LamChingFung-2425-Anti-FraudX

Feature: GPU Detection

Category: Secondary Version: 4.1 Last Updated: 2026-03-11

Overview

At startup, the backend detects whether an NVIDIA GPU is available and logs the result. In strict mode (FORCE_GPU=1), the server refuses to start without a GPU — ensuring Ollama always runs with acceleration.

Implementation

File: backend/utils/gpu_checker.py

def check_and_enforce_gpu() -> bool:
    # Runs: nvidia-smi --query-gpu=name,memory.total --format=csv,noheader
    # Returns True if GPU found, False otherwise
    # If FORCE_GPU=1 and no GPU: sys.exit(1)

Called from main.py during startup, before any router registration.

Behaviour by Config

`FORCE_GPU` value	GPU found	Behaviour
`0` (default)	Yes	Logs GPU name + VRAM, continues
`0` (default)	No	Logs warning, continues (CPU mode)
`1`	Yes	Logs GPU info, continues
`1`	No	Logs error, `sys.exit(1)`

Notes

gpu_status.json is not cached between runs — detection executes fresh on every startup
Only detects NVIDIA GPUs via nvidia-smi; AMD/Intel GPUs are not detected
Ollama uses the detected GPU automatically — no additional config needed

Relevant Files

backend/utils/gpu_checker.py   check_and_enforce_gpu()
backend/main.py                Calls gpu_checker at startup
backend/.env                   FORCE_GPU=0 (default)

llm-backend.md — Ollama benefits from GPU acceleration