can it run
Comfortably. 16GB of memory against a 6GB working set leaves headroom for context, the OS, and anything else you're running.
weights 4.1GB + KV cache 0.5GB at 4k context vs ~16GB usable → full. 4.6 GB needed of 16 GB usable — headroom for context
computed roofline: 320 GB/s × 0.48–0.72 efficiency window / 4.1GB (Q4_K_M) — bands, never points.
| model file | 4.1GB (GGUF Q4_K_M) |
| minimum memory | 6GB working set |
| this machine | 16GB GDDR6, 320 GB/s |
| params | 7B |
| license | apache-2.0 |
| price | $450 |
easy run: ollama pull mistral:7b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ every model the Tesla T4 (16GB, used) can run — full list
→ tesla-v100-16gb — the next step up from this machine's GPU