NoCodeAbode

Local AI tested on a normal gaming PC. Same six real jobs, every model, graded.

I'm Jon DiPilato. My test PC is an RTX 4070 SUPER with 12GB of VRAM and 32GB of RAM, on Windows. Every model gets the same six jobs, an invoice page fix, a calculator, a snake game and three writing jobs, and gets graded on what it actually finished. Not benchmarks.

What you'll see here:
• llama.cpp, Strata and Ollama on 12GB + 32GB, real speeds and memory limits
• Can a local model drive Claude Code, tested on real files
• Qwen, Flint, NVIDIA Bolt, GLM, DeepSeek and whatever drops next
• The misses too: the crashes, the out-of-memory errors, the loops

New test every Tuesday and Friday.

All glory to God and Jesus Christ. He is my everything, and my faith shapes how I work, including being honest about what these tools can and cannot do.

Subscribe and tell me what to test next.

Book a strategy call: https://cal.com/jon-dipilato/30min
Done-for-you systems: https://dipilatoautomations.com