Meta has developed 4 new chips to power its AI and recommendation systems

He was pronounced dead Wednesday They have developed four new computer chips that will be used for power Generative AI features and Content classification systems Among the technology giant’s applications. The devices will become part of the existing Meta chipset line known as MTIA, or Meta Training and Inference Accelerators.

Meta has partnered with Broadcom to develop state-of-the-art semiconductors, built on top of the open source RISC-V architecture. It is manufactured by Taiwan Semiconductor Manufacturing Company, a global company Leading chip manufacturer.

One of MTIA’s new chips, the MTIA 300, is already in production, and Meta says the other three — the MTIA 400, 450, and 500 — are expected to ship sometime between early and late 2027. Rolling out new silicon so quickly is unusual by most chip industry standards, but unheard of for a social media company that hasn’t historically produced its own physical computing infrastructure.

AI models evolve faster than traditional chip development cycles, so AI workloads may change dramatically by the time new devices reach production, says YJ Song, vice president of engineering at Meta. “Instead of betting and waiting for a long period of time, we are deliberately taking an iterative approach. Each MTIA generation builds on the last generation, using modular microchips and incorporating the latest AI workload insights and hardware technologies,” Song said in a statement. Blog post.

The MTIA 300 will be used primarily to train algorithms that classify and recommend content for the hundreds of millions of people who use apps like Facebook and Instagram every day. The other three chips are designed to support Inferencethe process of running trained AI models to produce output such as text or images.

The MTIA 400, which Meta claims delivers performance “competitive with leading commercial products,” has been tested and is expected to arrive in data centers soon. The MTIA 450 will have twice the amount of high-bandwidth memory as the MTIA 400, and should ship in early 2027. Meta says the MTIA 500, which is scheduled to arrive later next year, will have more memory than the MTIA 450 and include “innovations in low-resolution data.”

MTIA chips are part of Meta’s broader strategy to store as much as possible Computing power As much as possible in order to develop advanced artificial intelligence. Meta first shared details about its chip development plans in 2023, when it released its first product under the MTIA banner. As software companies and AI labs continue to train increasingly powerful AI models, they have begun announcing ambitious plans to build custom chips that serve their AI needs. OpenAI, for example, also said It’s a partnership with Broadcom To create custom accelerators, follow a path similar to the Meta path.

Earlier this year, Meta Downsizing has been reported Some of its internal efforts are to make high-end chips that will compete directly with leading players like Nvidia. The company now seems keen to dispel this narrative by announcing its new roadmap for MTIA chips. But manufacturing custom silicon is still very expensive and technically complex, which means Meta will likely continue to buy the majority of its AI hardware from other companies, at least for the near future.

This reality is reflected in the company’s recent chip buying spree. Meta unveiled the new MTIA chipset shortly after Announcing multi-billion dollar deals With Nvidia and AMD. The social media giant also signed an agreement to lease chips made by Google.