!eWOErHSaiddIbsUNsJ:nixos.org

NixOS CUDA

290 Members
CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda58 Servers

Load older messages


SenderMessageTime
16 Jan 2026
@connorbaker:matrix.orgconnor (burnt/out) (UTC-8)I don't follow; IIRC hwloc is pulled in by those directly and would use it untouched22:48:00
17 Jan 2026
@glepage:matrix.orgGaétan Lepage FYI, the onnxruntime bump was merged! 🎉
https://github.com/NixOS/nixpkgs/pull/450587
09:33:45
@kaya:catnip.eekaya 𖤐Hey. Im working on upstreaming packages for exllama & tabbyapi. I would appreciate if somebody from here could review the PR :) https://github.com/NixOS/nixpkgs/pull/48112119:11:25
18 Jan 2026
@glepage:matrix.orgGaétan Lepage Anyone could give a second opinion on https://github.com/NixOS/nixpkgs/pull/474676 please? 09:48:06
@ss:someonex.netSomeoneSerge (back on matrix)This particular regression could've been avoided with autopatchelf. Maybe switch to that?12:08:40
@glepage:matrix.orgGaétan Lepage Just adding autoPatchelfHook you mean? 22:29:11
@ss:someonex.netSomeoneSerge (back on matrix)And removing --set-rpath23:40:44
20 Jan 2026
@kaya:catnip.eekaya 𖤐 Gaétan Lepage: Thank you for the through PR review, much appreciated! 19:03:45
@glepage:matrix.orgGaétan LepageYou're welcome! I'll probably wait for a second committer to double check, but we should be able to merge it soon :)19:26:32
21 Jan 2026
@connorbaker:matrix.orgconnor (burnt/out) (UTC-8) Gaétan Lepageif you haven’t already, can you link any of the additional work you’ve had to do for the OpenCV version bump in the PR description? I remember yesterday you had mentioned needing to fix protobuf, but that might have been about ONNX Runtime. 17:20:11
@glepage:matrix.orgGaétan Lepage Indeed, this was for protobuf. The opencv effort is all contained in https://github.com/NixOS/nixpkgs/pull/479136 21:33:06
@glepage:matrix.orgGaétan Lepage The protobuf patching related to onnxruntime was merged into staging.
-> https://github.com/NixOS/nixpkgs/pull/480716
21:33:34
22 Jan 2026
@glepage:matrix.orgGaétan Lepage

Sadness.

I had to package the 12.8 version of cuda-bindings, because we're on cuda 12.8.
However, it dlopens libcuda.so and the driver (on my system at least) tells it that it uses cuda 13.1.
Hence, it errors out :/

https://github.com/NixOS/nixpkgs/pull/482638

(cuda-bindings is supposed to be required in the new torch update, not sure how critical it is though)

12:06:42
@justbrowsing:matrix.orgKevin Mittman (UTC-8) Hmm a bit confused about that issue, as nvidia-smi shows the maximum CUDA version supported and with "backwards compat" should always be able to use a newer NVIDIA driver than a CUDA Toolkit was released with 20:14:30
@glepage:matrix.orgGaétan Lepage The version detection logic happens here and errors out here. 20:48:39
@glepage:matrix.orgGaétan Lepage

Here's the stack trace when I re-enable the failing test:

______________________ ERROR collecting test_nvjitlink.py ______________________
tests/test_nvjitlink.py:62: in <module>
    not check_nvjitlink_usable(), reason="nvJitLink not usable, maybe not installed or too old (<12.3)"
        ^^^^^^^^^^^^^^^^^^^^^^^^
tests/test_nvjitlink.py:58: in check_nvjitlink_usable
    return inner_nvjitlink._inspect_function_pointer("__nvJitLinkVersion") != 0
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
cuda/bindings/_internal/nvjitlink.pyx:254: in cuda.bindings._internal.nvjitlink._inspect_function_pointer
    ???
cuda/bindings/_internal/nvjitlink.pyx:257: in cuda.bindings._internal.nvjitlink._inspect_function_pointer
    ???
cuda/bindings/_internal/nvjitlink.pyx:205: in cuda.bindings._internal.nvjitlink._inspect_function_pointers
    ???
cuda/bindings/_internal/nvjitlink.pyx:99: in cuda.bindings._internal.nvjitlink._check_or_init_nvjitlink
    ???
cuda/bindings/_internal/nvjitlink.pyx:57: in cuda.bindings._internal.nvjitlink.load_library
    ???
cuda/bindings/_internal/utils.pyx:137: in cuda.bindings._internal.utils.get_nvjitlink_dso_version_suffix
    ???
E   cuda.bindings._internal.utils.NotSupportedError: only CUDA 12 driver is supported
------------------------------- Captured stdout --------------------------------
+++++++++++++++ 13010
22:27:19
@glepage:matrix.orgGaétan Lepage The code block mentionned above returns 13010 which fails the 12000 <= driver_ver < 13000 check in cuda/bindings/_internal/utils.pyx. 22:28:37
23 Jan 2026
@glepage:matrix.orgGaétan Lepage Weird regression following the merge of staging-next... 10:42:06
@arilotter:matrix.orgAri Lotter

