Popular open vision-language model for image Q&A and captioning.
visionchat · sizes: 7B, 13B · Apache-2.0
| machine | best fit | verdict | speed |
|---|---|---|---|
| Apple M1 · 8GB | 13B Q4_K_M | ❌ Won't fit | — |
| Apple M1 Pro · 16GB | 13B Q4_K_M | 🐢 CPU-only (slow) | 0.9–2 tok/s |
| Apple M2 · 16GB | 13B Q4_K_M | 🐢 CPU-only (slow) | 0.5–0.8 tok/s |
| Apple M3 Pro · 18GB | 7B Q4_K_M | ✅ Runs (tight) | 10–16 tok/s |
| Apple M2 Pro · 32GB | 13B Q5_K_M | ✅ Runs (tight) | 7–11 tok/s |
| Apple M4 Pro · 48GB | 13B Q8_0 | ✅ Runs comfortably | 7–12 tok/s |
| Apple M3 Max · 36GB | 13B Q8_0 | ✅ Runs (tight) | 9–14 tok/s |
| Apple M2 Max · 64GB | 13B Q8_0 | ✅ Runs comfortably | 9–15 tok/s |
| Apple M4 Max · 128GB | 13B Q8_0 | ✅ Runs comfortably | 11–19 tok/s |
| Apple M2 Ultra · 192GB | 13B Q8_0 | ✅ Runs comfortably | 12–20 tok/s |
| NVIDIA GeForce RTX 3060 · 12GB VRAM | 7B Q5_K_M | ✅ Runs (tight) | 16–25 tok/s |
| NVIDIA GeForce RTX 4070 · 12GB VRAM | 7B Q5_K_M | ✅ Runs (tight) | 20–35 tok/s |
| NVIDIA GeForce RTX 4090 · 24GB VRAM | 13B Q8_0 | ✅ Runs (tight) | 20–35 tok/s |
| NVIDIA GeForce RTX 5080 · 16GB VRAM | 13B Q4_K_M | ✅ Runs (tight) | 25–45 tok/s |
| NVIDIA GeForce RTX 5090 · 32GB VRAM | 13B Q8_0 | ✅ Runs comfortably | 35–60 tok/s |
| AMD Radeon RX 7900 XTX · 24GB VRAM | 13B Q8_0 | ✅ Runs (tight) | 18–30 tok/s |
Shopping for a machine? What hardware do I need to run LLaVA? →
Check your exact rig: open the checker →
2 models share LLaVA's KV-cache geometry — they size memory the same way, so a rig that fits one tends to fit its same-size kin.
Explore the whole catalog by architecture: open the graph →