can it run
Comfortably. 256GB of memory against a 4GB working set leaves headroom for context, the OS, and anything else you're running.
weights 0.3GB + KV cache 2GB at 4k context vs ~32GB usable → full. 2.3 GB needed of 32 GB usable — headroom for context
computed roofline: 1792 GB/s × 0.55–0.80 efficiency window / 0.3GB (Q4_K_M) — bands, never points.
| model file | 0.3GB (GGUF Q4_K_M) |
| minimum memory | 4GB working set |
| this machine | 256GB VRAM (8× 32GB) |
| params | base |
| license | apache-2.0 |
| price | $100000 |
easy run: no stock runtime yet
no alternatives — every machine in the catalog can run this one.
→ rtx-pro-6000 — the next step up from this machine's GPU