RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu.
I am using CUDA 12.9, and I have also set "Allowed CUDA Versions" to 12.9, but sometimes an error occurs. Could you please tell me how to solve this problem? logs show RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu.
2 Replies
hey we started getting this as well. would be good to have some insights. im using runpod mainly for ffmpeg work so when the gpu drops im kinda stuck
ive asked this internally, will try to get a response for this.
@d1g17al / @billchen
I talked to some ppl internally, I believe that CUDA 12.9 is unstable in general. Try to use filters to specify 12.8 or lower.
(Did get a couple machines checked out specifically and they do have the cuda installed, but runs into some issues around memory etc)
https://docs.nvidia.com/datacenter/tesla/drivers/#cuda-toolkit-driver-and-architecture-matrix
12.9 I was told is a new feature branch, but is unstable. So working with partners to get things back to 12.8 where it should be more stable 🙂