# Nix Flake \`devShells\` CUDAExecutionProvider (NVIDIA) error with Python \`onnxruntime\`

**URL:** https://discourse.nixos.org/t/nix-flake-devshells-cudaexecutionprovider-nvidia-error-with-python-onnxruntime/75512
**Category:** Help
**Created:** [February 21, 2026, 2:20am UTC](https://discourse.nixos.org/t/nix-flake-devshells-cudaexecutionprovider-nvidia-error-with-python-onnxruntime/75512 "2026-02-21T02:20:00Z")
**Posts on this page:** 2
**Page:** 1

<div class="post-metadata">

### Author: ![S0AndS0](https://discourse.nixos.org/user_avatar/discourse.nixos.org/s0ands0/32/34008_2.png) [@S0AndS0](https://discourse.nixos.org/u/S0AndS0)
#### Post date: [February 21, 2026, 2:20am UTC](https://discourse.nixos.org/t/nix-flake-devshells-cudaexecutionprovider-nvidia-error-with-python-onnxruntime/75512/1 "2026-02-21T02:20:01Z")

</div>

Howdy y’all! I’m trying to, eventually, package the [Deep-Live-Cam](https://github.com/hacksider/Deep-Live-Cam/tree/07b4d66965262654afac59ae2fbefd291a2c4d7d) Python application for NixOS and am having the skill issues with NVIDIA/Cuda

To be totally clear it works, very slowly, without GPU and main aim is to make it work with GPU… ideally faster too

## Configuration files

> **Relevant system;
/etc/nixos/boot-stuff.nix**
>
> ```nix
> {
> config,
> lib,
> modulesPath,
> ...
> }:
> 
> {
> ## From: NixOS install via Virt-Manager
> imports = [
> (modulesPath + "/profiles/qemu-guest.nix")
> ];
> 
> ## Enable non-free drivers to make WiFi mostly work
> hardware.enableRedistributableFirmware = true;
> 
> ## Maybe fix WiFi not showing 5ghz networks
> hardware.enableAllFirmware = true;
> 
> ## Enable OpenGL
> hardware.graphics.enable = true;
> 
> ## Load nvidia driver for Xorg and Wayland
> services.xserver.videoDrivers = ["nvidia"];
> 
> boot.initrd.availableKernelModules = [
> "ahci"
> "rtsx_pci_sdmmc"
> "sd_mod"
> "sr_mod"
> "usb_storage"
> "virtio_pci"
> "virtio_scsi"
> "xhci_pci"
> ];
> 
> ## From: `nixos-generate-config`
> hardware.cpu.intel.updateMicrocode = lib.mkDefault config.hardware.enableRedistributableFirmware;
> 
> ## From: NixOS install via Virt-Manager
> ## https://wiki.nixos.org/wiki/PCI_passthrough
> boot.initrd.kernelModules = [
> "vfio"
> "vfio_pci"
> "vfio_iommu_type1"
> ];
> boot.kernelModules = ["kvm-intel"];
> 
> # Enables DHCP on each ethernet and wireless interface. In case of scripted networking
> # (the default) this is the recommended approach. When using systemd-networkd it's
> # still possible to use this option, but it's recommended to use it in conjunction
> # with explicit per-interface declarations with `networking.interfaces.<interface>.useDHCP`.
> networking.useDHCP = lib.mkDefault true;
> 
> nixpkgs.hostPlatform = lib.mkDefault "x86_64-linux";
> 
> # https://wiki.nixos.org/wiki/CUDA#Setting_up_CUDA_Binary_Cache
> nix.settings = {
> substituters = [
> "https://cache.nixos-cuda.org"
> ];
> trusted-public-keys = [
> "cache.nixos-cuda.org:74DUi4Ye579gUqzH4ziL9IyiJBlDpMRn9MBN8oNan9M="
> ];
> };
> }
> 
> ```

