VoicePad

NVIDIA GPU requirements

The production deployment requires a CUDA-capable NVIDIA GPU and Linux x86_64. Startup validates:

  • PyTorch CUDA availability;
  • stable NVIDIA device UUID;
  • FP16 compute capability;
  • physical and currently free memory;
  • complete CUDA placement after model load.

The maintained baseline is an RTX 3050 Laptop GPU in the 4 GB physical class. PyTorch exposes 3,953,393,664 bytes on that device, so admission uses a 3,900,000,000-byte floor plus a free-memory safety margin.

VoicePad does not fall back to CPU, lower precision, ONNX ASR, or another model. Close other GPU-heavy programs if admission reports insufficient free memory. A supported NVIDIA driver is a system prerequisite; a separate CUDA toolkit is not required.