A YouTube series · Local AI

PC Magik Lab

Local models, real GPUs.

A hands-on series about running AI models on real, local GPUs — no cloud, no abstractions. One rig, one method, and every prompt and result file published so you can see exactly what happens when the model meets the hardware.

How it's measured

One method, every time.

Comparing models only makes sense when the test itself is under control. That's the rule of this series.

Same prompt

Every model is tested with the same prompt. The input never changes — only the model under test.

Same rig

One machine, one GPU. The hardware stays fixed so results are actually comparable across episodes.

One variable per test

Only one variable is changed at a time, so the effect of the change is what you're actually seeing.

Model reloaded before every run

The model is reloaded before every run, so no leftover state skews the result.

Every prompt and result file published

Nothing is hidden. Every prompt and result file is published in the repository, so you can check the work for yourself.

The rig

One rig. No excuses.

Every test in the series runs on the same machine, served by the same local model server.

Hardware

NVIDIA RTX 3090

24 GB of VRAM — the single GPU behind every run in the series.

  • GPU · NVIDIA RTX 3090
  • VRAM · 24 GB
  • One rig · same machine every episode
About

Who's behind the lab.

PC Magik (Mateusz Piekut) is a computer service and homelab operator from Poland who turned his everyday lab into a testing ground for local AI.

PC Magik Lab is his YouTube series about running AI models on real, local GPUs. The premise is simple: local models, real GPUs — and a method that makes the comparisons honest, with every prompt and result file published in the open.

local AI real GPUs computer service homelab Poland