Local models, real GPUs.
Every test uses the exact same prompt to ensure apples-to-apples comparisons.
Tests are performed on the same hardware rig to eliminate environment variables.
Only one variable is changed per test to isolate the impact of specific changes.
The model is reloaded before every single run to ensure no state leakage.
Every prompt and result file is published for full transparency and verification.
GPU: NVIDIA RTX 3090 24 GB
Server: LM Studio
PC Magik Lab is a dedicated exploration of artificial intelligence performance in a local environment. The series focuses on the practical realities of running LLMs on consumer-grade hardware, moving beyond theoretical benchmarks to real-world execution.
By utilizing local GPUs and open-source tools, the lab aims to provide a clear, verifiable look at how different models and configurations perform under controlled conditions.