RTX 5080/3090 Local LLM Rig

TL;DR

A hardware configuration for running local LLMs at speeds that make cloud providers look like they are on dial-up.

Who is this actually for?

Self-hosting nerds and privacy-focused developers who want to stop paying API credits and run 27B models locally.

The Good

• 80 tokens per second on a 27B model is impressively fast for a home setup.
• Smartly leverages the 24GB VRAM of the older 3090 alongside the raw speed of the 5080.

The Catch (Potential Downsides)

Your electricity meter will spin fast enough to take flight during long sessions. Managing mixed-generation GPU drivers and thermal throttling in a single case is usually a total headache.