Learn the right VRAM for coding models, why an RTX 5090 is optional, and how to cut context cost with K-cache quantization.