Run Ordinary CUDA Applications
Applications should consume STRATUM GPU as an ordinary CUDA device. The llama.cpp device listing is a direct check that an unmodified CUDA application can discover it.
Use the GPU without application-specific STRATUM changes
The application remains an ordinary CUDA application. Guest Tools supplies the STRATUM driver-side integration; the application should not need a STRATUM wrapper, a manual library-path override, or a STRATUM-specific rebuild.
Use llama.cpp device discovery as a concrete compatibility check: if the CUDA backend lists the STRATUM GPU, the application is seeing the virtual accelerator through the intended guest path.
Interface tour
llama.cpp Detects STRATUM GPU
Run the application’s normal device-discovery command. In this example, llama.cpp lists the STRATUM GPU as a CUDA-capable device without a STRATUM-specific application build.
Commands
Run these checks inside the guest. They verify the supported customer path and intentionally omit development-only library-path, mock-mode, and native-port overrides.
./build/bin/llama-server --list-devicesThe outcome you should see
The application lists the STRATUM GPU as a CUDA-capable device without STRATUM-specific application changes.