can it run
Only with compromises. 80GB is below the 84GB working set — expect a lower quant, offloading, or heavy swap. Possible, not pleasant.
weights 67GB + KV cache 0.75GB at 4k context vs ~80GB usable → tight. 67.8 of 80 GB usable — keep context modest
computed roofline: 2039 GB/s × 0.46–0.76 efficiency window / 67.0GB (Q4_K_M) — bands, never points.
| model file | 67GB (GGUF Q4_K_M) |
| minimum memory | 84GB working set |
| this machine | 80GB HBM2e, 2 TB/s |
| params | Scout 109B A17B |
| license | llama4 (gated on HF) |
| price | $23000 |
easy run: ollama pull llama4:scout
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ NIMO AI Mini PC (128GB) — cheapest machine that runs it ($1999)
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — this model on the biggest machine
→ rtx-pro-6000 — the next step up from this machine's GPU
NVIDIA (used market) ↗ · Amazon ↗