Hash encoder requires CUDA v11.3 about sax-nerf

[BiTaNannnnn]

When I change the backend.py extra_cuda_cflags=[
'-O1',
'-arch=sm_70',
'-U__CUDA_NO_HALF_OPERATORS__', '-U__CUDA_NO_HALF_CONVERSIONS__', '-U__CUDA_NO_HALF2_OPERATORS__' # undefine flags, necessary!
],
It also has the same error.

If I change the '-arch=sm_70' to '-arch=sm_75/80/85'
It will show this:
"
nvcc fatal : Value 'sm_80' is not defined for option 'gpu-architecture'
ninja: build stopped: subcommand failed.
"

[caiyuanhao1998]

The error you encountered, "nvcc fatal: Unsupported gpu architecture 'compute_86'," indicates that the NVIDIA CUDA Compiler (nvcc) does not support the GPU architecture specified for compilation, which in this case is 'compute_86'. This issue can arise if you're using a CUDA toolkit version that does not support the specific compute capability of your GPU.

Please determine the compute capability of your GPU and check if it aligns with the CUDA version installed

If you find our repo useful, please help us star it. Thanks.

[BiTaNannnnn]

The final solution to this problem is Ninja version = 1.10.0.
I have tried 1.12 and 1.8 before and the problem still exists.

Today I tried version 1.10.0/1.10.2 and it was solved successfully.

[caiyuanhao1998]

Hi, really happy to see that you have fixed the problem. Good luck :)