How to set which GPU to use？

[kli-casia]

Thank you very much!

[madhawav]

Hi,
Currently this feature is not implemented in YOLO34Py. It would choose the GPU chosen by Darknet. I will look into implementing it in the future.
Thanks

[jantman]

I was also looking into this, but my really poor C knowledge doesn't make it easy.

It looks like the darknet example script both sets the gpu_index int and calls cuda_set_device(). It appears that this can also be used to disable using the GPU even if the code was compiled for GPU support, by setting the gpu_index to -1

[kli-casia]

still waiting for this feature, thx

[madhawav]

@kli-nlpr The new merge request #25 has this feature implemented. Can you try it out and let me know whether it works as expected? Then I would merge it to master and do a new release. Unfortunately, I do not have multiple GPUs to test the code.

1. Uninstall the version you have installed right now (pip uninstall yolo34py-gpu)
2. Clone the code in merge request Added set_cuda_device #25
3. Set environment variables GPU=1 and OPENCV=1.
4. Run pip install .
5. Use pydarknet.set_cuda_device to change the GPU.

@jantman Thank you for your support. The link you mentioned helped me!

[jantman]

Sure! For what it's worth, I was actually hoping to get this so that (I hope) it would work like the darknet option and setting the GPU device to zero would disable GPU mode and use CPU instead.

I've been having some issues with the detections using yolov3-tiny on GPU, and was looking to write something that comes back after-the-fact and runs my images through the full yolov3 in CPU mode using otherwise the same code.

[madhawav]

Hi @jantman, I believe setting the GPU to zero would make it use the Primary GPU. Why don't you use the version without GPU? (pip install yolo34py instead of pip install yolo34py-gpu).

[jantman]

@madhawav That's what I'm doing now, but it means that I need to use two different virtualenvs... my goal is to compare the results of the tiny GPU and non-tiny CPU.

For the command-line darknet binary, at least, setting GPU 1 is primary GPU and GPU 0 disables GPU mode.

[madhawav]

Option to select GPU is implemented with #25

[woojeonghoon]

have eight GPUs.

By the way, 'python3 video_demo' only uses the first GPU and nothing else

What should I do??