When I use pytorch, the pinned tensors it creates run normally. However, pinning tensors manually with torch.cuda.cudart().cudaHostRegister fails and I can’t access the allocations from the GPU on WSL in Windows 11.
Pinned system memory (example: System memory that an application makes resident for GPU accesses) availability for applications is limited.
For example, some deep learning training workloads, depending on the framework, model and dataset size used, can exceed this limit and may not work.
With a bit of web searching, you can find other possibly related reports/observations.
cudaHostAlloc works though, it is only the cudaHostRegister that doesn’t seem to be supported on WSL. Is there any fundamental limitation that limits supporting cudaHostRegister on WSL?