You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Some (all) our images only need CUDA Runtime and do not need the rest of the CUDA SDK. Therefore, images could be made smaller if we only install the runtime. The Python packages (Tensorflow, Pythorch, ...) should only need the runtime to function.
I think the suggestion is sensible if image sizes ever become a concern. Installing everything (dnf install -y cuda-toolkit-11-8) only takes up space. I think that the tradeoff (more disk/network usage in exchange for convenience) is good for now.
This was suggested by @ben-alkov (as a general idea regarding CUDA in images, without specific investigation of the OpenDataHub situation).
Feel free to close this if no near-term action is intended to be taken on this suggestion.
The text was updated successfully, but these errors were encountered:
Some (all) our images only need CUDA Runtime and do not need the rest of the CUDA SDK. Therefore, images could be made smaller if we only install the runtime. The Python packages (Tensorflow, Pythorch, ...) should only need the runtime to function.
I think the suggestion is sensible if image sizes ever become a concern. Installing everything (
dnf install -y cuda-toolkit-11-8
) only takes up space. I think that the tradeoff (more disk/network usage in exchange for convenience) is good for now.This was suggested by @ben-alkov (as a general idea regarding CUDA in images, without specific investigation of the OpenDataHub situation).
Feel free to close this if no near-term action is intended to be taken on this suggestion.
The text was updated successfully, but these errors were encountered: