Closed infermachine closed 5 years ago
Are you trying to build on Singularity Hub, or just on your host?
This issue to me looks like you want to report it here https://github.com/sylabs/singularity/issues with the Singularity software. My .02 - you'll need to install the exact same libraries in the container that you have on the host, and don't forget to use the --nv
flag! Other than that, you should open the issue on the board linked. Good luck!
Ok yeah I'll post it over there instead. I posted here because this is the only place I could find buildfile examples with cuda10 (e.g. https://singularity-hub.org/containers/6713)... but most of those give the "cannot open 385: No such file" error.
I think you're right that I may need to update the nvidia drivers on the kernel or something because I can't seem to find how to install 410.93 on the client.
Thanks!
Sure, glad to help, good luck!
I have a few nvidia 2080ti GPUs on a node in our cluster and I want to install cuda10 in a singularity container so I can work with them (primarily with pytorch). When I use this singularity base image:
From: nvidia/cuda:10.0-devel-ubuntu18.04
...the container seems to build fine, but when I try to singularity run/exec/shell the image I get this error:
/.singularity.d/actions/run: 7: /.singularity.d/env/10-docker.sh: cannot open 385: No such file
I also tried just building the barebones ubuntu and installing cuda manually, ie.
But then once I'm running the container and try running nvidia-smi I get the following error:
I think the node has nvidia kernel 410.93 installed and this is interfering somehow? Anyone have any ideas how to fix this?
Thanks! DB