> **Partially functional;
~/git/hub/hacksider/Deep-Live-Cam/flake.nix**
>
> ````nix
> {
> description = "Nix Flake for installing and running project in reproducible fashion";
> 
> inputs = {
> # nixpkgs.url = "github:nixos/nixpkgs?ref=nixos-unstable";
> };
> 
> outputs =
> { ... }:
> let
> system = "x86_64-linux";
> 
> nixpkgs = import <nixpkgs> {
> inherit system;
> config.allowUnfree = true;
> 
> ## WARN: enabling following without cache causes magma to rebuild opencv and other things :-|
> config.cudaSupport = true;
> config.allowUnfreePredicate =
> p:
> builtins.all (
> license:
> license.free
> || builtins.elem license.shortName [
> "CUDA EULA"
> "cuDNN EULA"
> "cuTENSOR EULA"
> "NVidia OptiX EULA"
> ]
> ) (if builtins.isList p.meta.license then p.meta.license else [p.meta.license]);
> };
> 
> python3Packages = nixpkgs.pkgs.python3Packages;
> python3Exec = nixpkgs.lib.getExe nixpkgs.pkgs.python3;
> 
> opennsfw2 = python3Packages.buildPythonPackage rec {
> pname = "opennsfw2";
> version = "0.10.2";
> src = python3Packages.fetchPypi {
> inherit pname version;
> hash = "sha256-xs6gcy3A8Y52YWXAg0JXechMpqAfEWm/pdDUqgUxHk8=";
> };
> doCheck = false;
> pyproject = true;
> build-system =
> with python3Packages;
> [
> wheel
> setuptools
> gdown
> matplotlib
> numpy
> opencv-python
> pillow
> scikit-image
> tensorflow
> tqdm
> ]
> ++ (with nixpkgs.pkgs; [
> ffmpeg
> ]);
> };
> 
> buildInputsDefault =
> with python3Packages;
> [
> numpy
> typing-extensions
> opencv4Full
> cv2-enumerate-cameras
> onnx
> onnxruntime
> insightface
> psutil
> tkinter
> customtkinter
> pillow
> opennsfw2
> protobuf
> torchvision
> ]
> ++ (nixpkgs.lib.optionals (system != "darwin") (
> with python3Packages;
> [
> tensorflow
> ]
> ))
> ++ (nixpkgs.lib.optionals (system == "darwin") (
> with python3Packages;
> [
> torch
> ]
> ));
> in
> {
> devShells.${system} = {
> /**
> Run with specified execution provider
> */
> run-with = {
> /**
> ## Slow but works
> 
> ```bash
> nix develop --impure .#run-with.cpu
> ```
> */
> cpu = nixpkgs.pkgs.mkShell {
> buildInputs =
> buildInputsDefault
> ++ (with python3Packages; [
> torch
> ]);
> 
> shellHook = ''
> ${python3Exec} run.py;
> exit;
> '';
> };
> 
> /**
> ## Nvidia or Cuda is broken here
> 
> ```bash
> nix develop --impure .#run-with.cuda
> ```
> */
> cuda = nixpkgs.pkgs.mkShell {
> buildInputs = buildInputsDefault;
> shellHook = ''
> ${python3Exec} run.py --execution-provider cuda;
> exit;
> '';
> };
> 
> /**
> ## Nvidia or Cuda is broken here too
> 
> ```bash
> nix develop --impure .#run-with.cuda-fhs
> ```
> */
> cuda-fhs =
> let
> pythonWithTkinter = nixpkgs.pkgs.python3.withPackages (ps: with ps; [
> numpy
> typing-extensions
> cv2-enumerate-cameras # cv2_enumerate_cameras==1.1.15
> onnx
> onnxruntime
> insightface
> psutil
> tkinter # tk==0.1.0
> customtkinter
> pillow
> opennsfw2
> protobuf
> torchvision
> torchWithCuda
> tensorflow
> ]);
> in
> (nixpkgs.pkgs.buildFHSEnv {
> name = "cuda-fhs";
> 
> profile = ''
> export LD_LIBRARY_PATH="${nixpkgs.pkgs.linuxPackages.nvidia_x11}/lib";
> export CUDA_PATH="${nixpkgs.pkgs.cudatoolkit}";
> export EXTRA_LDFLAGS="-L/lib -L${nixpkgs.pkgs.linuxPackages.nvidia_x11}/lib";
> export EXTRA_CCFLAGS="-I/usr/include";
> export PYTHONPATH="${pythonWithTkinter}/lib/python3.13/site-packages:$PYTHONPATH";
> '';
> 
> runScript = ''
> ${pythonWithTkinter}/bin/python3 run.py --execution-provider cuda;
> exit;
> '';
> }).env;
> };
> };
> };
> }
> 
> ````

