[Windows] CUDA installation fixes, add 11.1 and document absence of nvcuda.dll

[jaimergp]

Checklist

• Used a personal fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

I have been trying to provide nvcuda.dll in conda-forge/openmm-feedstock#36 only to realize it is not possible, so I am documenting it here in case somebody wants to try again.

• The CUDA toolkit installer is supposed to provide the drivers too, but it will silently fail to do that if no GPU is present
• I thought that maybe getting the Tesla/datacenter drivers would allow a GPU-less install. Well, it's not the case.
• Old versions of the drivers allowed to manually extract nvcuda64.dl_ into C:\Windows\system32 as nvcuda.dll, but that doesn't seem to be the case with most recent ones -- DLL will fail to load with error 193.

Basically the only way to provide nvcuda.dll involves breaking the EULA since you would have to redistribute the DLL from a working installation to bypass the hardware checks, and we can't do that.

Since I had to experiment with a few things to arrive to this conclusion, I am also adding a few fixes:

• Enable path to CUDA 11.1 (coming soon)
• Make sure we wait for the installers to finish!
• Install only the needed components (no samples, documentation, etc) to save time and space
• Make sure the CUDA failures are reflected as such in the CI status

Status

• (Linux only) Blocked by Add GHA draft implementation docker-images#161 & Add CUDA 11.1 docker-images#163
• (All platforms) cudatoolkit 11.1 is only available in conda-forge/label/test for now.
• Trying to fix [Windows] CUDA_HOME cannot be found under some circumstances nvcc-feedstock#53

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[jaimergp]

@conda-forge/core we need this for CUDA 11.1, now merged. How should I proceed here? Merge master? Rebase? New PR?

[beckermr]

Resolve the conflicts and make sure the version is ahead. Then we should be good to go.

[beckermr]

@conda-forge-admin rerender

Just to make sure the merge went ok.

[beckermr]

Note this won't build for travis. Also, we might want some code review from someone more knowledgeable than me on the CI setup package itself. cc @conda-forge/core

[github-actions]

Hi! This is the friendly automated conda-forge-webservice.
I tried to rerender for you, but it looks like I wasn't able to push to the windows-cuda-driver
branch of jaimergp-feedstocks/conda-forge-ci-setup-feedstock. Did you check the "Allow edits from maintainers" box?

[jaimergp]

I also forgot I need to add the CUDA 11.1 matrix entries, which will fail on Linux due to the missing Docker image (see conda-forge/docker-images#160).

[beckermr]

Can you fork to a non-org place? That way others can rerender.

[jaimergp]

Can you fork to a non-org place? That way others can rerender.

Oops, I didn't know that was a limitation. I'll transfer back to my personal account.

[jakirkham]

It is. Though one can also re-render locally.

[jaimergp]

@conda-forge/core this is ready for review!

[jaimergp]

May I ask for a merge if nobody is against it? 😄