can it run
Comfortably. 8GB 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 ~8GB usable → full. 2.3 GB needed of 8 GB usable — headroom for context
computed roofline: 192.3 GB/s × 0.28–0.50 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 | 8GB GDDR5, 192 GB/s |
| params | base |
| license | apache-2.0 |
| price | $100 |
easy run: no stock runtime yet
→ every model the Tesla P4 (8GB, used) can run — full list
→ tesla-p100-16gb — the next step up from this machine's GPU