I’m getting warning that my card which is sm_86 is not supported by nightly build. Are you removing all cards under sm_90 from support in new PyTorch?
No, sm_86 is still supported in all of our nightly binaries and we are not planning to remove it.
1 Like
OK, please look at below message.
Nightly build CUDA 12.6.3 and CUDA 12.8.0 message:
python3 -c “import torch; print(torch.cuda.is_available()); print(torch.cuda.device_count()); print(torch.cuda.get_device_name(0) if torch.cuda.is_available() else ‘No CUDA’)”
True
1
/usr/local/lib/python3.12/dist-packages/torch/cuda/init.py:266: UserWarning:
Found GPU0 NVIDIA RTX A4000 which is of cuda capability 8.6.
PyTorch no longer supports this GPU because it is too old.
The minimum cuda capability supported by this library is 9.0.
warnings.warn(
/usr/local/lib/python3.12/dist-packages/torch/cuda/init.py:291: UserWarning:
NVIDIA RTX A4000 with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_90.
If you want to use the NVIDIA RTX A4000 GPU with PyTorch, please check the instructions at Start Locally | PyTorch
warnings.warn(
NVIDIA RTX A4000
Are you using an ARM system? In this case it’s expected as we support Hopper and Blackwell. x86 binaries support Maxwell to Hopper using CUDA 12.6 and Turing to Blackwell using CUDA 12.8.
Yes, but not NVidia specific HW. I’m on AsrockRack ARM64 Neoverse-N1 motherboard.
Actually that is sad about ARM64 that you have removed GPUs like L40s, RTX 6000, etc.
RTX 6000 Blackwell would be my upgrade choice I believe, but this, I think, it’s expected in Q3 2025.
Pozdrawiam
Łukasz