Benchmarks
Real measurements using Ollama (median of 3 runs, 4K context, 256 generated tokens). Community submission rows are submitted by readers with the same script and reviewed before publishing, but we cannot re-check their hardware. Submit your results.
Measured by FitGPU rows were run by us on our own machines. Community-reported, unverified rows are published results we found on the web (llama.cpp discussion threads, GitHub repos). We copied the numbers exactly as posted and link each source, but we have not reproduced them: hardware, drivers, settings and test methods differ, so treat them as rough reference points, not like-for-like comparisons. GPU memory specs under each GPU come from official AMD, NVIDIA and Apple spec pages ("src" links); anything a vendor page does not state is left blank.
- Ryzen 9 7900X + Radeon RX 6950 XT 16 GB + 64 GB RAM (Windows) — measured
- MinisForum AI X1 Pro-470 (Ryzen AI 9 HX 470, 64 GB RAM, Windows) — measured
Results
| Source ↕ | GPU ↕ | CPU ↕ | VRAM ↕ | Model ↕ | Quant ↕ | Runtime / backend ↕ | Prompt tok/s ↕ | Gen tok/s ↕ | Date ↕ |
|---|---|---|---|---|---|---|---|---|---|
| No measured results yet. | |||||||||