can it run
Comfortably. 384GB of memory against a 4GB working set leaves headroom for context, the OS, and anything else you're running.
weights 2GB + KV cache 0.44GB at 4k context vs ~96GB usable → full. 2.4 GB needed of 96 GB usable — headroom for context
computed roofline: 1792 GB/s × 0.46–0.76 efficiency window / 2.0GB (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).
| model file | 2.0GB (GGUF Q4_K_M) |
| minimum memory | 4GB working set |
| this machine | 384GB VRAM (4× 96GB) |
| params | 3B |
| license | llama3.2 (gated on HF) |
| price | $91000 |
easy run: ollama pull llama3.2:3b
runs with: ollama · llama.cpp · koboldcpp · lm studio
no alternatives — every machine in the catalog can run this one.