search

marcoslucianops / DeepStream-Yolo

NVIDIA DeepStream SDK 7.0 / 6.4 / 6.3 / 6.2 / 6.1.1 / 6.1 / 6.0.1 / 6.0 / 5.1 implementation for YOLO models
MIT License
1.45k stars 357 forks source link

Multiple GPUs appear and GPUs are not used #374

Open lzylzylzy123456 opened 1 year ago

lzylzylzy123456 commented 1 year ago

Have you ever encountered this situation? I couldn't find the answer on the NVIDIA forum. I have already asked a question, and this is my question link. https://forums.developer.nvidia.com/t/multiple-gpus-appear-and-gpus-are-not-used/256306

marcoslucianops commented 1 year ago

The DeepStream doesn't use good multiple GPUs. It's better to run 2 codes, one in each GPU (using export CUDA_VISIBLE_DEVICES=0, export CUDA_VISIBLE_DEVICES=1, etc).

lzylzylzy123456 commented 1 year ago

The official response is that we still have some plugins that can set gpu-id, such as nvstreammux, videoconvert, etc. After modifying according to this method, it is indeed useful. However, as the number of pipelines increases, the GPU utilization rate is very low. I have also asked you this question. Please take a look. Thank you. The name of this issue is Low GPU utilization

marcoslucianops commented 1 year ago

The method I said is better than set the gpu-id. You will have more performance using one pipeline per GPU.