| 24 Oct 2022 |
|  waa left the room. | 08:10:13 |
| 31 Oct 2022 |
|  underpantsgnome changed their display name from underpantsgnome! to underpantsgnome. | 20:29:26 |
| 6 Nov 2022 |
 tpw_rules | where is a link to the hercules instance again? | 01:47:12 |
| tpw_rules | i found it but it seems like something is wedged: https://hercules-ci.com/github/SomeoneSerge/nixpkgs-unfree | 01:54:44 |
| tpw_rules | any idea why recent builds haven't completed? was there a staging merge a while ago? | 01:54:56 |
| tpw_rules | Someone S: ? | 02:05:42 |
 SomeoneSerge (back on matrix) | Oh wow. I'll have a look | 17:22:38 |
| SomeoneSerge (back on matrix) | Agent systemd service was somehow failing to resolve some domain names since October 22. I restarted it, seems to be online now | 17:27:19 |
| tpw_rules | it looks like things are better now? i guess now i just have to be patient for the builds | 19:14:31 |
| tpw_rules | https://github.com/NixOS/nixpkgs/pull/199910 | 21:59:28 |
| 8 Nov 2022 |
|  pbsds changed their profile picture. | 00:46:17 |
| 9 Nov 2022 |
|  breakds joined the room. | 08:01:07 |
|  eahlberg joined the room. | 16:24:14 |
| breakds | Hi! First I would like to say "thank you" to the Nix CUDA maintainers and the community! | 16:47:27 |
| breakds | I am trying to build pytorch with CUDA 11.08, and it complaints about cudnn:
at /nix/store/171cjmpyl45dz6dy818i3kf7x3nijkpg-source/pkgs/development/libraries/science/math/cudnn/generic.nix:28:1:
27|
28| assert useCudatoolkitRunfile || (libcublas != null);
| ^
29|
Being not quite familiar with how CUDA libraries are packaged, I am wondering since cudaPackages_11_8 does not seem to have libcublas, should I turn on useCudatoolkitRunfile (it is false by default) for CUDA 11.8? Judging from the name, it seems to suggest that I can find libcublas files from the cudatoolkit itself, am I wrong?
Thanks a lot!
| 16:47:30 |
| SomeoneSerge (back on matrix) | Ouch. This is actually a quite embarrassing overlook | 19:57:20 |
| breakds | Thanks for taking a look at this. Is it that cuda 11.8 should be handled differently? | 20:00:20 |
| breakds | https://github.com/NixOS/nixpkgs/blob/master/pkgs/development/libraries/science/math/cudnn/extension.nix#L11 I am looking at this line | 20:00:33 |
| breakds | Not sure why there is a difference between 11.4+ and 11.4- | 20:01:20 |
| SomeoneSerge (back on matrix) | No, it's much simpler than that:) https://github.com/NixOS/nixpkgs/pull/200426 | 20:02:29 |
| breakds | Thanks for the fix! I need to learn more about how cuda packaging works. | 20:03:59 |
| SomeoneSerge (back on matrix) | Regarding this particular issue: you probably already found this, but the redist packages' attributes (like libcublas) are populated from a json file, stored right in the repo. With every release we download the json "manifest" file from nvidia and keep it as is | 20:13:37 |
| breakds | In reply to @ss:someonex.net
Regarding this particular issue: you probably already found this, but the redist packages' attributes (like libcublas) are populated from a json file, stored right in the repo. With every release we download the json "manifest" file from nvidia and keep it as is Thanks for the explanation! It makes sense to me now. | 20:25:42 |
| breakds | A separate question, as I read from https://discourse.nixos.org/t/announcing-the-nixos-cuda-maintainers-team-and-a-call-for-maintainers/18074 , is x86_64-linux computing cycle still needed for github actions? I have a spare RTX 3080 not attached to any machine at this moment, not sure what is the best way to make it useful to the project. Shall I build a machine to run github actions? | 20:35:33 |
| SomeoneSerge (back on matrix) | Samuel Ainsworth had started building a gpu-enabled testsuite, but I'm not aware of anything like that running on a regular basis.
We do have builds running that populate cachix: https://app.cachix.org/cache/cuda-maintainers
These builds don't require (and can't use) a GPU.
We do, as a matter of fact, need more cpu compute just for the builds, or better yet a more sustainable option like a dedicated buildbox maintained by some committed organisation. We haven't got around to make any calls yet
And there's https://github.com/samuela/nixpkgs-upkeep which, I think, uses github workers
| 20:48:41 |
| SomeoneSerge (back on matrix) | GPU-enabled sanity checks (even for things like "torch being able to allocate a tensor on a device") would be very nice, but somebody has to implement them | 20:53:49 |
| breakds | Thanks for the introduction to how the builds and the packages are being tested. Agree that it might make sense to add some GPU-enabled tests such as training a very small model and verify it is successful. | 21:03:07 |
| breakds | And this also looks like something I am capable of implementing and setup a dedicate machine to run. Probably need some guidance on which repo to put them into and get started. | 21:04:24 |
| SomeoneSerge (back on matrix) | ...so far all of the "CI" infrastructure for CUDA has been implemented outside nixpkgs | 21:05:00 |
| breakds | Do you mean this is not ideal? | 21:05:39 |