hm i'm getting an infinite hang on pytestCheckPhase for python3.12-keras3.13.1 since the 3.13.1 update 🤔 https://github.com/NixOS/nixpkgs/commit/73806a397d9f9360f331ed98202e5cecc1230548
building python3.12-keras-3.13.1 (pytestCheckPhase): keras/src/trainers/trainer_test.py::TestTrainer::test_fit_with_data_adapter_py_dataset_infinite_cw PASSED [ 94%]
hang happens every time

nix-top shows

  UID     PID    PPID STIME     TIME COMMAND
30001 3422767 3422639 15:44 00:00:00 bash -e /nix/store/l622p70vy8k5sh7y5wizi5f2mic6ynpg-source-stdenv.sh /nix/store/shkw4qm9qcw5sc5n1k5jznc83ny02r39-default-builder.sh
30001 3425989 3422767 15:45 00:17:09 /nix/store/iizfzmqjdfvbyx82gqw92glb5zl6m4my-python3-3.12.12/bin/python3.12 -m pytest --ignore-glob=keras/src/layers/preprocessing/data_layer_test.py --ignore-glob=keras/src/layers/preprocessing/image_preprocessing/resizing_test.py --ignore-glob=keras/src/layers/preprocessing/rescaling_test.py --ignore-glob=keras/src/trainers/data_adapters/grain_dataset_adapter_test.py --ignore-glob=keras/src/export/onnx_test.py --ignore-glob=integration_tests/dataset_tests --ignore-glob=keras/src/applications/applications_test.py --ignore-glob=integration_tests/jax_custom_fit_test.py --ignore-glob=integration_tests/tf_distribute_training_test.py --ignore-glob=integration_tests/torch_custom_fit_test.py --ignore-glob=integration_tests/torch_workflow_test.py --ignore-glob=keras/src/backend/tensorflow/saved_model_test.py -k not (test_fit_with_data_adapter_grain_dataloader) and not (test_fit_with_data_adapter_grain_datast) and not (test_fit_with_data_adapter_grain_datast_with_len) and not (test_image_dataset_from_directory_binary_grain) and not (test_image_dataset_from_directory_color_modes_grain) and not (test_image_dataset_from_directory_crop_to_aspect_ratio_grain) and not (test_image_dataset_from_directory_follow_links_grain) and not (test_image_dataset_from_directory_manual_labels_grain) and not (test_image_dataset_from_directory_multiclass_grain) and not (test_image_dataset_from_directory_no_labels_grain) and not (test_image_dataset_from_directory_not_batched_grain) and not (test_image_dataset_from_directory_pad_to_aspect_ratio_grain) and not (test_image_dataset_from_directory_shuffle_grain) and not (test_image_dataset_from_directory_validation_split_grain) and not (test_sample_count_grain) and not (test_text_dataset_from_directory_binary_grain) and not (test_text_dataset_from_directory_follow_links_grain) and not (test_text_dataset_from_directory_manual_labels_grain) and not (test_text_dataset_from_directory_multiclass_grain) and not (test_text_dataset_from_directory_not_batched_grain) and not (test_text_dataset_from_directory_standalone_grain) and not (test_text_dataset_from_directory_validation_split_grain) and not (test_keras_imports) and not (test_reloading_default_saved_model)
30001 3797431 3425989 15:59 00:00:00 /nix/store/iizfzmqjdfvbyx82gqw92glb5zl6m4my-python3-3.12.12/bin/python3.12 -m pytest --ignore-glob=keras/src/layers/preprocessing/data_layer_test.py --ignore-glob=keras/src/layers/preprocessing/image_preprocessing/resizing_test.py --ignore-glob=keras/src/layers/preprocessing/rescaling_test.py --ignore-glob=keras/src/trainers/data_adapters/grain_dataset_adapter_test.py --ignore-glob=keras/src/export/onnx_test.py --ignore-glob=integration_tests/dataset_tests --ignore-glob=keras/src/applications/applications_test.py --ignore-glob=integration_tests/jax_custom_fit_test.py --ignore-glob=integration_tests/tf_distribute_training_test.py --ignore-glob=integration_tests/torch_custom_fit_test.py --ignore-glob=integration_tests/torch_workflow_test.py --ignore-glob=keras/src/backend/tensorflow/saved_model_test.py -k not (test_fit_with_data_adapter_grain_dataloader) and not (test_fit_with_data_adapter_grain_datast) and not (test_fit_with_data_adapter_grain_datast_with_len) and not (test_image_dataset_from_directory_binary_grain) and not (test_image_dataset_from_directory_color_modes_grain) and not (test_image_dataset_from_directory_crop_to_aspect_ratio_grain) and not (test_image_dataset_from_directory_follow_links_grain) and not (test_image_dataset_from_directory_manual_labels_grain) and not (test_image_dataset_from_directory_multiclass_grain) and not (test_image_dataset_from_directory_no_labels_grain) and not (test_image_dataset_from_directory_not_batched_grain) and not (test_image_dataset_from_directory_pad_to_aspect_ratio_grain) and