What hardware runs gpt-oss?

The smallest-memory cataloged machine that runs gpt-oss on the GPU is a AMD Radeon RX 6800 XT · 16GB VRAM. Here are the rigs that fit it, smallest memory first.

machine	best fit	verdict	speed
AMD Radeon RX 6800 XT · 16GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	45–70 tok/s
AMD Radeon RX 6950 XT · 16GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	50–80 tok/s
AMD Radeon RX 7800 XT · 16GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	55–90 tok/s
NVIDIA GeForce RTX 4080 · 16GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	65–110 tok/s
NVIDIA GeForce RTX 5080 · 16GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	85–140 tok/s
AMD Radeon RX 7900 XT · 20GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs comfortably	70–120 tok/s
AMD Radeon RX 7900 XTX · 24GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs comfortably	85–140 tok/s
NVIDIA GeForce RTX 3090 · 24GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs comfortably	90–150 tok/s
NVIDIA GeForce RTX 4090 · 24GB VRAM	20B-A3.6B (MoE) MXFP4	✅ Runs comfortably	95–160 tok/s
Apple M2 · 24GB	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	13–20 tok/s
Apple M4 · 24GB	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	15–25 tok/s
Apple M4 Pro · 24GB	20B-A3.6B (MoE) MXFP4	✅ Runs (tight)	30–50 tok/s

Minimum rig by size

The smallest cataloged machine that runs each size on the GPU (default 8K context). Bigger sizes need more memory — a rig that runs the small one may not touch the large.

size	minimum rig	verdict	speed
120B-A5.1B (MoE)	Apple M2 Max · 96GB	✅ Runs (tight)	~30–45 tok/s (fast)
20B-A3.6B (MoE)	AMD Radeon RX 6800 XT · 16GB VRAM	✅ Runs (tight)	~45–70 tok/s (very fast)

Already have a machine? See gpt-oss across popular rigs → or check your exact rig →

Planning a whole workflow? Rigs for a coding agent · local chat · reasoning · local RAG setup.

$npx runlocal-sh can-i-run gpt-oss --hardware <your-rig>

Verdicts use the same GQA-aware fit math as the CLI — a AMD Radeon RX 6800 XT verdict here matches what runlocal tells you on the machine.