Someone Squeezed 6 AI Chips Into a Regular PC Tower

A user on a tech forum posted photos and specs of a home-built AI server: a standard tower case stuffed with six Nvidia V100 graphics chips, plus an AMD server processor and a specialized motherboard. Normally, people building machines like this use open metal racks or rack-mounted server cases, because they are easier to cool and wire. This person deliberately chose a regular desktop tower because racks are huge and ugly, and they wanted the machine to look at home in a normal room.

Why does this matter to someone who doesn't build computers? Because those V100 chips are the same class of hardware that trains and runs AI systems inside corporate data centers. They are older now — released back in 2017 — so they show up on the secondhand market for a few hundred dollars each. That means a motivated hobbyist can assemble a genuinely capable AI machine for roughly the price of a decent used car, rather than the millions a company would spend.

The plan is to run Qwen, a free and open AI model from China, using a setup that splits the work across multiple chips at once — the abbreviations "TP2" and "PP3" just describe how the model gets divided up. Running a large AI model locally means your data never leaves your house, and you don't pay per-question fees to a cloud company.

Here's the honest catch. Six of these chips draw enormous electricity — likely well over 1,500 watts under load — and produce serious heat and fan noise. That's more power than a space heater running full blast. Most home circuits can't handle it without an electrician, and the electricity bill adds up fast. Also, older chips support fewer modern AI tricks than today's hardware. So this is a glimpse of the future, not a casual weekend project for most people.