> **Stack trace errors for;
\`\`\`bash
nix develop --impure .#run-with.cuda

\`\`\`**
>
> ```auto
> [ERROR:0@10.429] global obsensor_uvc_stream_channel.cpp:163 getStreamChannelGroup Camera index out of range
> [WARN:0@22.180] global cap_gstreamer.cpp:1173 isPipelinePlaying OpenCV | GStreamer warning: GStreamer: pipeline have not been created
> Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}, 'CUDAExecutionProvider': {'sdpa_kernel': '0', 'use_tf32': '1', 'fuse_conv_bias': '0', 'prefer_nhwc': '0', 'tunable_op_max_tuning_duration_ms': '0', 'enable_skip_layer_norm_strict_mode': '0', 'tunable_op_tuning_enable': '0', 'tunable_op_enable': '0', 'use_ep_level_unified_stream': '0', 'device_id': '0', 'has_user_compute_stream': '0', 'gpu_external_empty_cache': '0', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'cudnn_conv1d_pad_to_nc1d': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_alloc': '0', 'gpu_external_free': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'do_copy_in_default_stream': '1', 'enable_cuda_graph': '0', 'user_compute_stream': '0', 'cudnn_conv_use_max_workspace': '1'}}
> model ignore: /home/s0ands0/.insightface/models/buffalo_l/1k3d68.onnx landmark_3d_68
> Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}, 'CUDAExecutionProvider': {'sdpa_kernel': '0', 'use_tf32': '1', 'fuse_conv_bias': '0', 'prefer_nhwc': '0', 'tunable_op_max_tuning_duration_ms': '0', 'enable_skip_layer_norm_strict_mode': '0', 'tunable_op_tuning_enable': '0', 'tunable_op_enable': '0', 'use_ep_level_unified_stream': '0', 'device_id': '0', 'has_user_compute_stream': '0', 'gpu_external_empty_cache': '0', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'cudnn_conv1d_pad_to_nc1d': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_alloc': '0', 'gpu_external_free': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'do_copy_in_default_stream': '1', 'enable_cuda_graph': '0', 'user_compute_stream': '0', 'cudnn_conv_use_max_workspace': '1'}}
> model ignore: /home/s0ands0/.insightface/models/buffalo_l/2d106det.onnx landmark_2d_106
> Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}, 'CUDAExecutionProvider': {'sdpa_kernel': '0', 'use_tf32': '1', 'fuse_conv_bias': '0', 'prefer_nhwc': '0', 'tunable_op_max_tuning_duration_ms': '0', 'enable_skip_layer_norm_strict_mode': '0', 'tunable_op_tuning_enable': '0', 'tunable_op_enable': '0', 'use_ep_level_unified_stream': '0', 'device_id': '0', 'has_user_compute_stream': '0', 'gpu_external_empty_cache': '0', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'cudnn_conv1d_pad_to_nc1d': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_alloc': '0', 'gpu_external_free': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'do_copy_in_default_stream': '1', 'enable_cuda_graph': '0', 'user_compute_stream': '0', 'cudnn_conv_use_max_workspace': '1'}}
> find model: /home/s0ands0/.insightface/models/buffalo_l/det_10g.onnx detection [1, 3, '?', '?'] 127.5 128.0
> Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}, 'CUDAExecutionProvider': {'sdpa_kernel': '0', 'use_tf32': '1', 'fuse_conv_bias': '0', 'prefer_nhwc': '0', 'tunable_op_max_tuning_duration_ms': '0', 'enable_skip_layer_norm_strict_mode': '0', 'tunable_op_tuning_enable': '0', 'tunable_op_enable': '0', 'use_ep_level_unified_stream': '0', 'device_id': '0', 'has_user_compute_stream': '0', 'gpu_external_empty_cache': '0', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'cudnn_conv1d_pad_to_nc1d': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_alloc': '0', 'gpu_external_free': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'do_copy_in_default_stream': '1', 'enable_cuda_graph': '0', 'user_compute_stream': '0', 'cudnn_conv_use_max_workspace': '1'}}
> model ignore: /home/s0ands0/.insightface/models/buffalo_l/genderage.onnx genderage
> Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}, 'CUDAExecutionProvider': {'sdpa_kernel': '0', 'use_tf32': '1', 'fuse_conv_bias': '0', 'prefer_nhwc': '0', 'tunable_op_max_tuning_duration_ms': '0', 'enable_skip_layer_norm_strict_mode': '0', 'tunable_op_tuning_enable': '0', 'tunable_op_enable': '0', 'use_ep_level_unified_stream': '0', 'device_id': '0', 'has_user_compute_stream': '0', 'gpu_external_empty_cache': '0', 'cudnn_conv_algo_search': 'EXHAUSTIVE', 'cudnn_conv1d_pad_to_nc1d': '0', 'gpu_mem_limit': '18446744073709551615', 'gpu_external_alloc': '0', 'gpu_external_free': '0', 'arena_extend_strategy': 'kNextPowerOfTwo', 'do_copy_in_default_stream': '1', 'enable_cuda_graph': '0', 'user_compute_stream': '0', 'cudnn_conv_use_max_workspace': '1'}}
> find model: /home/s0ands0/.insightface/models/buffalo_l/w600k_r50.onnx recognition ['None', 3, 112, 112] 127.5 127.5
> set det-size: (320, 320)
> 2026-02-18 21:25:24.309620807 [E:onnxruntime:Default, cudnn_fe_call.cc:33 CudaErrString<cudnn_frontend::error_object>] execute(handle, plan->get_raw_desc(), variant_pack_descriptor.get_ptr()) failed with message: func(handle, stream, m, n, k, static_cast<const T_IN*>(d_B), ldb, static_cast<T_OUT*>(d_C), static_cast<const T_IN*>(d_A), parms, texB, texB_offset, alphaVal, betaVal, doBiasAct, static_cast<const T_OUT*>(zData), static_cast<const T_OUT*>(bias), doRelu, propNan, dilation_h, dilation_w, outputType, launch_params), and code: CUDNN_STATUS_EXECUTION_FAILED
> 2026-02-18 21:25:24.309686510 [E:onnxruntime:Default, cudnn_fe_call.cc:93 CudaCall] CUDNN_FE failure 11: CUDNN_BACKEND_API_FAILED ; GPU=0 ; hostname=nixos ; file=/build/source/onnxruntime/core/providers/cuda/nn/conv.cc ; line=483 ; expr=s_.cudnn_fe_graph->execute(cudnn_handle, s_.variant_pack, ws.get()); 
> 2026-02-18 21:25:24.309706058 [E:onnxruntime:, sequential_executor.cc:572 ExecuteKernel] Non-zero status code returned while running Conv node. Name:'Conv_0' Status Message: CUDNN_FE failure 11: CUDNN_BACKEND_API_FAILED ; GPU=0 ; hostname=nixos ; file=/build/source/onnxruntime/core/providers/cuda/nn/conv.cc ; line=483 ; expr=s_.cudnn_fe_graph->execute(cudnn_handle, s_.variant_pack, ws.get()); 
> Exception in thread Thread-2 (_processing_thread_func):
> Traceback (most recent call last):
> File "/nix/store/slhpx9glq7vl99bwi93bgrhn3syv98s1-python3-3.13.11/lib/python3.13/threading.py", line 1044, in _bootstrap_inner
> self.run()
> ~~~~~~~~ ^^
> File "/nix/store/slhpx9glq7vl99bwi93bgrhn3syv98s1-python3-3.13.11/lib/python3.13/threading.py", line 995, in run
> self._target(*self._args, **self._kwargs)
> ~~~~~~~~~~~~ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> File "/home/s0ands0/git/hub/hacksider/Deep-Live-Cam/wt/nix-flake/modules/ui.py", line 1015, in _processing_thread_func
> source_image = get_one_face(cv2.imread(modules.globals.source_path))
> File "/home/s0ands0/git/hub/hacksider/Deep-Live-Cam/wt/nix-flake/modules/face_analyser.py", line 38, in get_one_face
> face = get_face_analyser().get(frame)
> File "/nix/store/z1vg2l4397y5jj0ljyac9qsqml8q15nv-python3.13-insightface-0.7.3/lib/python3.13/site-packages/insightface/app/face_analysis.py", line 59, in get
> bboxes, kpss = self.det_model.detect(img,
> ~~~~~~~~~~~~~~~~~~~~~ ^^^^^
> max_num=max_num,
> ^^^^^^^^^^^^^^^^
> metric='default')
> ^^^^^^^^^^^^^^^^^
> File "/nix/store/z1vg2l4397y5jj0ljyac9qsqml8q15nv-python3.13-insightface-0.7.3/lib/python3.13/site-packages/insightface/model_zoo/retinaface.py", line 224, in detect
> scores_list, bboxes_list, kpss_list = self.forward(det_img, self.det_thresh)
> ~~~~~~~~~~~~ ^^^^^^^^^^^^^^^^^^^^^^^^^^
> File "/nix/store/z1vg2l4397y5jj0ljyac9qsqml8q15nv-python3.13-insightface-0.7.3/lib/python3.13/site-packages/insightface/model_zoo/retinaface.py", line 152, in forward
> net_outs = self.session.run(self.output_names, {self.input_name : blob})
> File "/nix/store/phxz157dmv9rxqv8bcmwq2k6gxnibi35-python3.13-onnxruntime-1.23.2/lib/python3.13/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 287, in run
> return self._sess.run(output_names, input_feed, run_options)
> ~~~~~~~~~~~~~~ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running Conv node. Name:'Conv_0' Status Message: CUDNN_FE failure 11: CUDNN_BACKEND_API_FAILED ; GPU=0 ; hostname=nixos ; file=/build/source/onnxruntime/core/providers/cuda/nn/conv.cc ; line=483 ; expr=s_.cudnn_fe_graph->execute(cudnn_handle, s_.variant_pack, ws.get()); 
> 
> ```

