TL;DR

Linus wrangled Nvidia’s $30K H200 “graphics” card—really a powerhouse generative-AI/HPC accelerator—because nobody would send it to him. He cracked it open, poked around the PCB, then ran it through paces: comparing raw AI/training chops against a dual-EPYC 9965 server and a flagship RTX 5090.

The verdict? It absolutely demolishes consumer GPUs on AI workloads (and holds its own against CPU clusters), but at that price it’s strictly for data centers and labs. If you’re into bleeding-edge AI performance (and have tens of thousands to burn), the H200 is a serious contender.

Watch on YouTube