Mak­ing the cloud smarter

NVIDIA and Mi­crosoft to boost AI cloud com­put­ing with new Hyper­scale GPU ac­cel­er­a­tor.

HWM (Malaysia) - - REPORT -

NVIDIA and Mi­crosoft re­cently un­veiled its blue­print to cre­ate a new hyper­scale GPU ac­cel­er­a­tor that would en­able both com­pa­nies to drive AI cloud com­put­ing.

Known as the HGX-1, the new Hyper­scale ac­cel­er­a­tor is an open-source de­sign re­leased in con­junc­tion with Mi­crosoft’s Project Olym­pus.

“AI is a new com­put­ing model that re­quires a new ar­chi­tec­ture,” Jen-Hsun Huang, CEO of NVIDIA said about the com­pany’s col­lab­o­ra­tion with Mi­crosoft. “The HGX-1 hyper­scale GPU ac­cel­er­a­tor will do for AI cloud com­put­ing what the ATX stan­dard did to make PCs per­va­sive to­day. It will en­able cloud-ser­vice providers to eas­ily adopt NVIDIA GPUs to meet surg­ing de­mand for AI com­put­ing.”

The HGX-1 will be pow­ered by eight NVIDIA Tesla P100 GPUs fit­ted in each chas­sis, and will uti­lize an in­no­va­tive switch­ing de­sign based on NVIDIA’s NVLink in­ter­con­nect tech­nol­ogy and PCIe stan­dard. Once com­pleted, the HGX1 will al­low for op­ti­mal per­for­mance, re­gard­less of the work­load.

On pa­per, the ma­chine will pro­vide up to 100 times faster deep learn­ing per­for­mance com­pared with legacy CPU-based servers, and it is fore­casted that it will only re­quire a fifth of the cost to run legacy CPU-based servers in or­der to con­duct AI train­ing and a tenth of the cost for AI in­fer­enc­ing.

Newspapers in English

Newspapers from Malaysia

© PressReader. All rights reserved.