Wrangling Datacenter GPUs Into a Desktop

A developer named Andrew is demonstrating how older datacenter GPUs can find new life in desktop systems. Hackaday reports he's using Tesla V100 graphics cards from twenty seventeen — hardware that might seem obsolete by today's standards — to run large language models like Gemma and Qwen completely locally. By building a custom PCI card to host two V100s, he's achieved competitive performance with online AI services while keeping inference private and local. The takeaway: as more datacenter hardware hits the secondhand market, companies relying on cloud AI may face competition from users who simply run models at home.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton