Meta Develops In-house AI Chips, Looks To Curb Nvidia Dependence
Meta has revealed the details of its next-generation artificial intelligence (AI) accelerator chip, which the company will develop in-house to address the demanding computing power necessary to run AI products on its popular platforms Facebook, Instagram and WhatsApp, according to a media report on Thursday.
Meta’s in-house chips were in talks for several months and are internally known as ‘Artemis’, developed to reduce the company’s reliance on Nvida's AI chips, the report said additionally.
In a blog post, the company highlighted its strategy for introducing AI chips which is significantly focused on reducing energy costs and striking the right balance of computing, memory bandwidth and memory capacity for serving ranking and recommendation models.
Additionally, the new Meta Training and Inference Accelerator (MTIA) chip underlines the organisation’s broader custom silicon efforts that include monitoring other hardware systems as well.
Meanwhile, beyond the chips and hardware building the company has also significantly invested in software developments to utilise the power of its infrastructure most efficiently.
