Same prompt
Every test is driven by the same prompt, so what you are comparing is the model — not the question.
YouTube series
PC Magik Lab
Local models, real GPUs.
A hands-on series about running AI models on real, local GPUs — measured the same way every time, on the same machine, with the prompts and result files published in the open.
RTX 3090 · 24 GB
Local
VRAM24 GB
ServerLM Studio
LocationLocal
Same rig, every episode
Method
How the series measures things
Every test follows the same rules, so results are comparable across models and episodes.
Same rig
One machine does all the work. There is no swapping hardware between tests to make a model look better.
One variable per test
One variable is changed per test — typically the model — while everything else stays fixed.
Model reloaded before every run
The model is reloaded before every run, so no run benefits from leftovers of a previous one.
Everything published
Every prompt and result file is published in the repository, so you can inspect exactly what happened.
The rig
Hardware
One rig, one model server. The whole series runs on the same setup.
NVIDIA RTX 3090
24 GB VRAM
The graphics card behind every run. Its 24 GB of VRAM is the ceiling for how big a local model can fit in memory.
LM Studio
Serves the models locally
LM Studio acts as the model server — the same server is used across the series, keeping the serving stack constant while models change.
There is only one rig in the lab — the RTX 3090 with LM Studio serving the models. No second machine, no cloud.
About
Who runs the lab
PC Magik (Mateusz Piekut) is a computer service and homelab operator from Poland. PC Magik Lab is his place to put local AI models to work on real hardware, measure them consistently, and publish everything that goes into the tests.