Wolfram Computation Meets Knowledge

What are the minimum requirements for GPU training of neural networks in the Wolfram Language?

Please note that CUDALink and GPU training of neural networks are separate features with separate requirements. If you are looking for the CUDALink requirements, please consult the appropriate article.

To use NetTrain with a GPU in the Wolfram Language, the NVIDIA GPU in your machine has to be compatible with a supported CUDA Toolkit version, and the NVIDIA driver has to be up to date.

Wolfram Language Version NVIDIA Driver Version Compute Capability
Linux Windows macOS
12.3 460.27 460.82 N/A 3.7
12.2 460.27 460.82 N/A 3.7
12.1 418.39 418.96 418.105 3.7
12.0 410.48 411.31 410.130 3.7
11.3 390.46 391.29 387.128 3.0

GPU training of neural networks is no longer supported on macOS starting with Wolfram Language 12.2, as Apple dropped macOS support of CUDA.

For more compatibility information, please see NVIDIA’s CUDA compatibility documentation.

In case of any inconsistency with their information, please contact NVIDIA.

Is this article helpful?
Yes
No

Any comments?

Thank you for your feedback.

Submit

Contact Support

Whether you have a question about billing, activation or something more technical, we are ready to help you.

1-800-WOLFRAM (+1-217-398-0700 for international callers)

Customer Support

Monday–Friday
8am–5pm US Central Time

  • Product registration or activation
  • Pre-sales information and ordering
  • Help with installation and first launch

Advanced Technical Support (for eligible customers)

Monday–Thursday
8am–7pm US Central Time

Friday
8:30–10am & 11am–5pm US Central Time

  • Priority technical support
  • Product assistance from Wolfram experts
  • Help with Wolfram Language programming
  • Advanced installation support