> **Logs of CPU doing fine via;
\`\`\`bash
nix develop --impure .#run-with.cpu

\`\`\`**
>
> ```auto
> [ERROR:0@10.370] global obsensor_uvc_stream_channel.cpp:163 getStreamChannelGroup Camera index out of range
> [WARN:0@24.603] global cap_gstreamer.cpp:1173 isPipelinePlaying OpenCV | GStreamer warning: GStreamer: pipeline have not been created
> Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}}
> model ignore: /home/s0ands0/.insightface/models/buffalo_l/1k3d68.onnx landmark_3d_68
> Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}}
> model ignore: /home/s0ands0/.insightface/models/buffalo_l/2d106det.onnx landmark_2d_106
> Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}}
> find model: /home/s0ands0/.insightface/models/buffalo_l/det_10g.onnx detection [1, 3, '?', '?'] 127.5 128.0
> Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}}
> model ignore: /home/s0ands0/.insightface/models/buffalo_l/genderage.onnx genderage
> Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}}
> find model: /home/s0ands0/.insightface/models/buffalo_l/w600k_r50.onnx recognition ['None', 3, 112, 112] 127.5 127.5
> set det-size: (320, 320)
> [DLC.FACE-SWAPPER] Loading face swapper model from: /home/s0ands0/git/hub/hacksider/Deep-Live-Cam/wt/nix-flake/models/inswapper_128.onnx
> Applied providers: ['CPUExecutionProvider'], with options: {'CPUExecutionProvider': {}}
> inswapper-shape: [1, 3, 128, 128]
> [DLC.FACE-SWAPPER] Face swapper model loaded successfully.
> 
> ```

