Inquiry on GPU utilization of pytorch/pytorch:1.8.1-cuda11.1-cudnn8-runtime image
When pytorch saw “cpu_mkl_py”, it only used the CPU.
The runtime pytorch image provided by pytorch seems to have worked well because it contains the “py3.10_cuda11.8_cudnn8.7.0_0” version of the package.
However, I also checked the GPU device in the redhat software “openshfit”, but unlike in existing python’s jupyter-notebook,
I did not configure the nvidia.com/gpu:1 device by filling in the deployment, but I brought the image of pytorch as is and checked it,
and the GPU was found. It comes with a card.
I have questions about how that part is different, but it was resolved well.
checking the device
When running in python, the device was also confirmed well.
import torch
print(torch.cuda.get_device_name(0))
=> NVIDIA GTX 6000
print(torch.cuda.is_available())
=> True