can it run
Comfortably. 16GB of memory against a 4GB working set leaves headroom for context, the OS, and anything else you're running.
weights 0.3GB + KV cache 2GB at 4k context vs ~16GB usable → full. 2.3 GB needed of 16 GB usable — headroom for context
computed roofline: 320 GB/s × 0.48–0.72 efficiency window / 0.3GB (Q4_K_M) — bands, never points.
| model file | 0.3GB (GGUF Q4_K_M) |
| minimum memory | 4GB working set |
| this machine | 16GB GDDR6, 320 GB/s |
| params | 82M |
| license | apache-2.0 |
| price | $450 |
easy run: no stock runtime yet
→ every model the Tesla T4 (16GB, used) can run — full list
→ tesla-v100-16gb — the next step up from this machine's GPU