Windows 11 Installation guide update. Fixed problems with tiny-cuda-nn by using older MSVS version.

[EyGy]

Hi everyone,
I spent an embarrising amount of time (a few days doesn't cut it) trying to make current nerfstudio version work on Windows 11 (GPU: A5000 RTX). The current documentation guide suggests using the the current MS Visual Studio version should work, which definitley was not the case for me. I experienced problems latests at the build stage for tiny-cuda-nn.

Is your feature request related to a problem? Please describe.
Build errors for tiny-cuda-nn using current MS Visual Studio version

Describe the solution you'd like
An update to the installation guide that simplifies things by using a fixed version of MS Visual Studio from https://learn.microsoft.com/en-us/visualstudio/releases/2022/release-history
e.g. tested working version: "17.08 professional"

Additional context

Using the inofficial wsl guide did work. However I experienced performance issues and for my current project WSl was not a good solution.

So here is what I found to be the way to get everything up and running on my windows 11 system (Last updated at 20 February 2025):

---> Make sure your nvidia display drivers are up to date (worked with 572.16) and you start clean (NO MS Visual Studio version or cuda tookit is installed): make sure you double check looking for "nvidia" and "visual studio" in the Windows menu for "Apps and features". Deinstall all existing versions first!

1. install git

2. install conda

3. install Visual Studio 2022 with an older version (e.g. 17.8)
   --> must NOT be 17.10 OR NEWER ---> choose from (https://learn.microsoft.com/en-us/visualstudio/releases/2022/release-history)
   tested working version is 17.08 professional
   make sure to set checkbox for C++ Desktop (DO NOT set additional checkbox for "MSVC 142 for Visual Studio 2019" as menitoned in official guide)

4. install cuda toolkit 11.8 manually (only cuda package required - you can deactivate the other two checkboxes in "custom installation")
   --> https://developer.nvidia.com/cuda-11-8-0-download-archive

--> from now on ONLY use "x64 Native Tools Command Prompt for VS 2022 LTSC 17.8" as terminal!!
Search in Windows Startmenu (Default location for me: C:\ProgramData\Microsoft\Windows\Start Menu\Programs\Visual Studio 2022\Visual Studio Tools\VC)

5. create and activate conda env + install pip:

conda create --name nerfstudio -y python=3.8
conda activate nerfstudio 
python -m pip install --upgrade pip

6. Install PyTorch 2.1.2 with CUDA 11.8:
   pip install torch==2.1.2+cu118 torchvision==0.16.2+cu118 --extra-index-url https://download.pytorch.org/whl/cu118
   DO NOT INSTALL cudatoolkit via conda (step: conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit) since we already did it manually after MS visualstudio install (see above)
   --> you can validate existing version with nvcc --version

7. Install tiny-cuda-nn
   pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch
   if it fails because of an error msg mentioning rust/cargo --> download and execute exe from here: https://rustup.rs/ and try again

8. Install required python packages:

pip install nerfstudio
ns-install-cli

9. If splatfacto fails try:

pip uninstall gsplat
set DISTUTILS_USE_SDK=1
pip install git+https://github.com/nerfstudio-project/gsplat.git