# LLM Hardware Guide

> PC and GPU builds and VRAM tiers for a local LLM, with every fit number shown as arithmetic from a primary source.

## Primary

- [LLM info](https://llmhardwareguide.com/llm-info): Machine-readable description of this site for an assistant, including the sister Mac site.
- [Start here](https://llmhardwareguide.com/start-here): How to route from this file, read raw markdown, and avoid inventing a spec.

## Key pages

- [LLM Quantization Explained: Q4 vs Q8, GGUF and Memory Math](https://llmhardwareguide.com/llm-quantization-explained): What Q4, Q5, Q8 and FP16 mean, how quantization changes model size, and the exact formula for memory, with sources from llama.cpp and Hugging Face.
- [LLM VRAM Requirements: 7B to 70B+ Models (Formula + Table)](https://llmhardwareguide.com/vram-requirements): How much VRAM do 7B, 14B, 32B and 70B models need at Q4, Q8 and FP16? A sourced table plus the formula, so you can check any model.
- [Best GPU for Local LLMs in 2026: VRAM-First Buying Guide](https://llmhardwareguide.com/best-gpu-for-local-llm): Pick a GPU for local LLMs by VRAM first: which cards hold 8B, 14B, 32B and 70B models, from NVIDIA, AMD and Intel spec sheets with the math shown.
- [24GB vs 16GB vs 12GB VRAM for Local LLMs: What Fits](https://llmhardwareguide.com/24gb-vs-16gb-vs-12gb-vram): What fits on 8, 12, 16 and 24GB GPUs for local LLMs, by arithmetic from model cards, with the cards in each tier from vendor spec pages.
- [RTX 3090 vs 4090 vs 5090 for Local LLMs: Specs and Fit](https://llmhardwareguide.com/rtx-3090-vs-4090-vs-5090-llm): RTX 3090, 4090 and 5090 compared for local LLMs from NVIDIA specs: VRAM, memory interface, power, and which model sizes fit on each.
- [Local LLM PC Build Guide: Parts, VRAM and Power (2026)](https://llmhardwareguide.com/local-llm-pc-build): Plan a PC for local LLMs: GPU VRAM first, then system RAM, PSU, motherboard lanes and storage. Parts reasoning from vendor specs, no fake benchmarks.
- [Run an LLM on CPU Only: RAM Offload and GPU Split Explained](https://llmhardwareguide.com/run-llm-on-cpu): Yes, LLMs run on CPU and system RAM. How CPU-only and partial GPU offload work in llama.cpp and Ollama, what limits them, and how to size RAM.
- [Methodology](https://llmhardwareguide.com/methodology): How this site cites GPU specs and model cards, shows VRAM fit math, and leaves a missing figure as unknown.
- [About LLM Hardware Guide](https://llmhardwareguide.com/about): What llmhardwareguide.com covers: PC and GPU builds and VRAM fit math for local LLMs, and where the Mac site takes over.

## About

- Owner: llmhardwareguide.com
- Standard followed: https://llmbestpractices.com
- Updated: 2026-10-12
- Sister site: https://llmformac.com
- Sister index: https://llmformac.com/llms.txt
- IndexNow key: https://llmhardwareguide.com/c14ebba93113262f105b01c742a2f94b.txt