can it run
No. 8GB of memory can't hold the 307GB working set this model needs. A smaller quant or a bigger machine is required.
weights 245GB + KV cache 0.75GB at 4k context vs ~8GB usable → no. 245.8 GB vs 8 GB usable — does not fit
computed roofline: 192.3 GB/s × 0.28–0.50 efficiency window / 245.0GB (Q4_K_M) — bands, never points.
| model file | 245GB (GGUF Q4_K_M) |
| minimum memory | 307GB working set |
| this machine | 8GB GDDR5, 192 GB/s |
| params | Maverick 400B A17B |
| license | llama4 (gated on HF) |
| price | $100 |
easy run: ollama pull llama4:maverick
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — cheapest machine that runs it ($60000)
→ every model the Tesla P4 (8GB, used) can run — full list
→ tesla-p100-16gb — the next step up from this machine's GPU