Can't get attribute for CU_DEVICE_ATTRIBUTE_D3D12_CIG_SUPPORTED

[apowers313]

When I call cuDeviceGetAttribute with cuda.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_D3D12_CIG_SUPPORTED I get the error <CUresult.CUDA_ERROR_INVALID_VALUE: 1>.

Seems like it might be an off-by-one error since it's the last value in the Enum CUdevice_attribute?

Here is some code to reproduce the problem.

[leofang]

What's your CUDA Python version and driver version?

• CUDA Python version can be accessed via either python -c "from cuda import __version__; print(__version__)" (but this will change later, see [EPIC] RFC: cuda namespace cleanup #75), pip list | grep cuda if you installed it via pip, or conda list cuda-python if via conda.
• Driver version is reported by nvidia-smi (we just need the "CUDA Version" here)

[apowers313]

CUDA Python: 12.6.0
CUDA Driver: 12.4

[leofang]

This happens because the driver support for CU_DEVICE_ATTRIBUTE_D3D12_CIG_SUPPORTED was added since CUDA 12.5, so to use it you need to update both your CUDA driver and CUDA sw (in this case, cuda-python) to >=12.5.

I am not sure why this is not mentioned in the CUDA Release Notes, but the way I checked this is to search CU_DEVICE_ATTRIBUTE_D3D12_CIG_SUPPORTED in the driver API docs, which are versioned so that you can search the same term by jumping across different minor versions. It takes a few clicks to get to the right URL, but you can see the version number is encoded as part of it, e.g. https://docs.nvidia.com/cuda/archive/12.5.1/cuda-driver-api/index.html. The first time this enumerator appeared was CUDA 12.5.0.

btw what's your use case of CIG? Do you have some graphics rendering pipeline written in Python?

[apowers313]

Use case was iterating all the enums to print out all the device capabilities to the screen.

Thanks for the help.