PC Magik Lab
Local models, real GPUs.
A YouTube series about running AI models on real, local GPUs — measured honestly, one variable at a time.
How the series measures things
Fair, repeatable tests. Same prompt, same rig, one variable changed per test — so every result actually means something.
Same prompt, same rig
Every run starts from an identical setup so the only thing that can change is what's being tested.
One variable per test
A single setting changes between runs. No guessing which change caused the difference in results.
Model reloaded every run
The model is reloaded before every test, so nothing leaks between runs and each measurement stands on its own.
Everything published
Every prompt and every result file is published, so the work can be checked and repeated by anyone.
One rig. Real hardware.
No cloud abstractions or hidden specs — just a single local machine running the models under test.
NVIDIA RTX 3090
24 GB of VRAM powers every local run on the single test rig.
LM Studio
Used as the model server that serves every model under test.
A series for people who run their own models
PC Magik Lab is a YouTube series about running AI models on real, local GPUs. No cloud fog machines — just a single rig and honest numbers.
The focus is on method: keep everything the same except one variable, reload the model before every run, and publish every prompt and result so anyone can verify what's shown.
If you care about reproducibility and want to know how your models actually behave on local hardware, this series is for you.
Local models, real GPUs.— the PC Magik Lab promise
Watch and dig in
Follow the series on YouTube and explore the published prompts and result files.