Running AI models locally — one variable at a time. Every prompt, every result, fully published.
Every test follows a strict methodology so results are reproducible and comparable.
Every model is evaluated with the exact same prompt, ensuring a fair apples-to-apples comparison.
One dedicated hardware setup is used for all tests — no swapping, no cloud, no variables.
Only one variable is changed per test — the model — keeping everything else constant.
The model is fully reloaded before every single run to eliminate caching bias.
Every prompt and every result file is published openly in the project repository.
One rig, one GPU, one model server. No cloud, no tricks.
The person behind the tests.
Computer service operator and homelab enthusiast from Poland. PC Magik Lab is a YouTube series dedicated to running AI models locally on real hardware — measuring performance with transparency and rigor.
Every prompt, every result, every configuration is published so anyone can reproduce the tests.