can it run
Comfortably. 128GB of memory against a 84GB working set leaves headroom for context, the OS, and anything else you're running.
weights 67GB + KV cache 0.75GB at 4k context vs ~126GB usable → full. 67.8 GB needed of 126 GB usable — headroom for context
computed roofline: 546 GB/s × 0.55–0.80 efficiency window / 67.0GB (Q4_K_M) — bands, never points.
| model file | 67GB (GGUF Q4_K_M) |
| minimum memory | 84GB working set |
| this machine | 128GB unified |
| params | Scout 109B A17B |
| license | llama4 (gated on HF) |
| price | $3499 |
easy run: ollama pull llama4:scout
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — this model on the biggest machine