Dual GPU enables larger local models, not just higher quants

The user found doubling VRAM from 24GB to 48GB primarily unlocked larger models like 70B, with quality gains from higher quants being marginal. Running qwen 27B at Q4-XL vs Q6/Q8 showed little difference in coding tasks.