CUDA_ARCHITECTURES is set to "native", but no NVIDIA GPU was detected

bash_bob

I am trying to install llama-cpp-python with GPU support. I installed Nvidia CUDA Toolkit v13.1, nvidia-smi shows that my graphics card - Geforce GTX 1050 Ti - supports CUDA v13, nvcc is installed correctly. Before calling pip I set some environment Bash variables: set CMAKE_ARGS=-DGGML_CUDA=ON -DCUDAToolkit_ROOT="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v13.1" -DCUDA_TOOLKIT_ROOT_DIR="C:/Program Files/NVIDIA GPU Computing...

Originally asked by Стебан on Stack Exchange (stackoverflow) (CC BY-SA 4.0).

uphilltech

Great question — answering as the UphillTech team:

Issue solved! Thanks to @paleonix and @jgonz in comments! I reinstalled CUDA Toolkit to version 12.9. As @paleonix wrote Pascal architecture on GTX 1050 Ti is not supported since v13.0. And I replaced one CMake argument: -DCMAKE_CUDA_FLAGS="-gencode arch=native" -> -DCMAKE_CUDA_ARCHITECTURES=61

Full answer by Стебан: https://stackoverflow.stackexchange.com/a/79873633 (CC BY-SA 4.0).

Need it done for your business? UphillTech.