can it run
Comfortably. 32GB of memory against a 4GB working set leaves headroom for context, the OS, and anything else you're running.
weights 1.7GB + KV cache 2GB at 4k context vs ~32GB usable → full. 3.7 GB needed of 32 GB usable — headroom for context
computed roofline: 1792 GB/s × 0.55–0.80 efficiency window / 1.7GB (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 | 1.7GB (GGUF Q4_K_M) |
| minimum memory | 4GB working set |
| this machine | 32GB GDDR7 |
| params | 2B |
| license | apache-2.0 |
| price | $5000 |
easy run: ollama pull moondream:1.8b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ every model the RTX 5090 (32GB) can run — full list
→ RTX PRO 6000 Blackwell (96GB) — upgrade path machine
B&H Photo ↗ · Amazon ↗ · Amazon ↗