can it run
Comfortably. 16GB of memory against a 5GB working set leaves headroom for context, the OS, and anything else you're running.
weights 3.3GB + KV cache 0.53GB at 4k context vs ~16GB usable → full. 3.8 GB needed of 16 GB usable — headroom for context
computed roofline: 900 GB/s × 0.46–0.76 efficiency window / 3.3GB (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 | 3.3GB (GGUF Q4_K_M) |
| minimum memory | 5GB working set |
| this machine | 16GB HBM2, 900 GB/s |
| params | 4B |
| license | gemma |
| price | $340 |
easy run: ollama pull gemma3:4b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ every model the Tesla V100 (16GB, used) can run — full list
→ RTX 3090 (24GB, used) — upgrade path machine