can it run
14.2 GB vs 12 GB usable — partial CPU offload, expect large speed loss
12GB · Pascal · model file 14GB · 20B MoE
computed roofline: 480.4 GB/s × 0.28–0.50 efficiency window / 2.1GB (Q4_K_M) — bands, never points.
MoE: file is 14GB (that decides fit) but only ~3B of experts activate per token (~2.1GB streamed) — band below is the active-expert estimate, optimistic; community MoE rows run several× the naive bandwidth math
→ Tesla M40 (24GB, used) — cheapest catalog machine that runs gpt-oss 20B MoE fully ($50)
→ RTX 3090 (24GB, used) — upgrade path machine
→ every model the NVIDIA Titan X (Pascal) 12GB can run — full list
→explore the full catalog90 machines indexed · live prices · what each one can run