How We Measure Things
Same Prompt, Every Time
The exact same prompt is used across every run. No variations, no improvisation — just consistent input for fair comparison.
Same Rig, Every Time
All benchmarks run on the same physical hardware. No cloud instances, no virtual machines — just real local GPUs.
One Variable Changed
Each test changes exactly one variable at a time, so every result maps cleanly to its cause.
Fresh Model Reload
The model is reloaded fresh before every run, eliminating cache artifacts and warmup bias.
Published Results
Every prompt and every result file is published in full. Complete transparency, open for review.
The Rig
NVIDIA RTX 3090
24 GB VRAM — Single Rig Setup
- GPU NVIDIA GeForce RTX 3090
- VRAM 24 GB GDDR6X
- Server LM Studio
- Type Local, physical hardware
What This Is
PC Magik Lab is a YouTube series dedicated to running AI language models locally on real, physical GPUs. No cloud abstractions, no serverless APIs — just raw hardware producing measurable output. Every benchmark follows the same rigorous method: one variable changed per test, fresh model reloads, and complete transparency with every prompt and result file published in full.