local intelligence

can it run

Can the NVIDIA RTX A6000 run qwen3.8 27B?

FULL

18.5 GB needed of 48 GB usable — headroom for context

48GB · Ampere · model file 16.5GB · 27B

~21–35 tok/s

computed roofline: 768 GB/s × 0.46–0.76 efficiency window / 16.5GB (Q4_K_M) — bands, never points.

Run qwen3.8 27B on instead

Mac mini M4 (32GB) — cheapest catalog machine that runs qwen3.8 27B fully ($999)

RTX 4090 (24GB, used) — upgrade path machine

every model the NVIDIA RTX A6000 can run — full list

explore the full catalog84 machines indexed · live prices · what each one can run
2026-09-17 · ← all models on the NVIDIA RTX A6000 · fit = weights + 4k KV vs usable memory