Meta reveals in-house AI chip family for data centre upgrades

Meta Platforms recently unveiled its in-house custom chip “family” aimed at enhancing artificial intelligence (AI) work. The company developed its first-generation chip in 2020 as part of the Meta Training and Inference Accelerator (MTIA) programme, to improve efficiency for recommendation models used in serving ads and other content in news feeds.

The first MTIA chip was designed exclusively for an AI process called inference, where algorithms trained on large amounts of data make judgments about which content to display next in a user’s feed.

Joel Coburn, a software engineer at Meta, explained that the company initially used graphics processing units (GPUs) for inference tasks but found them ill-suited for the job. He stated…

“Their efficiency is low for real models, despite significant software optimizations. This makes them challenging and expensive to deploy in practice. This is why we need MTIA.”

A spokesperson for Meta did not provide details on deployment timelines for the new chip or plans to develop chips for training models. The company has been working on upgrading its AI infrastructure in the past year after recognising that it lacked the necessary hardware and software to support AI-powered features. Consequently, Meta scrapped plans for a large-scale rollout of its in-house inference chip and began working on a more ambitious chip capable of performing both training and inference tasks.

While the first MTIA chip struggled with high-complexity AI models, it managed low- and medium-complexity models more efficiently than competitor chips. The chip consumed only 25 watts of power, significantly less than market-leading chips from suppliers like Nvidia Corp, and used an open-source chip architecture called RISC-V.

Meta also announced plans to redesign its data centres with modern AI-oriented networking and cooling systems, with the first facility set to break ground this year. The new design is expected to be 31% cheaper and built twice as fast as the company’s current data centres.

In addition, Meta has developed an AI-powered system to assist its engineers in creating computer code, similar to tools offered by Microsoft Corp, Amazon.com Inc, and Alphabet Inc, reported Channel News Asia.