can it run
Yes, but close to the edge. 8GB against a 6GB working set: it runs, with little room for long context windows or other processes.
weights 4.7GB + KV cache 2GB at 4k context vs ~6GB usable → offload-partial. 6.7 GB vs 6 GB usable — partial CPU offload, expect large speed loss
computed roofline: 68 GB/s × 0.25–0.65 efficiency window / 4.7GB (Q4_K_M) — bands, never points.
| model file | 4.7GB (GGUF Q4_K_M) |
| minimum memory | 6GB working set |
| this machine | 8GB unified |
| params | 7B |
| license | apache-2.0 |
| price | $499 |
easy run: ollama pull llava:7b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ Tesla P100 (16GB, used) — cheapest machine that runs it ($135)
→ every model the Jetson Orin Nano Super dev kit can run — full list