Nvidia and Amazon Upgrade Project Ceiba AI Supercomputer to Blackwell

Nvidia is fresh off the unveiling of its new Blackwell AI superchip, and it's wasting no time making plans to roll that hardware out. Nvidia and Amazon partnered up last year to build what was to be one of the fastest supercomputers in the world, known as Project Ceiba. Now, the companies have said Project Ceiba will get a Blackwell upgrade to make it up to six times faster than originally envisioned.

The version of Project Ceiba discussed last year was still a beast, featuring more than 16,000 H100 Hopper AI accelerators. Nvidia predicted the machine would have offered 65 exaflops of AI processing power when complete. The current leading supercomputer is the US Department of Energy's Frontier machine, which can hit 1.1 exaFLOPS with thousands of AMD Epyc CPUs and Radeon GPUs.

Switching to Blackwell for Project Ceiba will push performance into the stratosphere. The system is now slated to feature 10,386 Blackwell B200 superchips. Each of those comes with one Grace CPU and two Blackwell chips, for a total of 20,736 GPUs. The image above is Nvidia's vision for a Blacwell-based "AI factory." Project Ceiba could end up looking similar.

Nvidia says this machine could hit an impressive 414 exaFLOPS, but it's not clear how it arrived at that calculation. It's possible Nvidia used an AI-friendly inference test, which makes a direct comparison to traditional supercomputers difficult. The Frontier number comes from Linpack, which is a standard HPC benchmark.

The Blackwell B200 Superchip features two Blackwell GPUs and a Grace CPU. Credit: Nvidia

Nvidia sees Project Ceiba as a necessary part of its quest to build the omniverse. Its goal is to design products with perfect fidelity inside a simulation before any physical components come together. These "digital twins" were a big part of CEO Jensen Huang's GTC keynote this week. At the event, Huang said Nvidia is launching a simulated version of Earth appropriately called Earth II. In the case of Project Ceiba, Nvidia is talking about digital biology, robotics, climate prediction models, self-driving cars, and more, reports PCMag. It can also be used to train new large learning models like ChatGPT.

Project Ceiba hasn't been built yet, so the preliminary upgrade doesn't require any rebuilding. For AI supercomputers that are already live with Nvidia Hopper GPUs, Nvidia makes a single-chip version of Blackwell in a Hopper-like form factor. These cards can be slotted right into Hopper HGX servers.