meta The platform on Thursday shared new details on projects it was undertaking to improve its data centers to support artificial intelligence work, including a custom chip “family” that it said would Developing at home.
Facebook And Instagram The owner said in a series of blog posts that it designed the first-generation chip in 2020 as part of the Meta Training and Invention Accelerator (MTIA) program, which aims to improve recommendation models used to serve ads and other content. To improve efficiency. in News Feed.
Reuters previously reported that the company was not planning to widely deploy its first in-house AI chip and was already working on a successor. The blog post portrayed the first MTIA chip as a learning opportunity.
“From this initial program, we have learned invaluable lessons that we are incorporating into our roadmap,” it wrote.
The first MTIA chip was specifically focused on a oh The process is called inference, in which algorithms trained on large amounts of data decide whether to show a dance video or cat meme as the next post in a user’s feed.
A Meta spokesperson declined to comment on deployment timelines or elaborate on the company’s plans to develop chips that can also train models.
Meta engaged in a massive project to upgrade its AI infrastructure last year after executives realized it lacked the hardware and software needed to support the demand for product teams building AI-powered features. There is a shortage.
As part of that, the company scrapped plans for a large-scale rollout of an in-house inference chip and began work on a more ambitious chip capable of doing both training and inference, according to Reuters reporting.
Meta acknowledged in its blog post that its first MTIA chip stumbled with high-complexity AI models, though it said the chip handles low- and medium-complexity models more efficiently than competing chips.
The MTIA chip also used only 25 watts of power – a fraction of market-leading chips from suppliers such as NVIDIA consume — and used an open-source chip architecture called RISC-V, Meta said.
In addition to detailing its chip work, Meta also provided an update on plans to redesign its data centers around more modern AI-oriented networking and cooling systems, saying it will build its first such facility this year. Will be based on convenience.
An employee said in a video explaining the changes that the new design would be 31 percent cheaper and could be built up to twice as fast as the company’s current data centers.
© Thomson Reuters 2023