Pyyan / Local AI Hardware
Machines for running models at home. · ranked by tokens per second per dollar · verified 13 Aug 2026
| # | Name | Memory | Bandwidth | Price | 70B tok/s | Status |
|---|---|---|---|---|---|---|
| 1 | Framework | 128GB unified | ~256 GB/s | $1,999 | 12–15 | current |
| 2 | Mac Studio M3 Ultra Apple | up to 512GB | 546–819 GB/s | $3–6K | 20–30 | current |
| 3 | RTX 5090 NVIDIA | 32GB GDDR7 | 1,792 GB/s | $2–3K | — | current |
| 4 | RTX 3090 (used) NVIDIA | 24GB | — | ~$900 | — | current |
| 6 | RTX PRO 6000 Blackwell NVIDIA | 96GB | — | $13,250–20,000 | — | current |
| 9 | DGX Spark NVIDIA | 128GB LPDDR5x | 273 GB/s | $4,699 | 2.7 | current |