Nvidia unveils 1 exaflop supercomputer for productive AI

Nvidia unveiled its DGX GH200 AI supercomputer on Sunday, 28. powerful machine with 1 exaflop per second performanceprovides tech giants with tools to develop competitors for ChatGPT. This is one of the brand’s innovations at Computex 2023, which will open Tuesday (30) in Taipei (Taiwan).

To perform 1 quintillion operations per second, The platform features 256 Grace Hopper superchips, a combination that allows it to operate as a single GPU.. Additionally, the combination of components provides 144TB of shared memory, equivalent to approximately 500x more memory than the previous generation.

Combining an Nvidia Grace CPU based on the Arm architecture with an Nvidia H100 Tensor Core GPU, the machine also uses NVLink-C2C technology to connect superchips. According to the manufacturer, method increases the bandwidth between CPU and GPU by up to seven times and even reduces power consumption.

all this power new supercomputer Nvidia DGX GH200 AI It will be made available to customers of the US-based manufacturer to develop their next generation of productive artificial intelligence (AI) engines. The idea is to tackle the very successful OpenAI chatbot.

first customers

Microsoft, Meta and Google will be the first companies to test the powerful performance of the DGX GH200 AI., according to Nvidia, in developing more advanced models of generative AI. But the optimized performance of the machine will also be used in other segments such as cloud services.

“The new NVLink scale and shared memory of Grace Hopper Superchips addresses key bottlenecks in large-scale AI, and we look forward to exploring its capabilities for Google Cloud and our prolific AI initiatives,” said Mark Lohmeyer, Vice President of Computing Cloud at Google.

The Nvidia DGX GH200 AI supercomputer should be available by the end of this year when the actual performance of the device will be known.