Local models, real GPUs.
Every test isolates a single variable so you can see exactly what changes performance.
Every run uses the identical hardware setup and the same prompt. The only thing that changes is the single variable being tested.
Each benchmark changes exactly one factor — model, context window, or configuration — to isolate its impact on performance.
The model is reloaded fresh before every single run to eliminate warm-up bias and ensure consistent conditions.
Every prompt and every result file is published openly. No hidden methodology — anyone can reproduce the tests.
One machine. Real consumer hardware. No cloud instances, no VMs.
PC Magik Lab is a YouTube series dedicated to running local AI models on real, consumer-grade GPUs. No cloud instances, no lab servers — just a real desktop pushing real models on real hardware.
Each episode tests a different model or configuration, measuring performance in a transparent, reproducible way. Every prompt and every result file is published openly so anyone can verify the numbers.
The goal is simple: help you understand what your own GPU can do with local AI, and make informed choices about which models work for your setup.
Follow the series on YouTube and explore all prompts and results on GitHub.