Is PyTorch 2.11.0 compatible with CUDA 13.2?

Geremia · April 4, 2026, 1:25am

I get this warning:

DEBUG W0401 21:33:19.460000 8000 site-packages/torch/utils/cpp_extension.py:556] There are no g++ version bounds defined for CUDA version 13.2

nvcc --version: cuda_13.2.r13.2/compiler.37434383_0
g++ --version: g++ (GCC) 15.2.0
python -c "import torch; print(torch.__version__): 2.11.0a0

ptrblck · April 4, 2026, 1:41pm

I don’t know which exact commit you are trying to build as it seems you are using a pre-2.11.0 commit. In any case, nightly binaries are already available for 13.2. I haven’t checked if the tagged 2.11.0 branch is fully compatible as it was cut before the CUDA release. To be safe you could either use the nightly builds or a current commit.

amjames · April 9, 2026, 8:49pm

We introduced CUDA 13.2 as experimental after the 2.11 release, so it is currently in nightlies and will be in the 2.12 release. See the dev mailing list post for more info!