If you are unsure of what CUDA version of our plugin builds that you should download and run, you’ve come to the right page. Hopefully the chart below will make it more clear what we recommend but also what other options that you have. Good to have with you is a couple of GPU examples and their compute capabilities:
GTX1080: sm_6.1
RTX2080: sm_7.5
RTX3080: sm_8.6
RTX A4000: sm_8.6
RTX4080: sm_8.9
To see a full fledged list of what compute capability your specific GPU is, please visit the CUDA page on Wikipedia and scroll down to the “GPUs supported” section.

Nuke
15.0/15.1
Nuke 14.1Nuke 14.0Nuke 13.xNuke 11.3-12.2cuDNN versionRequire NVIDIA driver versionNative compute capability (sm_) supportLatest native supported architectureWill JIT-compile the kernels/PTX code
CUDA 11.88.4.1520.61+ (Linux)
520.06+ (Windows)
3.5, 5.0, 5.2, 6.0, 6.1, 7.0, 7.5, (8.0*), 8.6, 8.9,(9.0*)
Ada Lovelace, e.g. RTX4080>=sm_9.0
CUDA 11.28.1.0460.27+ (Linux)
460.89+ (Windows)
3.5, 5.0, 5.2, 6.0, 6.1, 7.0, 7.5, 8.0, 8.6Ampere, e.g. RTX3080 or RTX A4000>=sm_8.9
CUDA 11.18.4.1455.32+ (Linux)
456.81+ (Windows)
3.5, 5.0, 5.2, 6.0, 6.1, 7.0, 7.5, 8.0, 8.6Ampere, e.g. RTX3080 or RTX A4000>=sm_8.9
CUDA 10.18.0.5418.87+ (Linux)
426.00+ (Windows)
3.5, 5.0, 5.2, 6.0, 6.1, 7.0, 7.5Turing, e.g. RTX2080>=sm_8.0
* The compute capabilities 8.0 and 9.0 for the CUDA11.8 builds are natively supported with the Nuke11.3-14.0 versions. They are NOT natively supported for the Nuke14.1/15.0 builds since for these versions we are instead linking against the included PyTorch version that comes bundled together with Nuke (which again are not supporting these specific compute capabilities). If you want to run Nuke14.1/15.0 with compute capabilities 8.0 or 9.0, we do have versions compiled with our statically linked PyTorch version instead. Please use the contact form to request a direct download link.
Recommended CUDA version to use since it’s matching exactly what Nuke is built against, hence having full compatibility with Foundry’s own AIR tools (such as the CopyCat node).
A later CUDA version than what Nuke was built against. This works well, and gives you native support for more modern GPUs, but with the drawback of not being fully compatible with Foundry’s AIR tools (such as the CopyCat node) . Please be careful with mixing our plugins with the AIR tools in the same scripts.
Since these Nuke versions were shipped without Foundry’s AIR tools (as they weren’t invented yet), there are no compatibility problems and you can choose CUDA version freely.
This build combination doesn’t exist.