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 · Turing TU116 · model file 4.7GB · 7B
computed roofline: 336 GB/s × 0.48–0.72 efficiency window / 4.7GB (Q4_K_M) — bands, never points.
→ Raspberry Pi 5 + AI Kit (Hailo-8L) — cheapest catalog machine that runs llava 7B fully ($110)
→ rtx-5060-ti-16 — the next step up from NVIDIA GTX 1660 Super
→ every model the NVIDIA GTX 1660 Super can run — full list
→explore the full catalog84 machines indexed · live prices · what each one can run