NVIDIA-SMI 516.94 RuntimeError: No CUDA GPUs are available

vision
1

image
cuda
Hi, I’ve the following issue in my program. NVIDIA GeForce RTX 3080 with driver version 516.94 have been installed on my PC.
I tried uninstall and reinstall cudatookit, pytorch, and reinstalling. torch.cuda.is_available() is still False though torch.version.cuda can check as
11.3.
Any guidance would be warmly welcome and thanks in advance.

2

Your system is most likely not able to communicate with the driver, which could happen e.g. if you didn’t restart the machine after a driver update.
Make sure other CUDA samples are running first, then check PyTorch again.

1 Like
3

Thanks for your suggestion. Now the CUDA is available but I’d like to know why CUDA device count 1 is displayed in red color?
image

4

That’s your IDE’s color scheme and isn’t defined by PyTorch.

1 Like
5

Hey @naychelynn , how did you solve the error? I am using GeForce RTX 3090.

6

I exactly follow @ptrblck 's suggestion “Restart my PC”.
I’ve tested on other environments too and it’s the same. ( installing torch without restarting then the same problem and CUDA is available after restart process).