Independent by design

Local AI Frontier exists because every "best GPU for local AI" list I read was either untested, undated, or quietly paid for. I wanted a source that showed its work — so I built one.

Who runs this

My name is Edgar. I'm the product owner, lab lead, and final publisher. AI agents help with research, drafting, and engineering, but every published benchmark is tested first-hand, reviewed, and signed off by me. No AI agent publishes, purchases, or makes external commitments on the site's behalf. Read the full author bio →

What's in the lab

The test fleet currently includes 4 named machines, covering NVIDIA, AMD, and Intel Arc consumer hardware. Every machine is documented in the open — no anonymized "test rig."

• Jitori PC — Codex Librarian: Intel Arc B60 Pro (24GB VRAM), vulkan backend.
• Ray — Marshal Alpha + Marshal Beta: 2× AMD Radeon AI PRO R9700 (64GB VRAM), vulkan backend.
• EvoX2 — Architect + Router: AMD Radeon RX 7900 XT (gfx1100, 20GB) + Ryzen AI MAX+ 395 iGPU (gfx1151, unified memory) (20GB VRAM), vulkan backend.
• Victor (msi-command) — NVIDIA CUDA reference (mobile): NVIDIA GeForce RTX 5070 + RTX 5070 Ti Laptop (Blackwell sm_120) (24GB VRAM), cuda backend.

Why independent

We don't accept payment for placement. We do have commercial interests — we own MoreYess, and some hardware links may be affiliate — but they are fully disclosed and structurally separated from benchmark results. The firewall is real: read it.

What we publish

• Benchmark reports — 12 and counting, each with raw data and limitations.
• Model Fit — a transparent estimator with the math shown.
• Tested builds — complete rigs by budget and workload.
• Field guides — decision-focused explainers.

Cadence

We aim for at least one new evidence-led report per week. Subscribe to the Frontier Dispatch to get it first.

Contact

Email lab@localaifrontier.com with questions, correction reports, or partnership inquiries. Note: we do not accept paid placement in benchmarks.