not (test_image_dataset_from_directory_shuffle_grain) and not (test_image_dataset_from_directory_validation_split_grain) and not (test_sample_count_grain) and not (test_text_dataset_from_directory_binary_grain) and not (test_text_dataset_from_directory_follow_links_grain) and not (test_text_dataset_from_directory_manual_labels_grain) and not (test_text_dataset_from_directory_multiclass_grain) and not (test_text_dataset_from_directory_not_batched_grain) and not (test_text_dataset_from_directory_standalone_grain) and not (test_text_dataset_from_directory_validation_split_grain) and not (test_keras_imports) and not (test_reloading_default_saved_model)
30001 3797433 3425989 15:59 00:00:00 /nix/store/iizfzmqjdfvbyx82gqw92glb5zl6m4my-python3-3.12.12/bin/python3.12 -m pytest --ignore-glob=keras/src/layers/preprocessing/data_layer_test.py --ignore-glob=keras/src/layers/preprocessing/image_preprocessing/resizing_test.py --ignore-glob=keras/src/layers/preprocessing/rescaling_test.py --ignore-glob=keras/src/trainers/data_adapters/grain_dataset_adapter_test.py --ignore-glob=keras/src/export/onnx_test.py --ignore-glob=integration_tests/dataset_tests --ignore-glob=keras/src/applications/applications_test.py --ignore-glob=integration_tests/jax_custom_fit_test.py --ignore-glob=integration_tests/tf_distribute_training_test.py --ignore-glob=integration_tests/torch_custom_fit_test.py --ignore-glob=integration_tests/torch_workflow_test.py --ignore-glob=keras/src/backend/tensorflow/saved_model_test.py -k not (test_fit_with_data_adapter_grain_dataloader) and not (test_fit_with_data_adapter_grain_datast) and not (test_fit_with_data_adapter_grain_datast_with_len) and not (test_image_dataset_from_directory_binary_grain) and not (test_image_dataset_from_directory_color_modes_grain) and not (test_image_dataset_from_directory_crop_to_aspect_ratio_grain) and not (test_image_dataset_from_directory_follow_links_grain) and not (test_image_dataset_from_directory_manual_labels_grain) and not (test_image_dataset_from_directory_multiclass_grain) and not (test_image_dataset_from_directory_no_labels_grain) and not (test_image_dataset_from_directory_not_batched_grain) and not (test_image_dataset_from_directory_pad_to_aspect_ratio_grain) and not (test_image_dataset_from_directory_shuffle_grain) and not (test_image_dataset_from_directory_validation_split_grain) and not (test_sample_count_grain) and not (test_text_dataset_from_directory_binary_grain) and not (test_text_dataset_from_directory_follow_links_grain) and not (test_text_dataset_from_directory_manual_labels_grain) and not (test_text_dataset_from_directory_multiclass_grain) and not (test_text_dataset_from_directory_not_batched_grain) and not (test_text_dataset_from_directory_standalone_grain) and not (test_text_dataset_from_directory_validation_split_grain) and not (test_keras_imports) and not (test_reloading_default_saved_model)
16:42:06
@arilotter:matrix.orgAri Lotterah lol still happens on 3.13.0 - problem is that it's not binary cached and i think just hangs forever on this system (that weird intel server i have) 😅17:48:54
@arilotter:matrix.orgAri Lotter does nixpkgs have a mechanism for marking a package's checks as broken on a specific CPU generation? 18:01:03
@connorbaker:matrix.orgconnor (burnt/out) (UTC-8)Nix barely knows the platform; it doesn’t know what generation the hardware is (or at least doesn’t make it available)19:42:59
@apyh:matrix.orgapyhyeah, makes sense. 's frustrating, means i have to maintain some out of tree list of overrides for checks to disable for specific packages 20:06:37
@sporeray:matrix.orgRobbie BuxtonHave you tried nixbuild.net?20:11:49
@sporeray:matrix.orgRobbie Buxton I use it and it’s fantastic 20:11:58
@sporeray:matrix.orgRobbie Buxton Never had an issue building anything there 20:12:43
@connorbaker:matrix.orgconnor (burnt/out) (UTC-8)There’s some support for architecture feature sets but you’ll have to rebuild everything20:30:36
@apyh:matrix.orgapyhyeah, i use garnix.io for CI stuff, this is more frustrating that i have a ton of compute available to me & i have to reserve a gpu node anyways to run the stuff I'm testing, but I can't build it on the same box :p20:47:22
@apyh:matrix.orgapyhprobably something to report w the upstream package though :)20:47:33
4 Aug 2022
@winterqt:nixos.devWinter (she/her) joined the room.03:26:42

Show newer messages


Back to Room ListRoom Version: 9