Google announces AI specialized supercomputer 'A3' equipped with NVIDIA H100 GPU



At the developer event ``

Google I / O '' on May 11, 2023, Google announced the supercomputer `` A3 '' specializing in AI training and operation.

Introducing A3 supercomputers with NVIDIA H100 GPUs | Google Cloud Blog
https://cloud.google.com/blog/products/compute/introducing-a3-supercomputers-with-nvidia-h100-gpus/?hl=en



Google Launches AI Supercomputer Powered by Nvidia H100 GPUs | Tom's Hardware

https://www.tomshardware.com/news/google-a3-supercomputer-h100-googleio



At this conference, Google announced a large-scale language model 'PaLM 2' that strengthens the company's chat AI Bard. This is the foundation for adding AI-infused features across Google's products, including Google Maps, Google Photos and Gmail.

Google announces large-scale language model 'PaLM 2', already installed in 25 Google services - GIGAZINE



As these capabilities become more prevalent, there will be an enormous number of users accessing vast AI applications, requiring significant processing power in the cloud. The focus is on Google's newly announced supercomputer, the A3 GPU VM. According to the announcement, the A3 supercomputer is equipped with eight H100 GPUs that use NVIDIA's Hopper architecture, and the throughput is three times higher than before.

At its core, it uses 2TB of DDR5-4800 memory and Intel's 4th generation Xeon Scalable processors, and each of the eight GPUs is connected with NVIDIA's NVSwitch and NVLink 4.0 with a bisection bandwidth of 3.6TB per second. increase.

With this purpose-built machine learning-specific design, A3 delivers up to 26 EFLOPS of AI performance, significantly reducing the time and cost of training large models.



According to Google, the A3 will be the first product to introduce a production-level GPU-to-GPU data interface that can bypass the host CPU and share data at 200Gbps. With this interface, called IPU (Infrastructure Processing Unit), A3's VMs can utilize 10 times more network bandwidth than A2 .

'Google Cloud's A3 VMs, powered by next-generation NVIDIA H100 GPUs, accelerate the training and deployment of generative AI applications,' said Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA. However, following Google Cloud's recent launch of G2 instances, we are proud to work with Google Cloud to help transform enterprises around the world with fit-for-purpose AI infrastructure.'

in Software,   Hardware, Posted by log1l_ks