can it run
Comfortably. 16GB of memory against a 8GB working set leaves headroom for context, the OS, and anything else you're running.
weights 6.2GB + KV cache 2GB at 4k context vs ~16GB usable → full. 8.2 GB needed of 16 GB usable — headroom for context
computed roofline: 160.4 GB/s × 0.28–0.50 efficiency window / 6.2GB (Q4_K_M) — bands, never points.
| model file | 6.2GB (GGUF Q4_K_M) |
| minimum memory | 8GB working set |
| this machine | 16GB GDDR5 (2×8 GB), 160 GB/s |
| params | large-v3 |
| license | apache-2.0 |
| price | $35 |
easy run: no stock runtime yet
→ every model the Tesla M60 (16GB, used) can run — full list
→ tesla-p100-16gb — the next step up from this machine's GPU