Same prompt
Every test starts from the same prompt, so the model — not the question — is what we are comparing.
prompt: fixedLocal models, real GPUs.
A hands-on series about running AI models on real, local GPUs. The same prompt, the same rig — one variable changed per test. Every prompt and every result file published, so you can check the work yourself.
model reloaded before every run — same prompt, same rig, one variable changed
Comparisons only mean something when the conditions are controlled. The lab keeps everything identical except the thing under test.
Every test starts from the same prompt, so the model — not the question — is what we are comparing.
prompt: fixedOne machine, one GPU, one model server. The hardware never changes between tests.
rig: singleOnly one thing changes per test, so the result can be attributed to that change — nothing else.
Δ = 1 variableThe model is reloaded before every run, so no run inherits state from the one before it.
reload: alwaysEvery prompt and every result file is published, so each test can be checked and re-run independently.
published: 100%With the same prompt, the same rig and the same procedure, the setup is built so results stand on their own.
verify: open sourceThe whole series runs on a single, fixed setup. No cloud, no rented inference — just local metal.
one rig · one GPU · one server — the hardware is held constant across the entire series
Mateusz Piekut
PC Magik (Mateusz Piekut) is a computer service and homelab operator from Poland who runs PC Magik Lab — a YouTube series about running AI models on real, local GPUs.
The lab is built around a simple idea: test local models the way a hardware person would. One fixed rig, one fixed procedure, and the results put out in the open — prompts, result files and all.
If you follow the series, the best way to keep up is the YouTube channel, and every prompt and result file lives in the public repository.
Two places hold everything: the channel for the videos, the repository for the prompts and result files.