Reuters reported earlier this yr that Meta deliberate to deploy a brand new model of a customized knowledge middle chip to handle the swelling quantity of computing energy essential to run AI merchandise in Fb, Instagram and WhatsApp. The chip, referred to internally as “Artemis,” will assist Meta cut back its reliance on Nvida’s AI chips and cut back its power prices total.
Elevate Your Tech Prowess with Excessive-Worth Talent Programs
| Providing School | Course | Web site |
|---|---|---|
| MIT | MIT Expertise Management and Innovation | Go to |
| Indian Faculty of Enterprise | ISB Skilled Certificates in Product Administration | Go to |
| IIT Delhi | IITD Certificates Programme in Information Science & Machine Studying | Go to |
“This chip’s structure is basically targeted on offering the proper steadiness of compute, reminiscence bandwidth, and reminiscence capability for serving rating and suggestion fashions,” the corporate wrote in a weblog put up.
The brand new Meta Coaching and Inference Accelerator (MTIA) chip is a part of a broad customized silicon effort on the firm that features taking a look at different {hardware} programs too. Past constructing the chips and {hardware}, Meta has made vital investments in growing the software program essential to harness the ability of its infrastructure in probably the most environment friendly means.
The corporate can also be spending billions on shopping for Nvidia and different AI chips: This yr CEO Mark Zuckerberg mentioned the corporate deliberate to accumulate roughly 350,000 flagship H100 chips from Nvidia. Mixed with different suppliers, Meta plans to build up the equal of 600,000 H100 chips this yr, he mentioned. THE NUMBERS
Taiwan Semiconductor Manufacturing Co will produce the brand new chip on its “5nm” course of. Meta mentioned it’s able to thrice the efficiency of its first technology processor. WHAT IS NEXT
Uncover the tales of your curiosity
The chip has been deployed within the knowledge middle and is engaged in serving AI purposes. The corporate mentioned it has a number of applications underway “aimed toward increasing the scope of MTIA, together with assist of (generative AI) workloads.”
