RTX 5070 + Ubuntu successfully installed submodules

Author: PangpangjuCreated Dec 31, 2025Updated May 26, 2026

I can see alot of people(including myself) had or are having trouble installing submodules, due to failed-building-wheels or subprocess-exited-with-error.

Initializing the project with environment.yml always ended up with issues for many environments in my case (Ubuntu and Windows both) So I share my solution so it might help other people going through the same issue.

bash


conda create -n "gaussian-splatting" python=3.9 ipython
conda activate gaussian-splatting
conda install cuda-toolkit=12.8 -c nvidia
conda install -c conda-forge plyfile
conda install tqdm
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
pip install opencv-python
pip install joblib

Open up submodules/diff-gaussian-rasterization/cuda_rasterizer/rasterizer_impl.h at the header section include this line below #include <cstdint>

Move to submodules/diff-gaussian-rasterization

bash

pip install . --no-build-isolation
cd ../simple-knn
pip install . --no-build-isolation
cd ../fused-ssim
pip install . --no-build-isolation

If the following error pops up in the training process after successfully installing the submodule

Traceback (most recent call last):
  File "train.py", line 216, in <module>
    training(lp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint, args.debug_from)
  File "train.py", line 35, in training
    scene = Scene(dataset, gaussians)
  File "C:\gaussian-splatting\scene\__init__.py", line 83, in _init_
    self.gaussians.create_from_pcd(scene_info.point_cloud, self.cameras_extent)
  File "C:\gaussian-splatting\scene\gaussian_model.py", line 134, in create_from_pcd
    dist2 = torch.clamp_min(distCUDA2(torch.from_numpy(np.asarray(pcd.points)).float().cuda()), 0.0000001)
MemoryError: bad allocation: cudaErrorMemoryAllocation: out of memory

You need to reinstall the submodules by replacing the pip command with the option pip install . --no-build-isolation with TORCH_CUDA_ARCH_LIST="6.0 7.0 7.5 8.0 8.6+PTX" pip install . --no-build-isolation

Reinstall the package after moving into each submodule directory and uninstalling each submodule with

bash

pip uninstall (diff_gaussian_rasterization)
python setup.py clean

Replace it with the package name of each submodule directory (check with pip list)

So your reinstall command would look like

bash

cd submodules/diff-gaussian-rasterization
TORCH_CUDA_ARCH_LIST="6.0 7.0 7.5 8.0 8.6+PTX" pip install . --no-build-isolation
cd ../simple-knn
TORCH_CUDA_ARCH_LIST="6.0 7.0 7.5 8.0 8.6+PTX" pip install . --no-build-isolation
cd ../fused-ssim
TORCH_CUDA_ARCH_LIST="6.0 7.0 7.5 8.0 8.6+PTX" pip install . --no-build-isolation

Source: graphdeco-inria/gaussian-splatting

View original on GitHub View discussion on GitHub