Local models, real GPUs.
PC Magik Lab is a YouTube series that measures how AI models actually perform on a single, home-built rig — no cloud, no hype, just numbers you can trust.
Every test is designed so the only thing that changes between runs is the one variable being measured. Everything else stays locked.
Every model is asked the exact same question, so responses are compared apples to apples.
All runs happen on one home-built rig, so hardware never skews the comparison.
A single factor changes between runs, so every result can be traced to a clear cause.
The model is reloaded before each test, so nothing leaks between runs.
Every prompt and every result file is published, so the work can be checked by anyone.
All benchmarking happens on a single home-built setup — no cloud providers, no remote servers.
24 GB of VRAM — the muscle behind every run in the series.
The local model server that loads and serves every model under test.
A single rig, one GPU, one model server — so the only thing that ever changes between tests is the variable being measured.
PC Magik Lab is a YouTube series about running AI models on real, local GPUs. The goal is simple: cut through the marketing and show how models behave on a genuine home setup.
The whole approach rests on a strict method — same prompt, same rig, one variable changed per test — with every prompt and result file published for anyone to verify.
Follow the series on YouTube and dig into every prompt and result file in the repository.