## Kernel Module and DMesg logs and device metadata

> **Module looks to be loaded;
\`\`\`bash
lsmod | grep nvidia

\`\`\`**
>
> ```auto
> nvidia_drm 143360 0
> nvidia_modeset 1933312 1 nvidia_drm
> nvidia_uvm 3858432 0
> nvidia 111611904 3 nvidia_uvm,nvidia_drm,nvidia_modeset
> drm_ttm_helper 20480 1 nvidia_drm
> video 81920 4 dell_wmi,dell_laptop,i915,nvidia_modeset
> 
> ```

> **And \`dmesg\` doesn't show too many concerning things concerning GPU;
\`\`\`bash
sudo dmesg | grep -iE 'gpu|graphics|nvidia'

\`\`\`**
>
> ``` [0.000000] Command line: initrd=\EFI\nixos\m3hmc5rvpdjp37m051dhypsmz30clh4a-initrd-linux-6.18.8-initrd.efi init=/nix/store/7x2bsmz2mms74vrfcnqvdgh90k1zc08a-nixos-system-nixos-26.05.20260204.00c21e4/init i915.enable_guc=2 fsck.mode=force fsck.repair=preen intel_iommu=on iommu=pt loglevel=4 lsm=landlock,yama,bpf nvidia-drm.modeset=1 nvidia-drm.fbdev=1 [0.059658] Reserving Intel graphics memory at [mem 0x79000000-0x7cffffff] [0.067512] Kernel command line: initrd=\EFI\nixos\m3hmc5rvpdjp37m051dhypsmz30clh4a-initrd-linux-6.18.8-initrd.efi init=/nix/store/7x2bsmz2mms74vrfcnqvdgh90k1zc08a-nixos-system-nixos-26.05.20260204.00c21e4/init i915.enable_guc=2 fsck.mode=force fsck.repair=preen intel_iommu=on iommu=pt loglevel=4 lsm=landlock,yama,bpf nvidia-drm.modeset=1 nvidia-drm.fbdev=1 [1.565082] stage-1-init: [Fri Feb 20 16:50:07 UTC 2026] loading module virtio_gpu... [13.287757] RAPL PMU: hw unit of domain pp1-gpu 2^-14 Joules [13.444790] nvidia: module license 'NVIDIA' taints kernel. [13.444803] nvidia: module license taints kernel. [14.075989] nvidia-nvlink: Nvlink Core is being initialized, major device number 511 [14.085347] nvidia 0000:02:00.0: enabling device (0006 -\> 0007) [14.312659] NVRM: loading NVIDIA UNIX x86_64 Kernel Module 580.126.09 Wed Jan 7 22:59:56 UTC 2026 [14.546130] nvidia_uvm: module uses symbols nvUvmInterfaceDisableAccessCntr from proprietary module nvidia, inheriting taint. [14.803094] nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms 580.126.09 Wed Jan 7 22:32:52 UTC 2026 [14.821338] [drm] [nvidia-drm] [GPU ID 0x00000200] Loading driver [15.051255] [drm] Initialized nvidia-drm 0.0.0 for 0000:02:00.0 on minor 0 [15.051273] nvidia 0000:02:00.0: [drm] No compatible format found [15.051276] nvidia 0000:02:00.0: [drm] Cannot find any crtc or sizes ```

> **Output of;
\`\`\`bash
nix-info -m

\`\`\`**
>
> ```auto
> - system: `"x86_64-linux"`
> - host os: `Linux 6.18.8, NixOS, 26.05 (Yarara), 26.05.20260204.00c21e4`
> - multi-user?: `yes`
> - sandbox: `yes`
> - version: `nix-env (Nix) 2.31.3`
> - nixpkgs: `/nix/store/ih9vmk2a3mrk6vhmibqzji6kjc6parzp-source`
> 
> ```

