February 26, 2024

nvidia at the moment introduced the next-generation Nvidia GH200 Grace Hopper platform, a chip powered by HBM3e processors and constructed for the age of accelerated computing and generative Synthetic Intelligence (AI).

The corporate has spent months engaged on merchandise targeted on this know-how, in addition to in depth language fashions and recommender techniques, which just a few months in the past certified that “they’re the digital engines of the trendy financial system”.

An instance of that is the brand new mannequin of the DGX GH200 supercomputer, introduced final Could and which has with 56 GH200 tremendous chips that mix with NVLin and NVLink Swap System interconnect applied sciences to operate as a single GPU.

The {hardware} maker has now shared a brand new platform, the GH200 Grace Hopper, a next-generation chip for the age of accelerated computing and generative AI, that includes a 282 GB HBM3e processor that’s 50 p.c sooner than the included one within the present era.

This platform, “constructed to deal with the world’s most advanced generative AI workloads” and spanning a number of language fashions and vector databases, is obtainable in numerous configurations.

The twin configuration, for its half, gives as much as 3.5 occasions extra reminiscence capability and 3 times extra bandwidth than the present era of platforms, along with comprising a single server of 144 Arm Neoverse cores and eight petaflops of AI efficiency.

Nvidia has commented that the next-generation Grace Hopper Superchip platform is totally compliant with the COMPUTEX MGX server specification and that early distributors are anticipated to combine this platform into their techniques beginning in November. second quarter of 2024.