Will PyTorch run on my GPU? — compute capability to prebuilt wheel checker
Type an NVIDIA GPU model and get its CUDA compute capability plus every PyTorch release whose prebuilt Linux and Windows wheels were actually compiled for that capability, with the pip index URL, the minimum NVIDIA driver, and the quoted line from NVIDIA's and PyTorch's own pages behind each.
Every compute capability, wheel and driver minimum here is quoted from NVIDIA's or PyTorch's own published page, and it is not advice about what to install: their release notes decide what is supported, and a test run on your own machine decides whether it works.
The records this page works from: NVIDIA GPU CUDA compute capability by model (checked 2026-11-15); PyTorch release, CUDA build and NVIDIA driver pairings (checked 2026-11-15). Every record there has its own page, with the source it was read from and the sentence quoted from it.