RuntimeError: Cuda error: 209[cudaFuncGetAttributes(&attr, (void*)fineRasterKernel);]

[Joceima]

Hello,
(It's my first time writing an issue, maybe i didn't express well my problem. Please be patient ! )

I install and build nvdiffrast (no error)

build_log.txt

from source in a Python 3.12 environment with CUDA 12.8 and PyTorch 2.9.1+cu128. But when I tried to run sample script from

nvdiffrast/samples/torch/triangle.py

for example, the execution failed.

Here the output :

Traceback (most recent call last):
  File "/home/mao/nvdiffrast/samples/torch/triangle.py", line 20, in <module>
    glctx = dr.RasterizeCudaContext()
            ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/mao/nvdiffrast/.venv/lib/python3.12/site-packages/nvdiffrast/torch/ops.py", line 67, in __init__
    self.cpp_wrapper = _nvdiffrast_c.RasterizeCRStateWrapper(cuda_device_idx)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Cuda error: 209[cudaFuncGetAttributes(&attr, (void*)fineRasterKernel);] 

I tried to replace dr.RasterizeCudaContext()by dr.RasterizeGLConstext(), but I ended up with a similar error :

$ python3 triangle.py 
/home/mao/nvdiffrast/samples/torch/triangle.py:17: DeprecationWarning: RasterizeGLContext has been deprecated and uses RasterizeCudaContext internally
  glctx = dr.RasterizeGLContext()
Traceback (most recent call last):
  File "/home/mao/nvdiffrast/samples/torch/triangle.py", line 17, in <module>
    glctx = dr.RasterizeGLContext()
            ^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/mao/nvdiffrast/.venv/lib/python3.12/site-packages/nvdiffrast/torch/ops.py", line 553, in __init__
    super().__init__(device=device)
  File "/home/mao/nvdiffrast/.venv/lib/python3.12/site-packages/nvdiffrast/torch/ops.py", line 67, in __init__
    self.cpp_wrapper = _nvdiffrast_c.RasterizeCRStateWrapper(cuda_device_idx)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Cuda error: 209[cudaFuncGetAttributes(&attr, (void*)fineRasterKernel);]

Here some more additional informations about what I did for the installation :

1. I followed the steps installation from "Custom install on Linux" https://nvlabs.github.io/nvdiffrast/#installation

2. (maybe useless info in that specific case) When I tried to this command pip install setuptools wheel ninja
   it failed :

$pip install setuptools wheel ninja
error: externally-managed-environment

× This environment is externally managed
╰─> To install Python packages system-wide, try apt install
    python3-xyz, where xyz is the package you are trying to
    install.
    
    If you wish to install a non-Debian-packaged Python package,
    create a virtual environment using python3 -m venv path/to/venv.
    Then use path/to/venv/bin/python and path/to/venv/bin/pip. Make
    sure you have python3-full installed.
    
    If you wish to install a non-Debian packaged Python application,
    it may be easiest to use pipx install xyz, which will manage a
    virtual environment for you. Make sure you have pipx installed.
    
    See /usr/share/doc/python3.12/README.venv for more information.

note: If you believe this is a mistake, please contact your Python installation or OS distribution provider. You can override this, at the risk of breaking your Python installation or OS, by passing --break-system-packages.
hint: See PEP 668 for the detailed specification.

so I created virtual environment, with the following steps :
$python3 -m venv .venv
$source .venv/bin/activate

Then i was been able to follow the rest of nvdiffrast' steps. At the end with pip install . --no-build-isolation I encounter no error like I said in the beginning of my post. I don't think the virtual environment plays a role in this problem.

Infos about versions and materials :

1. Driver version

$nvidia-smi
Tue May 12 15:21:14 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.58.03              Driver Version: 595.58.03      CUDA Version: 13.2     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA RTX PRO 1000 Blac...    Off |   00000000:01:00.0 Off |                  N/A |
| N/A   39C    P4              7W /   35W |      14MiB /   8151MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1208      G   /usr/lib/xorg/Xorg                        4MiB |
+-----------------------------------------------------------------------------------------+

2. nvcc version

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Wed_Jan_15_19:20:09_PST_2025
Cuda compilation tools, release 12.8, V12.8.61
Build cuda_12.8.r12.8/compiler.35404655_0

3. Python version

Python: 3.12.3
PyTorch: 2.9.1+cu128
CUDA interne à Torch: 12.8
CUDNN: 91002

I tried to look up the other issues and I asked to AI (Gemini and Perplexity), but I'm having difficulty. Thank for your time and help !

[s-laine]

If you followed the custom Linux installation example in the docs, you probably set TORCH_CUDA_ARCH_LIST in a way that does not support the GPU you have installed. This would explain the error code 209 (= cudaErrorNoKernelImageForDevice).

The safest option is to skip the export TORCH_CUDA_ARCH_LIST=... command entirely, so that the installer compiles for the architectures that are detected in the system.

[Joceima]

Hello, I apologize for responding late and thank you for your help. I've been trying to re install nvdiffrast and skipping export TORCH_CUDA_ARCH_LIST=.... But I end up with other type of errors, similar to #170.
I'll try to set up the environment well this time and try skipping export TORCH_CUDA_ARCH_LIST=.... I'll keep this post update, if it works or not.