Incorrect Open3D PyTorch and CUDA version installed

[RichardSnijder]

Checklist

• [X] I have searched for similar issues.
• [X] For Python issues, I have tested with the latest development wheel.
• [X] I have checked the release documentation and the latest documentation (for master branch).

My Question

Hello,

Recently, I have installed Open3D, for the o3d.ml functionalities. In order to do so I am planning to use PyTorch and CUDA for GPU usability.

I have installed Pytorch with CUDA via the pip installation command found here: https://pytorch.org/ Whilst trying to run a o3d.ml related command I received the following warning:

--------------------------------------------------------------------------------

 Using the Open3D PyTorch ops with CUDA 11 may have stability issues!

 We recommend to compile PyTorch from source with compile flags
   '-Xcompiler -fno-gnu-unique'

 or use the PyTorch wheels at
   https://github.com/isl-org/open3d_downloads/releases/tag/torch1.8.2


 Ignore this message if PyTorch has been compiled with the aforementioned
 flags.

 See https://github.com/isl-org/Open3D/issues/3324 and
 https://github.com/pytorch/pytorch/issues/52663 for more information on this
 problem.

--------------------------------------------------------------------------------

Warning: Open3D was built with CUDA 11.0 butPyTorch was built with CUDA 11.1. Falling back to CPU for now.Otherwise, install PyTorch with CUDA 11.0.

From this message I understand the Open3D has been built using CUDA 11.0. However, when trying to install this version, using the following command in the terminal:

pip install torch==1.8.1+cu110 torchvision==0.9.1+cu110 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

It won't let me install it, yielding in the following message in the terminal:

ERROR: Could not find a version that satisfies the requirement torch==1.8.1+cu110 (from versions: 1.4.0, 1.4.0+cpu,
1.4.0+cu100, 1.4.0+cu92, 1.5.0, 1.5.0+cpu, 1.5.0+cu101, 1.5.0+cu92, 1.5.1, 1.5.1+cpu, 1.5.1+cu101, 1.5.1+cu92, 1.6.0,
1.6.0+cpu, 1.6.0+cu101, 1.6.0+cu92, 1.7.0, 1.7.0+cpu, 1.7.0+cu101, 1.7.0+cu110, 1.7.0+cu92, 1.7.1, 1.7.1+cpu, 1.7.1+cu101,
1.7.1+cu110, 1.7.1+cu92, 1.7.1+rocm3.7, 1.7.1+rocm3.8, 1.8.0, 1.8.0+cpu, 1.8.0+cu101, 1.8.0+cu111, 1.8.0+rocm3.10,
1.8.0+rocm4.0.1, 1.8.1, 1.8.1+cpu, 1.8.1+cu101, 1.8.1+cu102, 1.8.1+cu111, 1.8.1+rocm3.10, 1.8.1+rocm4.0.1, 1.9.0, 1.9.0+cpu,
1.9.0+cu102, 1.9.0+cu111, 1.9.0+rocm4.0.1, 1.9.0+rocm4.1, 1.9.0+rocm4.2, 1.9.1, 1.9.1+cpu, 1.9.1+cu102, 1.9.1+cu111,
1.9.1+rocm4.0.1, 1.9.1+rocm4.1, 1.9.1+rocm4.2, 1.10.0, 1.10.0+cpu, 1.10.0+cu102, 1.10.0+cu111, 1.10.0+cu113, 
1.10.0+rocm4.0.1, 1.10.0+rocm4.1, 1.10.0+rocm4.2, 1.10.1, 1.10.1+cpu, 1.10.1+cu102, 1.10.1+cu111, 1.10.1+cu113, 
1.10.1+rocm4.0.1, 1.10.1+rocm4.1, 1.10.1+rocm4.2, 1.10.2, 1.10.2+cpu, 1.10.2+cu102, 1.10.2+cu111, 1.10.2+cu113, 
1.10.2+rocm4.0.1, 1.10.2+rocm4.1, 1.10.2+rocm4.2, 1.11.0, 1.11.0+cpu, 1.11.0+cu102, 1.11.0+cu113, 1.11.0+cu115, 
1.11.0+rocm4.3.1, 1.11.0+rocm4.5.2)
ERROR: No matching distribution found for torch==1.8.1+cu110

I have read the installation guide from the READ_ME.md (https://github.com/isl-org/Open3D-ML) but here, a command is called pip install -r requirements-torch-cuda.txt. Unfortunately this results in an error: ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements-torch-cuda.txt'.

Versions currently installed:

• Distribution: Ubuntu 20.04.4 LTS
• Open3D 0.15.2
• PyTorch 1.8.2+cu111 (+ CUDA 11.1)

Can someone give me some advice on how to get the right CUDA and PyTorch version installed? Please let me know if you require additional information.

Any help would be highly appreciated!

Kind regards Richard

[haixuanTao]

I think that there is no prebuilt version for Torch 1.8.2 + CUDA 11.0 You should probably try to build torch from source and specify the cuda version you want.

But with the latest open3d, it should be possible to use: CUDA 11.x You should try to run it again.