A YouTube series about running AI models on real, local GPUs.
Local models, real GPUs.
Every benchmark follows a strict protocol so results are fair, transparent, and reproducible.
Every model is tested with identical prompts so there's no ambiguity about what it was asked to do.
All tests run on a single, fixed hardware setup. No swapping parts between benchmarks.
Each test isolates a single variable — usually the model — so differences in results are meaningful.
The model is fully reloaded before every single run to eliminate caching or state carryover.
Every prompt and every result file is published openly so anyone can verify the findings.
One rig. No variables. Just raw local inference.
PC Magik Lab is a YouTube series dedicated to running AI models locally — no cloud, no shortcuts. Each video puts real language models through real prompts on real hardware, so you can see how they actually perform when deployed on your own machine.
All prompts, results, and test files are published openly in the companion repository so you can reproduce every benchmark yourself.
Watch the videos or dig into the data.