Hardcore Mod: Water-Cooled Quad 7900 XTX Delivers 96GB VRAM for Local LLMs

tictacturkey · reddit · 2026-08-07

A Reddit user shared a hardcore custom PC build designed specifically for local LLM inference. By utilizing a custom water-cooling loop, the builder combined four AMD Radeon 7900 XTX GPUs to achieve a massive 96GB of VRAM. This setup breaks single-GPU memory limits, offering a cost-effective hardware reference for running large AI models locally.

Original post →

More from Infra

Infra channel →