Open wjmenu opened 1 week ago
Here's a summary I've made of all past issues relating to simple-knn and diff-gaussian-rasterization. It eventually worked for me after 4 hours, 10+ reboots, and back-rolling VS2022 Community to an earlier version.
Environment setup issue (github.com)
"diff-gaussian-rasterization" and "simple-knn" not installing properly.
Always use the conda terminal and work under the git repo’s directory.
CUDA toolkit 11.7 was installed system-wide before all the following took place.
VS2022 v17.10.3 fails regardless, whilst v17.6.4 works.
Delete these before cloning the repo and creating conda environment again: C:\Users\YOURNAME.conda\envs\gaussian_splatting C:\Users\YOURNAME\gaussian-splatting
Make sure cloning is recursive (otherwise the hyperlinked folders in the github repo doesn’t get cloned):
git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive
Need “C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\Hostx64\x64” to be in PATH for the build command to find cl.exe
Also need C++ devtools ticked in VS installer.
CUDA and PyTorch versions need to match: maybe use pip instead of conda. For example, for CUDA 11.7 (11.7.1 works):
conda create -n gaussian_splatting python=3.7
conda activate gaussian_splatting
conda install -c conda-forge vs2022_win-64
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url
https://download.pytorch.org/whl/cu117 (or alternatively? conda install pytorch==1.13.1 torchvision==0.14.1
torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia)
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn
pip install plyfile
pip install tqdm
Note that cudatoolkit version in environment.yml is WRONG. Change it to be 11.7 or 11.8, not 11.6. Some reported problems using 11.8, but some succeeded with it.
Check if PyTorch can access CUDA:
(gaussian_splatting) c:\dev\gaussian-splatting>python
>>> import torch
>>> torch.cuda.is_available()
True
(then press Ctrl+Z to exit)
To restart an aborted installation:
conda env update --file environment.yml --name gaussian_splatting
Edit 2024-06-27:
Environment: Windows 11 home version CPU: Intel i7-12700H GPU: NVIDIA RTX 3070 16GB RAM x64 system Windows 11 home version
Thanks for your reply! I tried your tips, but sadly to no avail. However, this line caught my attention :
nvcc fatal : Unsupported gpu architecture 'compute_89'
I believe I have to update the nvcc to a newer version, so that it supports my compute architecture. At the moment, this is not possible. When it does I'll let you know if it worked 👍
Here's a summary I've made of all past issues relating to simple-knn and diff-gaussian-rasterization. It eventually worked for me after 4 hours, 10+ reboots, and back-rolling VS2022 Community to an earlier version.
Environment setup issue (github.com)
"diff-gaussian-rasterization" and "simple-knn" not installing properly.
- Always use the conda terminal and work under the git repo’s directory.
- CUDA toolkit 11.7 was installed system-wide before all the following took place.
- VS2022 v17.10.3 fails regardless, whilst v17.6.4 works.
- Delete these before cloning the repo and creating conda environment again: C:\Users\YOURNAME.conda\envs\gaussian_splatting C:\Users\YOURNAME\gaussian-splatting
- Make sure cloning is recursive (otherwise the hyperlinked folders in the github repo doesn’t get cloned):
git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive
- Need “C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\Hostx64\x64” to be in PATH for the build command to find cl.exe
- Also need C++ devtools ticked in VS installer.
- CUDA and PyTorch versions need to match: maybe use pip instead of conda. For example, for CUDA 11.7 (11.7.1 works):
conda create -n gaussian_splatting python=3.7
conda activate gaussian_splatting
conda install -c conda-forge vs2022_win-64
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url
https://download.pytorch.org/whl/cu117 (or alternatively? conda install pytorch==1.13.1 torchvision==0.14.1
torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia)
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn
pip install plyfile
pip install tqdm
- Note that cudatoolkit version in environment.yml is WRONG. Change it to be 11.7 or 11.8, not 11.6. Some reported problems using 11.8, but some succeeded with it.
- Check if PyTorch can access CUDA:
(gaussian_splatting) c:\dev\gaussian-splatting>python
>>> import torch
>>> torch.cuda.is_available()
True
(then press Ctrl+Z to exit)- To restart an aborted installation:
conda env update --file environment.yml --name gaussian_splatting
Edit 2024-06-27:
Environment: Windows 11 home version CPU: Intel i7-12700H GPU: NVIDIA RTX 3070 16GB RAM x64 system Windows 11 home version
Thanks for this detailed description of your troubleshooting.
Could you talk a bit about how you downgrade to VS2022 v17.6.4? I am unable to find a way to downgrade the software from anything other than v17.10.3 which it installs by default.
Could you also mention where you got the“C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\Hostx64\x64” folder from? I am unable to add it to PATH since it doesn't exist on my computer.
Here's a summary I've made of all past issues relating to simple-knn and diff-gaussian-rasterization. It eventually worked for me after 4 hours, 10+ reboots, and back-rolling VS2022 Community to an earlier version. Environment setup issue (github.com) "diff-gaussian-rasterization" and "simple-knn" not installing properly.
- Always use the conda terminal and work under the git repo’s directory.
