What hardware runs gpt-oss 20B?
21B parameters. GGUF files: Q4_K_M 13.8 GB, Q8_0 13.8 GB (GGUF repo, model card). Est. VRAM incl. ~4k context: 16.2 GB (Q4) / 16.2 GB (Q8).
Hardware that fits at Q4 (real UK prices first, then cheapest sample price)
- Radeon AI PRO R9700 (32 GB) £1539.98 REAL
- RTX 5090 (32 GB) £5339.99 REAL
- Tesla P40 (24 GB) ~$300 sample
- RX 7900 XT (20 GB) ~$699 sample
- RTX 3090 (24 GB) ~$800 sample
- RX 7900 XTX (24 GB) ~$899 sample
- RTX 4090 (24 GB) ~$1800 sample
- Ryzen AI Max+ 395 mini-PC (128GB unified) (128 GB) ~$1999 sample
- Mac mini M4 Pro (max 64GB unified) (64 GB) ~$2199 sample
- RTX A6000 (48 GB) ~$4500 sample
- Mac Studio M3 Ultra (max 512GB unified) (512 GB) ~$9499 sample