gpu-discovery

name	gpu-discovery
description	Detect GPU hardware and CUDA environment, output a structured report for downstream skills. Use this skill before any GPU-intensive task (training, inference, rendering) to determine available hardware and choose appropriate model sizes, batch sizes, and device settings.
compatibility	{"requires":["Bash","Python (>= 3.8)"],"dependencies":[]}

GPU Discovery

Detect the GPU environment and produce a machine-readable report that other skills can consume to make hardware-aware decisions (model size, batch size, device selection).

Workflow

1. Run detection script → 2. Read report → 3. Choose configuration

---

Step 1: Run Detection Script

Run the detection script — it has zero external dependencies (stdlib only):

python scripts/detect_gpu.py

The script writes gpu_info.json and prints a human-readable summary to stdout.

---

Step 2: Read the Report

Read gpu_info.json. The schema:

GPU available:

{
  "mode": "gpu",
  "gpu_count": 1,
  "gpus": [{
    "index": 0,
    "name": "NVIDIA GeForce RTX 4090",
    "memory_total_mb": 24564,
    "memory_free_mb": 22000,
    "cuda_version": "12.8",
    "driver_version": "560.35.03"
  }]
}

No GPU:

{
  "mode": "cpu",
  "gpu_count": 0,
  "gpus": []
}