Last Wednesday, Meta revealed information about its next-gen in-house AI chip, designed to enhance the performance of its AI models.
Before the unveiling, there were reports from tech blogs such as Reuters that the Facebook parent company Meta is planning to deploy a new custom data center chip.
The AI chip is called Artemis, designed to address the increasing amount of computing power essential for running AI products on WhatsApp, Facebook, and Instagram.
“Artemis will also help Meta reduce its reliance on Nvidia’s AI chip, lowering its energy costs on the balance sheet,” says Meta. “This chip’s architecture is focused on providing the right balance of compute, memory bandwidth, and memory capacity for serving ranking and recommendation models,” the company added.
The new Meta Training and Inference Accelerator (MTIA) chip represents a pivotal component in the company’s comprehensive custom silicon initiative, which involves exploring other hardware systems.
Aside from building better chips and hardware, Meta is also investing in developing the required software to enable users to harness the power of its hardware infrastructure efficiently.
Taiwan Semiconductor Manufacturing Co will produce the new chip on its 5nm process, which, according to Meta, provides three times the performance of its first-gen processors.
The chip currently serves AI applications in the data center. Meta highlights that several programs are in development to expand the scope of the new AI chip, including support for generative artificial intelligence workloads.
