# LLM Hardware - llmhardware.io

> Find the perfect hardware for running local LLMs. Match any Hugging Face model to optimal hardware with specs, benchmarks, and pricing.

## About

LLM Hardware is a free tool that helps users find the right hardware (GPUs, Apple Silicon systems, PC components) for running large language models locally. It calculates memory requirements based on model parameters and quantization level, then recommends compatible hardware ranked by value, performance, and budget.

## How It Works

1. Select a model (from Hugging Face) or specify parameters
2. Choose quantization level (Q4, Q8, FP16)
3. Get hardware recommendations ranked by: Best Value, Best Performance, Budget Pick

Memory formula: Model memory ≈ parameters × bytes_per_param × overhead
- Q4: ~0.53 GB per billion parameters
- Q8: ~1.07 GB per billion parameters
- FP16: ~2.13 GB per billion parameters

## Key Pages

- Homepage: https://llmhardware.io/
- Hardware Recommender: https://llmhardware.io/recommend
- Browse All Hardware: https://llmhardware.io/hardware
- Browse All Models: https://llmhardware.io/models
- About: https://llmhardware.io/about

## Hardware Database

- Mac Mini M4 (16GB): $599, 16GB unified, 120 GB/s bandwidth, max 28B params at Q4, tier: budget → https://llmhardware.io/hardware/mac-mini-m4-16gb
- Mac Mini M4 (32GB): $799, 32GB unified, 120 GB/s bandwidth, max 60B params at Q4, tier: budget → https://llmhardware.io/hardware/mac-mini-m4-32gb
- Mac Mini M4 Pro (24GB): $1399, 24GB unified, 273 GB/s bandwidth, max 44B params at Q4, tier: mid → https://llmhardware.io/hardware/mac-mini-m4-pro-24gb
- Mac Mini M4 Pro (48GB): $1599, 48GB unified, 273 GB/s bandwidth, max 92B params at Q4, tier: mid → https://llmhardware.io/hardware/mac-mini-m4-pro-48gb
- Mac Studio M2 Ultra (64GB): $3999, 64GB unified, 800 GB/s bandwidth, max 124B params at Q4, tier: high → https://llmhardware.io/hardware/mac-studio-m2-ultra-64gb
- Mac Studio M2 Ultra (192GB): $5999, 192GB unified, 800 GB/s bandwidth, max 380B params at Q4, tier: ultra → https://llmhardware.io/hardware/mac-studio-m2-ultra-192gb
- Mac Studio M4 Ultra (192GB): $6999, 192GB unified, 960 GB/s bandwidth, max 380B params at Q4, tier: ultra → https://llmhardware.io/hardware/mac-studio-m4-ultra-192gb
- NVIDIA GeForce RTX 4060 Ti 16GB: $449, 16GB gddr6x, 288 GB/s bandwidth, max 28B params at Q4, tier: budget → https://llmhardware.io/hardware/rtx-4060-ti-16gb
- NVIDIA GeForce RTX 4070 Ti Super 16GB: $799, 16GB gddr6x, 672 GB/s bandwidth, max 28B params at Q4, tier: mid → https://llmhardware.io/hardware/rtx-4070-ti-super-16gb
- NVIDIA GeForce RTX 4080 Super 16GB: $999, 16GB gddr6x, 736 GB/s bandwidth, max 28B params at Q4, tier: high → https://llmhardware.io/hardware/rtx-4080-super-16gb
- NVIDIA GeForce RTX 4090 (24GB): $1599, 24GB gddr6x, 1008 GB/s bandwidth, max 44B params at Q4, tier: high → https://llmhardware.io/hardware/rtx-4090-24gb
- NVIDIA GeForce RTX 5090 (32GB): $1999, 32GB gddr7, 1792 GB/s bandwidth, max 60B params at Q4, tier: ultra → https://llmhardware.io/hardware/rtx-5090-32gb
- AMD Ryzen 9 7950X3D: $549, 0GB ddr5, 83 GB/s bandwidth, max 0B params at Q4, tier: high → https://llmhardware.io/hardware/ryzen-9-7950x3d
- 128GB DDR5-6000 Kit (4x32GB): $299, 128GB ddr5, 96 GB/s bandwidth, max 0B params at Q4, tier: high → https://llmhardware.io/hardware/ddr5-128gb-6000
- Samsung 990 Pro 2TB NVMe SSD: $179, 0GB ddr5, 7 GB/s bandwidth, max 0B params at Q4, tier: high → https://llmhardware.io/hardware/samsung-990-pro-2tb
- Corsair RM1000x 1000W PSU: $189, 0GB ddr5, 0 GB/s bandwidth, max 0B params at Q4, tier: high → https://llmhardware.io/hardware/corsair-rm1000x

## Supported Models



## API-Like Usage

To get recommendations for a specific model, visit:
https://llmhardware.io/recommend?model={huggingface_model_id}

To filter by use case:
https://llmhardware.io/recommend?useCase={coding|reasoning|image_generation|video_generation|speech|general}

## Categories

- Apple Silicon: Mac Mini, MacBook Pro, Mac Studio, Mac Pro with M4/M4 Pro/M4 Max/M4 Ultra chips
- NVIDIA GPUs: RTX 4060 through RTX 5090, consumer and prosumer cards
- PC Components: RAM, storage, and other build components

## Contact

hello@llmhardware.io