By Katie Paul and Stephen Nellis
NEW YORK (Reuters) – Metaplatforms on Thursday unveiled a project it is advancing to make its data centers better able to support artificial intelligence capabilities, including a custom chip “family” it said it was developing in-house. shared new details.
In a series of blog posts, Facebook and Instagram owners said they designed the first-generation chip in 2020 as part of the Meta-Training and Inference Accelerator (MTIA) program. The program aims to improve the efficiency of the recommendation models used for serving. Ads and other content in your news feed.
Reuters had previously reported that the company had no plans to widely deploy its first homegrown AI chip and was already working on a successor chip. In his blog post, he described his first MTIA chip as a learning opportunity.
“From this first program, we have learned valuable lessons and are incorporating them into our roadmap,” it reads.
The first MTIA chips focused solely on an AI process called inference. For inference, algorithms trained on vast amounts of data determine whether to show a video of her dancing, a cat meme, etc. as the next post in a user’s feed. Said.
A Meta spokesperson declined to comment on the deployment schedule or provide details about the company’s plans to develop a chip that can train models.
Meta upgraded its AI infrastructure last year after executives realized it lacked the hardware and software needed to support demand from product teams building AI-powered features. I have been working on a large scale project.
As part of that, Reuters reports that the company has scrapped plans to roll out an in-house inference chip at scale and set out to develop a more ambitious chip capable of both training and inference.
In a blog post, Meta admitted that its first MTIA chip stumbled on high-complexity AI models, but the chip is more efficient at processing low- and medium-complexity models than its competitors’ chips. He said it was handled properly.
The MTIA chip also consumes only 25 watts, a fraction of the power consumed by chips from market-leading suppliers such as NVIDIA Corp, and uses an open-source chip architecture called RISC-V. said Mehta.
In addition to details on how its chips work, Meta updates on plans to redesign its data center around more modern AI-driven networking and cooling systems, breaking ground on the first such facility this year. said he planned to
The new design will be 31% cheaper than the company’s current data center and can be built twice as fast, an employee said in a video explaining the changes.
(Reporting by Katie Paul in New York and Stephen Nellis in San Francisco; Editing by Kenneth Lee and Chizu Nomiyama)
