VRAM guides
These guides use the same calculator as the rest of LocalLM Fit. Every model is evaluated at an 8,192 token context, with published quantization sizes, calculated KV cache, and 1.5 GB runtime overhead.
8 GB class
Eight gigabytes is a strict local-AI tier: small models fit with useful headroom, while most 20B-class packages do not. This guide favors honest compatibility over aspirational model names.
12 GB class
Twelve gigabytes opens the door to compact 8B-class models and a few efficient mixtures of experts, but the difference between a clean fit and a desktop spill is small.
16 GB class
Sixteen gigabytes is the first broadly useful enthusiast tier. It can hold several modern 12B-class models cleanly, while larger 20B to 35B models depend heavily on quantization and context.
24 GB class
Twenty-four gigabytes supports larger dense models and several efficient MoE builds without immediately falling back to CPU memory. It is a practical workstation tier for local experimentation.
32 GB class
Thirty-two gigabytes is a premium consumer tier where quantization quality becomes the main tradeoff. Many models already fit below this capacity, so the additional memory is best spent on context, higher-bit weights, or future model growth.
Dedicated VRAM and Apple unified memory are shown separately in hardware profiles, but both are compared against the same estimated model requirement. A capacity label is not a performance guarantee or a live price comparison.