Local models, real GPUs.
Every test follows a strict, repeatable protocol so you can trust the results.
The identical prompt is used across all runs, ensuring a fair and apples-to-apples comparison between models and configurations.
All tests are run on the same hardware. Only one variable is changed per test — the model or a single parameter — so the impact is clear.
Each model is freshly loaded before every single test to eliminate cache effects and guarantee consistent starting conditions.
All prompts and result files are open and accessible so anyone can verify, replicate, or extend the tests.
One machine. Real consumer hardware. No cloud instances, no lab-grade equipment.
24 GB VRAM
Every test in this series is run on one machine — a local PC equipped with an NVIDIA RTX 3090 with 24 GB of VRAM, using LM Studio as the model server.
No cloud VMs, no lab-grade servers. Just a consumer GPU running real models with real prompts, the way you would at home.
PC Magik Lab is a YouTube series dedicated to testing and comparing local AI models on real, consumer-grade GPUs.
PC Magik Lab is a YouTube series that puts AI models through their paces on real, local GPUs — the kind you'd have in your own home lab.
Every episode follows a strict methodology: the same prompt, the same rig, one variable changed per test. Models are reloaded before every run, and every prompt and result file is published for full transparency.
The goal is simple: honest, reproducible benchmarks that anyone can verify and build on. No cloud instances, no lab-grade equipment — just consumer hardware and real results.
Watch episodes on YouTube or browse the repository with all prompts and result files.