Build / compile failures

llama.cpp CUDA build: unsupported GNU version! gcc versions later than X are not supported

Q: How do you fix "llama.cpp CUDA build: unsupported GNU version! gcc versions later than X are not supported"?

**1. Install a supported gcc version side-by-side:** ```bash # Ubuntu sudo apt install gcc-12 g++-12 # Fedora sudo dnf install gcc-toolset-12 ``` **2. Point the build at it:** ```bash # CMake cmake -B build -DGGML_CUDA=ON \ -DCMAKE_C_COMPILER=gcc-12 \ -DCMAKE_CXX_COMPILER=g++-12 \ -DCMAKE_CUDA_HOST_COMPILER=g++-12 # Make make GGML_CUDA=1 \ CC=gcc-12 CXX=g++-12 \ NVCCFLAGS="-ccbin g++-12" -j ``` **3. Or update CUDA** to a newer toolkit that supports your default gcc. CUDA 12.6+ supports gcc 14. **4. Quick override** if you don't want to change the system gcc: ```bash export NVCC_PREPEND_FLAGS='-allow-unsupported-compiler' ``` This is a hack — works often, fails on subtle ABI incompatibilities. Prefer a real fix.

error: unsupported GNU version! gcc versions later than 13 are not supported

By Eruo Fredoline · Last verified May 8, 2026

Cause

Each CUDA toolkit version supports a bounded range of host compilers. CUDA 12.4 supports gcc up to 13; CUDA 12.6 supports up to 14. Modern distros (Ubuntu 24.04, Fedora 40) ship gcc 14 by default, which fails against an older CUDA toolkit.

Solution

1. Install a supported gcc version side-by-side:

# Ubuntu
sudo apt install gcc-12 g++-12

# Fedora
sudo dnf install gcc-toolset-12

2. Point the build at it:

# CMake
cmake -B build -DGGML_CUDA=ON \
  -DCMAKE_C_COMPILER=gcc-12 \
  -DCMAKE_CXX_COMPILER=g++-12 \
  -DCMAKE_CUDA_HOST_COMPILER=g++-12

# Make
make GGML_CUDA=1 \
  CC=gcc-12 CXX=g++-12 \
  NVCCFLAGS="-ccbin g++-12" -j

3. Or update CUDA to a newer toolkit that supports your default gcc. CUDA 12.6+ supports gcc 14.

4. Quick override if you don't want to change the system gcc:

export NVCC_PREPEND_FLAGS='-allow-unsupported-compiler'

This is a hack — works often, fails on subtle ABI incompatibilities. Prefer a real fix.

Related errors

Did this fix it?

If your case was different, email Contact support 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.