Google launches A3 Virtual Machines supercomputer: Nvidia H100 with 26 exaFlops AI performance

Thanks to CTOnews.com netizens Xiao Zhan cut, Mr. Aviation's clue delivery! CTOnews.com, May 11, at the Google I / O conference this morning, Google spent more than an hour talking about its many advances in artificial intelligence, such as the new PaLM 2 large language model (LLM) for its generative AI Bard. This is the basic pillar of adding artificial intelligence features to Google's portfolio, including Google Maps, Google photos, Gmail and so on.

In addition, Google also launched a supercomputer with excellent performance, the A3 Virtual Machines. It uses a fourth-generation Intel Xeon scalable processor, 2TB DDR5-4800 memory, and 8 Invida H100 "Hopper" GPU, which achieves a bisection bandwidth of 3.6 TBps through NVLink 4.0 and NVSwitch.

Google said the new A3 supercomputer was "designed to train and serve the most demanding artificial intelligence models that power today's generation of artificial intelligence and large-scale language model innovation". According to reports, this supercomputer can provide 26 exaFlops of artificial intelligence performance.

According to Google, A3 represents the first production deployment of its GPU-to-GPU data interface, allowing data to be shared at 200Gbps speeds while bypassing the host CPU. This interface, which Google calls the Infrastructure processing Unit (Infrastructure Processing Unit, IPU), can increase the available network bandwidth of the A3 virtual machine (VM) by as much as 10 times that of the A2 virtual machine.

Ian Barker, vice president of ultra large-scale and high-performance computing at Nvidia, said: "Google's A3 virtual machine, powered by the next-generation Nvidia H100 GPU, will accelerate training and services for generative artificial intelligence applications." After Google Cloud recently launched G2 instances, we are proud to continue to work with Google Cloud to help transform global businesses through a specially built artificial intelligence infrastructure. "

If you want to try A3, you can apply to join the early access program through Google A3 preview. But CTOnews.com must remind you here that filling out the form is just an application, and there is no guarantee that you will pass the review and be eligible to use A3.