The value pick
A used RTX 3090: 24GB of VRAM for the price of a phone. The secondhand door into real local AI.
The commenters are right: at $3k there are two honest answers. The Mac is silent and simple; the 3090 is raw memory per dollar. A used 3090 still holds 24GB of VRAM, which was flagship-class in 2020 and is still enough for the 27B class, most decision models, and image generation today. This build is for the person who enjoys the hunt: buy used, save $700, and spend it on electricity you will not use.
You like the idea of a used pickup: not pretty, hauls anything. You are comfortable buying a used GPU with a bit of research (test it under load the day it arrives), and you want maximum VRAM per dollar. Gamers upgrading from a 2080Ti: this card is a sidegrade for games and a massive upgrade for AI.
first check: next Monday
The parts
What it runs
The 300B+ MoE class (24GB of VRAM is 24GB). And used cards have no warranty: buy from sellers with return policies, test with a memory stress test the day it arrives, and accept that the card has history.
About 350W under AI load and it will be heard: the 3090 is a loud card, and this is the main trade against the Mac. A decent 850W PSU and a case with actual airflow are part of the build.
This is the tier where local AI becomes boring, which is the highest compliment. The box is silent, the answers are fast, and the models are big enough that you stop asking what a model can do and start asking what you want done. Your writing assistant, your code reviewer, your document librarian, your form-filler: all local, all day, on hardware that uses less power than the monitor next to it.
See the $5k tier →