## Attributions

- [GitHub - hacksider/Deep-Live-Cam: real time face swap and one-click video deepfake with only a single image · GitHub](https://github.com/hacksider/Deep-Live-Cam)
- [Python - Official NixOS Wiki](https://wiki.nixos.org/wiki/Python)
- [opennsfw2 · PyPI](https://pypi.org/project/opennsfw2/)
- [Allow unfree in flakes](https://discourse.nixos.org/t/allow-unfree-in-flakes/29904)
- [Packaging/Python - Official NixOS Wiki](https://wiki.nixos.org/wiki/Packaging/Python)
- [Pytorch with cuda support](https://discourse.nixos.org/t/pytorch-with-cuda-support/51189)
- [CUDA Cache for Nix Community](https://discourse.nixos.org/t/cuda-cache-for-nix-community/56038)
- [https://unix.stackexchange.com/questions/529047/is-there-a-way-to-have-hibernate-and-encrypted-swap-on-nixos](https://unix.stackexchange.com/questions/529047/is-there-a-way-to-have-hibernate-and-encrypted-swap-on-nixos)
- [OpenCV - Official NixOS Wiki](https://wiki.nixos.org/wiki/OpenCV)
- [How to give opencv dependency to python package](https://discourse.nixos.org/t/how-to-give-opencv-dependency-to-python-package/16949)
- [opencv-python · PyPI](https://pypi.org/project/opencv-python/)
- [OpenCV with Cuda in nix-shell](https://discourse.nixos.org/t/opencv-with-cuda-in-nix-shell/7358)
- [Massive rebuild of packages that use CUDA despite cuda-maintainers and nix-community caches](https://discourse.nixos.org/t/massive-rebuild-of-packages-that-use-cuda-despite-cuda-maintainers-and-nix-community-caches/72394)
- [CUDA - Official NixOS Wiki](https://wiki.nixos.org/wiki/CUDA#Setting_up_CUDA_Binary_Cache)
- [CUDA setup on NixOS](https://discourse.nixos.org/t/cuda-setup-on-nixos/1118)
- [CUDA working with poetry2nix - #8 by DavidRConnell](https://discourse.nixos.org/t/cuda-working-with-poetry2nix/58208/8)
- [PoC of JAX, PyQt, numpy, tensorboard etc. in nix · GitHub](https://gist.github.com/erolm-a/01586f294f3bae45020c85002652401f)
- [Painlessly setting up ML tooling on NixOS - lavafroth](https://lavafroth.is-a.dev/post/cuda-on-nixos-without-sacrificing-ones-sanity/)
- [https://github.com/NixOS/nixpkgs/issues/238990](https://github.com/NixOS/nixpkgs/issues/238990)

---

<div class="post-metadata">

### Author: ![S0AndS0](https://discourse.nixos.org/user_avatar/discourse.nixos.org/s0ands0/32/34008_2.png) [@S0AndS0](https://discourse.nixos.org/u/S0AndS0)
#### Post date: [March 1, 2026, 1:57am UTC](https://discourse.nixos.org/t/nix-flake-devshells-cudaexecutionprovider-nvidia-error-with-python-onnxruntime/75512/2 "2026-03-01T01:57:24Z")

</div>

For any that happen to stumble across this set of software sins, here be a starter example for building-up Cuda dependent packages

```nix
let
  pkgs = (import <nixpkgs> { config.cudaSupport = true; }).pkgs;
  python3Packages = pkgs.python312Packages;
in
{
  fetchFromGitHub ? pkgs.fetchFromGitHub,
  buildPythonPackage ? python3Packages.buildPythonPackage,
  torch ? python3Packages.torch,
  torchWithCuda ? python3Packages.torchWithCuda,
  config ? { cudaSupport = true; },
  lib ? pkgs.lib,
}:

buildPythonPackage (finalAttrs: {
  version = "4.19.68.1336";
  pname = "python_importable_name_here";

  dependencies = [
    # Common dependencies here
  ]
  ++ lib.lists.optionals (!config.cudaSupport) [
    torch
  ]
  ++ lib.lists.optionals config.cudaSupport [
    torchWithCuda
  ];

  # Usual attributes to follow...
}

```

… After testing the `let`/`in` can get dropped as well as defaulting via `?` stuff

I’m still sorting out all the many dependencies for the OP but maybe, one day _soon™_, there’ll be an associated PR 😉