-jocher]

@daigang896 great!!

[github-actions[bot]]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!

[Robotatron]

I saw no difference in training time with --cache or without when using a SSD, interesting. Also using a smaller image size (e.g. 160 with batch size of 960) was training SLOWER then using a bigger image size (e.g. 240 with bs of 221)

[bartlomiejgadzicki-digica]

Hi there @glenn-jocher, do symlinks affect training speed? I keep multiple versions of my datasets and this way I can avoid storing the same images multiple times. Do you think it can be harmful in any way?

[github-actions[bot]]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!

[glenn-jocher]

@bartlomiejgadzicki-digica symlinks generally do not affect training speed significantly, as they are simply pointers to the original data. However, they can introduce a slight overhead during data loading, so their impact on training speed might be negligible. Maintaining multiple dataset versions through symlinking is a smart storage solution. As long as your data loading and training procedures are not impacted, feel free to continue using symlinks to efficiently manage your datasets.