Meta engaged on new AI chip, next-gen GPUs for video workloads

NEW DELHI: Meta (previously Fb) is constructing its first-generation customized silicon chip for operating synthetic intelligence (AI) fashions, saying its AI compute wants will develop dramatically over the subsequent decade as we break new floor in AI analysis.

Known as MTIA (Meta Coaching and Inference Accelerator), the in-house, customized accelerator chip will present higher compute energy and effectivity than CPUs, and is customised for inside workloads.

“By deploying each MTIA chips and GPUs, we’ll ship higher efficiency, decreased latency and higher effectivity for every workload,” mentioned Santosh Janardhan, VP and Head of Infrastructure at Meta.

The corporate additionally plans a brand new AI-optimised information centre design and the second part of its 16,000 GPU supercomputer for AI analysis.

“These efforts — and extra initiatives nonetheless underway — will allow us to develop bigger, extra subtle AI fashions after which deploy them effectively at scale,” Janardhan added.

The following-generation information centre can be an AI-optimised design, supporting liquid-cooled AI {hardware} and a high-performance AI community connecting hundreds of AI chips collectively for information centre-scale AI coaching clusters.

“It is going to even be quicker and less expensive to construct, and it’ll complement different new {hardware} resembling our first in-house-developed ASIC answer, MSVP (Meta Scalable Video Processor), which is designed to energy the consistently rising video workloads at Meta,” Janardhan knowledgeable.

Meta’s Analysis SuperCluster (RSC) AI supercomputer, which the corporate believes is likely one of the quickest AI supercomputers on the earth, was constructed to coach the subsequent era of huge AI fashions to energy new augmented actuality instruments, content material understanding programs, real-time translation expertise and extra.

It options 16,000 GPUs, all accessible throughout the 3-level Clos community material that gives full bandwidth to every of the two,000 coaching programs.

“Customized-designing a lot of our infrastructure allows us to optimize an end-to-end expertise from the bodily layer to the digital layer to the software program layer to the precise consumer expertise,” mentioned Meta.