- CUDA toolkit 11.7 was installed system-wide before all the following took place.
- VS2022 v17.10.3 fails regardless, whilst v17.6.4 works.
- Delete these before cloning the repo and creating conda environment again: C:\Users\YOURNAME.conda\envs\gaussian_splatting C:\Users\YOURNAME\gaussian-splatting
- Make sure cloning is recursive (otherwise the hyperlinked folders in the github repo doesn’t get cloned):
git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive
- Need “C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\Hostx64\x64” to be in PATH for the build command to find cl.exe
- Also need C++ devtools ticked in VS installer.
- CUDA and PyTorch versions need to match: maybe use pip instead of conda. For example, for CUDA 11.7 (11.7.1 works):
conda create -n gaussian_splatting python=3.7
conda activate gaussian_splatting
conda install -c conda-forge vs2022_win-64
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url
https://download.pytorch.org/whl/cu117 (or alternatively? conda install pytorch==1.13.1 torchvision==0.14.1
torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia)
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn
pip install plyfile
pip install tqdm
- Note that cudatoolkit version in environment.yml is WRONG. Change it to be 11.7 or 11.8, not 11.6. Some reported problems using 11.8, but some succeeded with it.
- Check if PyTorch can access CUDA:
(gaussian_splatting) c:\dev\gaussian-splatting>python
>>> import torch
>>> torch.cuda.is_available()
True
(then press Ctrl+Z to exit)- To restart an aborted installation:
conda env update --file environment.yml --name gaussian_splatting
Edit 2024-06-27: Environment: Windows 11 home version CPU: Intel i7-12700H GPU: NVIDIA RTX 3070 16GB RAM x64 system Windows 11 home version
Thanks for this detailed description of your troubleshooting.
- Could you talk a bit about how you downgrade to VS2022 v17.6.4? I am unable to find a way to downgrade the software from anything other than v17.10.3 which it installs by default.
- Could you also mention where you got the“C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\Hostx64\x64” folder from? I am unable to add it to PATH since it doesn't exist on my computer.
Same issue here. It seems like Microsoft does not allow you to downgrade the Visual Studio C compiler without a paid subscription. v17.10.3 is the latest and that's all you get the Community bundle. If you have Visual Studio installed, there should be either a 2019 or 2022 folder at C:\Program Files (x86)\Microsoft Visual Studio\
Update: The rasterization submodule did build properly for me using the VS 2019 compiler here: https://visualstudio.microsoft.com/thank-you-downloading-visual-studio/?sku=Community&rel=16
Here's a summary I've made of all past issues relating to simple-knn and diff-gaussian-rasterization. It eventually worked for me after 4 hours, 10+ reboots, and back-rolling VS2022 Community to an earlier version.
Environment setup issue (github.com)
"diff-gaussian-rasterization" and "simple-knn" not installing properly.
- Always use the conda terminal and work under the git repo’s directory.
- CUDA toolkit 11.7 was installed system-wide before all the following took place.
- VS2022 v17.10.3 fails regardless, whilst v17.6.4 works.
- Delete these before cloning the repo and creating conda environment again: C:\Users\YOURNAME.conda\envs\gaussian_splatting C:\Users\YOURNAME\gaussian-splatting
- Make sure cloning is recursive (otherwise the hyperlinked folders in the github repo doesn’t get cloned):
git clone https://github.com/graphdeco-inria/gaussian-splatting --recursive
- Need “C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\Hostx64\x64” to be in PATH for the build command to find cl.exe
- Also need C++ devtools ticked in VS installer.
- CUDA and PyTorch versions need to match: maybe use pip instead of conda. For example, for CUDA 11.7 (11.7.1 works):
conda create -n gaussian_splatting python=3.7
conda activate gaussian_splatting
conda install -c conda-forge vs2022_win-64
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url
https://download.pytorch.org/whl/cu117 (or alternatively? conda install pytorch==1.13.1 torchvision==0.14.1
torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia)
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn
pip install plyfile
pip install tqdm
- Note that cudatoolkit version in environment.yml is WRONG. Change it to be 11.7 or 11.8, not 11.6. Some reported problems using 11.8, but some succeeded with it.
- Check if PyTorch can access CUDA:
(gaussian_splatting) c:\dev\gaussian-splatting>python
>>> import torch
>>> torch.cuda.is_available()
True
(then press Ctrl+Z to exit)- To restart an aborted installation:
conda env update --file environment.yml --name gaussian_splatting
Edit 2024-06-27:
Environment: Windows 11 home version CPU: Intel i7-12700H GPU: NVIDIA RTX 3070 16GB RAM x64 system Windows 11 home version
you are my hero
Error Description: Hey, I have some issues with the installation of the submodule diff_gaussian_rasterization (and likewise simple_knn) occuring from installing the environment.yml .
CUDA 11.5 Ubuntu 22.04.4 RTX 4090 compute capability 8.9
There is a warning about the mismatch with the version that was used to compile PyTorch (11.6) and the detected CUDA version (11.5), but I do not think that this is the problem.
Error Output:
If anyone knows a fix for this, please let me know. Thanks in advance.