can it run
6.7 GB vs 6 GB usable — partial CPU offload, expect large speed loss With a q8 (quantized) KV cache it comes to the edge (5.7GB) — possible with modest context, at some quality cost.
6GB · Alchemist · model file 4.7GB · 7B
computed roofline: 192 GB/s × 0.20–0.50 efficiency window / 4.7GB (Q4_K_M) — bands, never points.
→ Tesla M60 (16GB, used) — cheapest catalog machine that runs llava 7B fully ($35)
→ arc-b580 — the next step up from Intel Arc A380
→ every model the Intel Arc A380 can run — full list
→explore the full catalog91 machines indexed · live prices · what each one can run