Guides · 7
Local LLM guides
The concepts behind the numbers, in plain words. Every figure here is the same one the calculator uses, with sources. Validated 2026-09-14.
- How local LLMs work
The basics: weights, memory, quantization and which runtime to use.
- Quantization explained
What Q4_K_M, Q8_0 and FP16 mean, and which to pick.
- VRAM vs RAM vs unified memory
The three kinds of memory a model runs in, and why the gigabytes are not equal.
- Mixture of experts
Why a 30B MoE runs at 3B speed but needs 30B of memory.
- Which LLMs fit in 16, 24 or 32GB
The counted answer per memory tier, from measured GGUF sizes.
- Is 6GB of VRAM enough?
What a 6GB card really runs, where 7B breaks, and what 8GB adds.
- Can you run Claude locally?
What Claude Code can do with a local Ollama model, and what it cannot make local.
Ready to check your hardware? Browse by memory →