GPU
NVIDIA RTX 4060 Ti 16 GB — cheapest card in the database with at least 8 GB.
AI Stack Builder
What should I actually build? Pick a VRAM tier, your OS and what you want to run — and get a matched GPU, model and app.
Model
Llama 3.1 8B Instruct — best rated model that fits 8 GB.
Application
llama.cpp — runs on windows, experience level: advanced.
Rough estimates (Q4 quantization, ~1.5 GB overhead). Treat this as a starting point, then verify with the VRAM Calculator.