[Docs] How I Got MMCV to Build on RTX 5090 (CUDA 12.8)

[YuchenZHANG2]

📚 The doc issue

I had some trouble installing MMCV on my RTX 5090 at first, but I was eventually able to get it working. I’m sharing the steps that worked for me below and hope this helps anyone facing similar issues.

Suggest a potential alternative/fix

# Create environment
conda create -n mmcv_py310 python=3.10 -y
conda activate mmcv_py310

# Install PyTorch (CUDA 12.8)
pip install torch torchvision torchaudio \
  --index-url https://download.pytorch.org/whl/cu128

# Install CUDA compiler + toolkit (12.8)
conda install -c nvidia cuda-nvcc=12.8 -y
conda install -c nvidia cuda-toolkit=12.8 -y

# Set CUDA paths (for conda CUDA layout)
export CUDA_HOME=$CONDA_PREFIX
export CUDACXX=$CONDA_PREFIX/bin/nvcc
export CPATH=$CONDA_PREFIX/targets/x86_64-linux/include:$CPATH
export LIBRARY_PATH=$CONDA_PREFIX/targets/x86_64-linux/lib:$LIBRARY_PATH

# Clone mmcv 
git clone https://github.com/open-mmlab/mmcv.git
cd mmcv
pip install -r requirements/optional.txt

# Clean previous build (if any)
rm -rf build mmcv.egg-info

# Build mmcv from source
MAX_JOBS=1 python setup.py develop

# Verify installation
python .dev_scripts/check_installation.py