Methodology
Controlled Variables
We ensure consistent results by using the same prompt and the same hardware rig for every single test.
One Variable at a Time
Only one parameter or model is changed per test, allowing for precise measurement of impact.
Fresh Environment
The model is reloaded from scratch before every run to prevent caching biases.
Full Transparency
Every prompt used and every result file generated is published for community inspection.
Hardware Rig
The Setup
NVIDIA RTX 3090
24 GB VRAM
LM Studio
Model Server
About the Series
The Goal
Exploring the real-world performance of local AI models on consumer-grade, high-VRAM hardware.
The Focus
Quantifying how different models, quantization levels, and prompts perform in local environments.
The Data
Open results and reproducible workflows for the AI enthusiast and developer community.