llama.cpp Failure to Utilize GPU Hardware

To confirm that llama.cpp is failing to use your GPU, observe the system behavior during inference. If the CPU utilization remains at or near 100% while GPU memory (VRAM) usage remains flat or at idle levels, the application is performing CPU-only inference. You can verify device detection by executing the command './llama-server --list-devices' in your terminal; if no GPU is returned, the system is not communicating with the hardware.

Sources