can it run
Yes, but close to the edge. 256GB against a 205GB working set: it runs, with little room for long context windows or other processes.
weights 164GB + KV cache 2GB at 4k context vs ~32GB usable → no. 166.0 GB vs 32 GB usable — does not fit
computed roofline: 1792 GB/s × 0.55–0.80 efficiency window / 9.2GB (Q4_K_M) — bands, never points.
MoE: file is 164GB (that decides fit) but only ~18B of experts activate per token (~9.2GB streamed) — band below is the active-expert estimate, optimistic; community MoE rows run several× the naive bandwidth math
| model file | 164GB (GGUF Q4_K_M) |
| minimum memory | 205GB working set |
| this machine | 256GB VRAM (8× 32GB) |
| params | 320B-A18B MoE |
| license | MIT |
| price | $100000 |
easy run: no stock runtime yet
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — cheapest machine that runs it ($60000)
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — this model on the biggest machine
→ rtx-pro-6000 — the next step up from this machine's GPU