About

The AI Desk

Independent reviews of hardware for running AI locally — GPUs, Mini PCs, and AI accessories. We focus on real-world performance for LLM inference, Stable Diffusion, and home AI server workloads.

Review Methodology

We evaluate hardware on five criteria: memory capacity (VRAM or unified memory), memory bandwidth (tokens/second throughput), power efficiency (performance per watt), software ecosystem compatibility (CUDA, ROCm, Metal), and value for money relative to the use case.

Benchmark figures are based on publicly available data, manufacturer specs, and community-tested results from llama.cpp, Ollama, and ComfyUI benchmarking threads. We do not receive hardware from manufacturers for review.

For the complete benchmark protocol, test conditions, and data sources, see our full methodology page →

Affiliate Disclosure

As an Amazon Associate I earn from qualifying purchases.

We earn a commission when you click our product links and make a purchase on Amazon, at no additional cost to you. This commission funds our independent research. We are not paid by manufacturers for positive reviews — editorial opinions are our own.

Who Writes This

Reviews and benchmarks are written by Alex Voss, an independent hardware reviewer specializing in local AI inference since 2022.

FTC Compliance

In compliance with FTC guidelines, all affiliate links are marked with (paid link) adjacent to the link. A site-wide disclosure banner appears on every page. See our Privacy Policy and Terms of Service for full details.