questions about cuda version

gubei1998 commented 1 year ago

I find that cuda 11.5 does not support pytorch 1.7.1，so I want to know how did you installed pytorch 1.7.1 with cuda 11.5?Thank you very much,and the errors reported are as follows.

$ conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.5 -c pytorch Collecting package metadata (current_repodata.json): done Solving environment: failed with initial frozen solve. Retrying with flexible solve. Collecting package metadata (repodata.json): done Solving environment: failed with initial frozen solve. Retrying with flexible solve.

PackagesNotFoundError: The following packages are not available from current channels:

cudatoolkit=11.5

Current channels:

To search for alternate channels that may provide the conda package you're looking for, navigate to

https://anaconda.org

and use the search bar at the top of the page.

yichen928 commented 1 year ago

Thanks for your interest.

You can use pip to install pytorch and torchvision. For example, "pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html".

Even if this pytorch is for CUDA 11.0, it is also compatible with CUDA 11.X (e.g. 11.3, 11.5, 11.7).

gubei1998 commented 1 year ago

ok,I will try it,thank you very much!

gubei1998 commented 1 year ago

When I install mmdetection3d==0.11.0,it will report an error as follows, 16 errors detected in the compilation of "mmdet3d/ops/voxel/src/scatter_points_cuda.cu". error: command '/usr/local/cuda-11.5/bin/nvcc' failed with exit code 1 error: subprocess-exited-with-error

× python setup.py develop did not run successfully.
│ exit code: 1
╰─> See above for output.

note: This error originates from a subprocess, and is likely not a problem with pip.
And then I search it and find that it is an bug with cuda11.x+pytorch 1.7.x + mmdet3d==0.11.0,so I want to know

Have you encountered this problem and how did you solve it? Thank you for your answer

yichen928 commented 1 year ago

Can you share the complete error message?

yichen928 commented 1 year ago

I would recommend you to install mmdet3d from their source code instead of directly using pip install.

yichen928 commented 1 year ago

Since there is no further reply, I temporally close this issue. Feel free to re-open it, if you have further questions.

gubei1998 commented 1 year ago

I would recommend you to install mmdet3d from their source code instead of directly using pip install.

yes,I truely install mmdet3d from their source code, but it still reports this error.

yichen928 commented 1 year ago

It would be helpful if you can give the entire error message, and version of your cuda, pytorch, mmcv, mmdet, mmdet3d. Your above information does not include many details.

Moyago commented 7 months ago

When I install mmdetection3d==0.11.0,it will report an error as follows, 16 errors detected in the compilation of "mmdet3d/ops/voxel/src/scatter_points_cuda.cu". error: command '/usr/local/cuda-11.5/bin/nvcc' failed with exit code 1 error: subprocess-exited-with-error
× python setup.py develop did not run successfully.
│ exit code: 1
╰─> See above for output.

note: This error originates from a subprocess, and is likely not a problem with pip.
And then I search it and find that it is an bug with cuda11.x+pytorch 1.7.x + mmdet3d==0.11.0,so I want to know 
Have you encountered this problem and how did you solve it? Thank you for your answer

Hi, I meet similar problem which the log like this. ` mmdet3d/ops/voxel/src/scatter_points_cuda.cu(249): error: type name is not allowed

mmdet3d/ops/voxel/src/scatter_points_cuda.cu(249): error: expected an expression

mmdet3d/ops/voxel/src/scatter_points_cuda.cu(249): error: no instance of function template "coors_id_kernel" matches the argument list
            argument types are: (int16_t *, <error-type>, int64_t *, const int, const int)

16 errors detected in the compilation of "mmdet3d/ops/voxel/src/scatter_points_cuda.cu".
error: command '/usr/local/cuda-11.4/bin/nvcc' failed with exit status 1
[end of output]`

then, I following this issue: https://github.com/pytorch/pytorch/issues/44069. I solved this problem on my side. I've only changed max_value to amax.

auto coor_space_dim = coors.amax(0) + 1; //coors.max_values(0) + 1;

Hope it helps.

yichen928 / SparseFusion

questions about cuda version #3