GPU not available repeatedly when run torch.cuda_is_available() | Runpod | Page 1

hexed oasis Jan 28, 2026, 7:48 AM

#

In last 5 hours, I've been running on the same problem: GPU not available when using torch cuda on different GPU, different pods. Are some error in your GPU pods?

NVIDIA-SMI returned all specs succesfully
this error I found on H100 GPU

spiral tundraBOT Jan 28, 2026, 7:48 AM

#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

wicked wraithBOT Jan 28, 2026, 7:48 AM

#

marsh wharf Jan 28, 2026, 2:43 PM

#

hexed oasis In last 5 hours, I've been running on the same problem: GPU not available when u...

Check if nvidia-smi shows it, if so its usually a mismatch between your cuda version and the driver

#

Where the cuda your using either is to old for that GPU or the cuda version is higher than the cuda version shown in nvidia-smi

hallow barn Jan 28, 2026, 11:09 PM

#

@hexed oasis can you give me output of these?

ls -l /dev/nvidia* || echo "NO /dev/nvidia* inside container" which nvidia-smi nvidia-smi

#

also do you know which version of torch are you running? or give me your template name

#GPU not available repeatedly when run torch.cuda_is_available()