Build / compile failures

llama.cpp build fails: nvcc not found

Q: How do you fix "llama.cpp build fails: nvcc not found"?

**Check if nvcc exists:** ```bash which nvcc nvcc --version ``` **Install the CUDA Toolkit:** Linux (Ubuntu): ```bash wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb sudo dpkg -i cuda-keyring_1.1-1_all.deb sudo apt update && sudo apt install cuda-toolkit-12-6 ``` Then add to PATH: ```bash echo 'export PATH=/usr/local/cuda/bin:$PATH' >> ~/.bashrc echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc source ~/.bashrc ``` **Windows:** Install the CUDA Toolkit installer from developer.nvidia.com/cuda-downloads. Use the visual studio integration option. Open the "x64 Native Tools Command Prompt for VS" — nvcc will be on PATH there. **Build llama.cpp:** ```bash make clean GGML_CUDA=1 make -j ``` **Skip CUDA** if you don't actually need it (CPU-only build is fast and works): ```bash make -j ```

GGML_USE_CUDA defined but nvcc not found in PATH

By Fredoline Eruo · Last verified May 6, 2026

Cause

Building llama.cpp with CUDA requires the full CUDA Toolkit (nvcc compiler), not just the GPU driver. The driver alone is enough to RUN CUDA programs, but you need the toolkit to COMPILE them.

Solution

Check if nvcc exists:

which nvcc
nvcc --version

Install the CUDA Toolkit:

Linux (Ubuntu):

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update && sudo apt install cuda-toolkit-12-6

Then add to PATH:

echo 'export PATH=/usr/local/cuda/bin:$PATH' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
source ~/.bashrc

Windows: Install the CUDA Toolkit installer from developer.nvidia.com/cuda-downloads. Use the visual studio integration option. Open the "x64 Native Tools Command Prompt for VS" — nvcc will be on PATH there.

Build llama.cpp:

make clean
GGML_CUDA=1 make -j

Skip CUDA if you don't actually need it (CPU-only build is fast and works):

make -j

Did this fix it?

If your case was different, email hello@runlocalai.co with what you saw and we'll update the page. If it worked but took different commands on your platform, we want to know that too.