search

NVlabs / nvdiffrast

Nvdiffrast - Modular Primitives for High-Performance Differentiable Rendering
Other
1.31k stars 139 forks source link

RuntimeError: Cuda error: 209[cudaLaunchKernel #75

Closed jiaxiangshang closed 2 years ago

jiaxiangshang commented 2 years ago

Hi, grouper,

My cuda is 10.1 or 11.0 The card is RTX 5000, Pytorch version is 1.6.0

The error I get is out, out_da = _get_plugin().interpolate_fwd(attr, rast, tri) RuntimeError: Cuda error: 209[cudaLaunchKernel(func, gridSize, blockSize, args, 0, stream);]

It seems the conflict of cuda, card, pyt versions, can you give some suggestions?

Best Jiaxiang

s-laine commented 2 years ago

Quadro RTX 5000 is supported by Cuda 10.0 and later, but PyTorch 1.6.0 is too old to support Cuda 11.0 if that's what you have installed. Error 209 indicates that the Cuda compilation did not produce a kernel for the installed GPU architecture, and this compilation is handled by PyTorch, so that's where the problem probably is.

If upgrading PyTorch doesn't solve the problem, please try setting verbose=True in nvdiffrast/torch/ops.py line 84 and paste the output here. You may need to clear the PyTorch extension compilation directory to force a recompile.

jiaxiangshang commented 2 years ago

Hi s-laine,

It is solved by upgrading PyTorch, thank you very much.

Best Jiaxiang