RuntimeError: CUDA error: no kernel image is available for execution on the device

1

Hi everyone,
I know there is a lot of questions about “CUDA error no kernel…”, but I think I have a different issue.
I’m working on the university cluster, and I open virtual-env and install all the packages include torch-Cuda 9.2. When I try to run the program I get the error: RuntimeError: CUDA error: no kernel image is available for execution on the device.

Now, I try to figure out if I need to install a regular torch without Cuda like torch 1.0.0.
Can someone having this problem ?

Eran

2

What GPUs is your cluster using?
Note that the binaries ship with pre-built CUDA, cudnn, etc. for GPUs with compute capability >=3.7.
If the GPUs you are using have an older cc, you might need to build PyTorch from source.

3

Yes I know that there is a lot of clusters, my question is if I install the simple version like torch == 1.0.0
it will install Cuda?
Because I think I don’t need to install Cuda, there is some version of Cuda on the clusters.

4

The binaries ship with their own CUDA, cudnn etc. so your local CUDA installation won’t be used.
Note that I was referring to the compute capability of your GPUs, not CUDA, in case there is a misunderstanding.