[llama.cpp Windows 11 Pro 25H2] The CUDA compiler nvcc.exe is not able to compile a simple test program.

[andreiramani]

I can not load qwen3 model (Qwen3-8B-Q4_K_M.gguf) with llama.cpp.

So, I decide to compile manually using VS & cuda toolkit (both install successfully).

cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES="86;89" -DCMAKE_GENERATOR_TOOLSET="cuda=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8" -DLLAMA_CURL=OFF

cmake --build build --config Release -j 8

error raise when testing CUDA:

...
-- CUDA Toolkit found
-- Using CUDA architectures: 86;89
-- The CUDA compiler identification is NVIDIA 12.8.61 with host compiler MSVC 19.50.35719.0
-- Detecting CUDA compiler ABI info
-- Detecting CUDA compiler ABI info - failed
-- Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.8/bin/nvcc.exe
-- Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.8/bin/nvcc.exe - broken
CMake Error at D:/install/Microsoft Visual Studio/18/Community/Common7/IDE/CommonExtensions/Microsoft/CMake/CMake/share/cmake-4.1/Modules/CMakeTestCUDACompiler.cmake:59 (message):
  The CUDA compiler

    "C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.8/bin/nvcc.exe"

  is not able to compile a simple test program.

  It fails with the following output:

    Change Dir: 'E:/gpt/llama/build/CMakeFiles/CMakeScratch/TryCompile-i5fdet'

    Run Build Command(s): "D:/install/Microsoft Visual Studio/18/Community/MSBuild/Current/Bin/amd64/MSBuild.exe" cmTC_828b3.vcxproj /p:Configuration=Debug /p:Platform=x64 /p:VisualStudioVersion=18.0 /v:n
    MSBuild version 18.0.5+e22287bf1 for .NET Framework
    Build started 11/27/2025 6:40:43 PM.
...
...

Any suggestion to fix this issue much appreciated .

note: Windows 11 Pro 25H2, 32GB RAM, RTX 3050 6GB, CUDA 12.8, llama.cpp source code tag e509411