TL;DR

Linus Sebastian finally got his hands on Nvidia’s elusive H200 NVL—a $30,000 generative AI and high-performance computing GPU Nvidia wouldn’t normally send reviewers. After an epic teardown (complete with multi-layer heatsink and HBM3 stacks), he pits the H200 against a dual AMD Epyc 9965 server and a flagship RTX 5090, walking us through power draw, thermal chops, and raw compute throughput.

Benchmarks show the H200 absolutely crushes AI workloads and slimly edges out data-center CPUs, though it’s a tough sell on price-per-watt compared to mainstream GPUs. In short: the H200 is a monster for enterprise AI, but its eye-watering cost keeps it firmly in the datacenter realm.

Watch on YouTube