local intelligence

can it run

Can the NVIDIA H200 141GB run qwen3 4B?

FULL

3.2 GB needed of 141 GB usable — headroom for context

141GB · Hopper · model file 2.6GB · 4B

~850–1400 tok/s

computed roofline: 4800 GB/s × 0.46–0.76 efficiency window / 2.6GB (Q4_K_M) — bands, never points. Small models (under ~4.5B) run BELOW these bands — kernel-launch overhead keeps them from saturating bandwidth (measured against community medians, 09-06).

Run qwen3 4B on instead

Tesla M60 (16GB, used) — cheapest catalog machine that runs qwen3 4B fully ($35)

NVIDIA DGX Station (GB300) — upgrade path machine

every model the NVIDIA H200 141GB can run — full list

explore the full catalog90 machines indexed · live prices · what each one can run
2026-09-24 · ← all models on the NVIDIA H200 141GB · fit = weights + 4k KV vs usable memory