Skip to content

cudaPackages_11_8: init, with support for jetsons#194791

Closed
SomeoneSerge wants to merge 4 commits intoNixOS:masterfrom
SomeoneSerge:cuda-11.8
Closed

cudaPackages_11_8: init, with support for jetsons#194791
SomeoneSerge wants to merge 4 commits intoNixOS:masterfrom
SomeoneSerge:cuda-11.8

Conversation

@SomeoneSerge
Copy link
Contributor

Description of changes

CUDA 11.8 has been released, this time including redist packages for jetsons.
This PR adds cuda 11.8 redist packages with "x86_64-linux" and "aarch64-linux" platforms.
This also packages cudnn 8.6.0 (x86_64 only), since 8.4.0 doesn't officially support cuda 11.8.

Jetson support is still very limited in that there's still not redist packages for CUDNN on aarch64-linux (-> we cannot build pytorch).
I haven't meaningfully tested jetson support yet.

Things done
Notify maintainers

@NixOS/cuda-maintainers

defaultBuild = { "tensorrt" = allBuilds.${computeName tensorRTDefaultVersion}; };
in allBuilds // defaultBuild;
defaultBuildName = computeName tensorRTDefaultVersion;
defaultBuild =
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm only touching this because of eval errors.
But I'm also really overwhelmed reading this piece

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fix seems good to me.

We create an attribute tensorr_${version} for every available tensorrt version. Then we add a tensorrt attribute (so without version, however, for cudatoolklit 11.8 there is no compatible tensorrt`, and hence we cannot add a default.

In the future we might need a fix like this for the other extensions as well.

@samuela
Copy link
Member

samuela commented Oct 6, 2022

also note that there is #194705 open as well

@SomeoneSerge
Copy link
Contributor Author

SomeoneSerge commented Oct 6, 2022

Ouch, missed it, didn't get a notification. I guess I'll wait for the other PR and then poke the jetson support

@SomeoneSerge SomeoneSerge added the 6.topic: cuda Parallel computing platform and API label Oct 6, 2022
@dguibert dguibert mentioned this pull request Oct 7, 2022
13 tasks
defaultBuildName = computeName tensorRTDefaultVersion;
defaultBuild =
if
builtins.hasAttr defaultBuildName allBuilds then
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

could use the ? operator

Copy link
Contributor Author

@SomeoneSerge SomeoneSerge Oct 12, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I thought so too, but then I realized I'd have to use it with an interpolation: ? ${defaultBuildName}. Wasn't sure if that's better than hasAttr

Anyway, @ dguibert handles this differently in # 194705 and I'm yet to have looked at it thoroughly

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

6.topic: cuda Parallel computing platform and API

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants