 | NixOS CUDA | 279 Members |
| CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda | 57 Servers |
| NixOS CUDA | 279 Members |
| CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda | 57 Servers |
| Sender | Message | Time |
|---|---|---|
| 20 Nov 2025 | ||
 ser(ial) | does it mean that i need full libraries in nixos container which are with identical version as on debian host? | 13:32:26 |
 connor (burnt/out) (UTC-8) | GaƩtan Lepage: I've got to package ONNX/ONNX Runtime/ONNX TensorRT for C++; if I upstream the PR do you think you'd have the bandwidth to look at it? I'd likely follow what I did here: https://github.com/ConnorBaker/cuda-packages/tree/8a317116a07717b13e0608f47b78bd6d75f8bb99/pkgs/development/libraries That is, the sort of cursed double-build in a single derivation which produces both the C++ binaries and a python wheel, so the python3Packages entry essentially turns into installing a wheel. | 14:04:07 |
 teto | are there differences between https://nix-community.cachix.org and https://cache.nixos-cuda.org . My goal is to gain access to cuda-enable packages for unstable | 14:24:20 |
| connor (burnt/out) (UTC-8) | community cache is no longer being populated, use the latter | 14:27:28 |
| connor (burnt/out) (UTC-8) | * community cache is no longer being populated with CUDA packages, use the latter | 14:27:35 |
| connor (burnt/out) (UTC-8) | note to self whenever I find the time (š« ) update cudnn-frontend; looks like the version I made here https://github.com/ConnorBaker/cuda-packages/blob/8a317116a07717b13e0608f47b78bd6d75f8bb99/pkgs/development/cuda-modules/packages/cudnn-frontend.nix is newer than what we have in master and requires fewer patches. Also should add brokenAssertions since IIRC cudnn-frontend has docs about the versions of cuDNN it supports | 14:30:28 |
 SomeoneSerge (back on matrix) | Yeah you can have the CDI hook mount the driver to /run/opengl-driver/lib, or, IIRC, the hook on NixOS even adds the driver to LD_LIBRARY_PATH | 14:31:39 |
 GaƩtan Lepage | Yes connor (burnt/out) (UTC-8), I can look at it. But don't we already have onnxruntime and onnx? | 15:33:58 |
| connor (burnt/out) (UTC-8) | IIRC only onnxruntime is available as both a C++ and Python library, ONNX is only available as a python package | 15:34:30 |
| GaƩtan Lepage | Yes true. | 15:35:04 |
| 21 Nov 2025 | ||
 Jeremy Fleischman (jfly) | i'm trying out programs.nix-required-mounts.presets.nvidia-gpu.enable, and am running into https://github.com/NixOS/nix/issues/9272 with remote builds. is this the best workaround? https://github.com/nix-community/infra/pull/1807#discussion_r2458816021 | 20:03:32 |
| connor (burnt/out) (UTC-8) | Yikes I mean I guess that would work? | 20:17:04 |
| Jeremy Fleischman (jfly) | i'd love to hear a better one! | 20:17:37 |
| Jeremy Fleischman (jfly) | (btw, this programs.nix-required-mounts.presets.nvidia-gpu.enable stuff is super cool. seems to be 99% of the way there) | 20:18:35 |
| Jeremy Fleischman (jfly) | also, i'd like to buy a beer for whoever wrote this error message:
| 20:19:44 |
| Jeremy Fleischman (jfly) | Ah, SomeoneSerge (back on matrix), looks like I owe you a beer: https://github.com/nixos/nixpkgs/commit/3a0d777486191cbdd2ffc7e31f25156da94c4831 | 20:22:04 |
| 22 Nov 2025 | ||
| connor (burnt/out) (UTC-8) | I'm going to revisit https://github.com/NixOS/nixpkgs/pull/459416 today and do some final fixups; I'd really like if we could merge it soon SomeoneSerge (back on matrix) GaƩtan Lepage given its absence is making my life more difficult | 14:07:58 |
| GaƩtan Lepage | I'm in a call with Serge. We'll have a look at it now. | 14:20:00 |
| SomeoneSerge (back on matrix) | Let's have a look | 14:20:23 |
| SomeoneSerge (back on matrix) | I'm still unconvinced by the approach... | 14:20:41 |
| SomeoneSerge (back on matrix) | But since we're not actively benefitting form the stubs right now, maybe we at least merge it for now (without promises to keep) | 14:21:01 |
| 23 Nov 2025 | ||
| SomeoneSerge (back on matrix) | We've had a peek with Gaetan yesterday and LOL turns out the messages aren't even defined in the stubs, but in downstream libraries like in cudart | 07:49:49 |
 jaredmontoya | Recently firefox and thunderbird started being rebuilt with cudaSupport enabled because they depend on some other package that now has cuda support. For my laptop with 32GB of ram it takes ages to compile either of them and it eventually runs out of RAM when it's time to link the binary. It seems that they were added to release-cuda in https://github.com/NixOS/nixpkgs/pull/437809 so they should be cached by the nix-community cache. Yet even after waiting for the new commit to be cached nix still tries to build thunderbird and firefox. This seems to only be possible if hydra.nixos-cuda.org doesn't push to nix-community or if it signs things with a key I don't trust. Here is my configuration: Does anyone know what I have to do to get build outputs from the cuda hydra? | 10:05:21 |
| jaredmontoya | * Recently firefox and thunderbird started being rebuilt with cudaSupport enabled because they depend on some other package that now has cuda support. For my laptop with 32GB of ram it takes ages to compile either of them and it eventually runs out of RAM when it's time to link the binary. It seems that they were added to release-cuda in https://github.com/NixOS/nixpkgs/pull/437809 so they should be cached by the nix-community cache. Yet even after I waited for half a week for the new commit to be cached nix still tries to build thunderbird and firefox. This seems to only be possible if hydra.nixos-cuda.org doesn't push to nix-community or if it signs things with a key I don't trust. Here is my configuration: Does anyone know what I have to do to get build outputs from the cuda hydra? | 10:07:38 |
| jaredmontoya | * Recently firefox and thunderbird started being rebuilt with cudaSupport enabled because they depend on some other package that now has cuda support. For my laptop with 32GB of ram it takes ages to compile either of them and it eventually runs out of RAM when it's time to link the binary. It seems that they were added to release-cuda in https://github.com/NixOS/nixpkgs/pull/437809 so they should be cached by the nix-community cache. Yet even after I waited for half a week or more for the new commit to be cached nix still tries to build thunderbird and firefox. This seems to only be possible if hydra.nixos-cuda.org doesn't push to nix-community or if it signs things with a key I don't trust. Here is my configuration: Does anyone know what I have to do to get build outputs from the cuda hydra? | 10:07:55 |
| jaredmontoya | * Recently firefox and thunderbird started being rebuilt with cudaSupport enabled because they depend on some other package that now has cuda support. For my laptop with 32GB of ram it takes ages to compile either of them and it eventually runs out of RAM when it's time to link the binary even with --max-jobs 1. It seems that they were added to release-cuda in https://github.com/NixOS/nixpkgs/pull/437809 so they should be cached by the nix-community cache. Yet even after I waited for half a week or more for the new commit to be cached nix still tries to build thunderbird and firefox. This seems to only be possible if hydra.nixos-cuda.org doesn't push to nix-community or if it signs things with a key I don't trust. Here is my configuration: Does anyone know what I have to do to get build outputs from the cuda hydra? | 10:16:54 |
| GaƩtan Lepage | Indeed, it does not push to the nix-community Cachix instance.You can use our new cache instead. | 10:51:20 |
| jaredmontoya | Thank you! | 10:54:45 |
| jaredmontoya | * | 10:55:02 |
| SomeoneSerge (back on matrix) | connor (burnt/out) (UTC-8): {isDeclared{Array,Map},sortArray}.bash very nice | 11:57:04 |