Meta Introduces Advanced AI Chip for Enhanced Computing Power

Meta Platforms has revealed details about its latest in-house artificial intelligence accelerator chip, signaling a significant advancement in its efforts to improve computing capabilities.

Ad

The introduction of the new chip, internally named "Artemis," demonstrates Meta's strategic move to lessen dependence on external AI chips, such as Nvidia's, while also reducing energy costs across its platforms, including Facebook, Instagram, and WhatsApp.

The Meta Training and Inference Accelerator (MTIA) chip features an architecture designed to optimize computing power, memory bandwidth, and capacity, particularly suited for ranking and recommendation models, as highlighted in a recent company blog post.

Advertisement

This unveiling follows Meta's broader initiative in custom silicon development, which includes not only chip manufacturing but also refining software to efficiently use its infrastructure. Additionally, the company's significant investments in hardware procurement, notably from Nvidia, underscore its commitment to enhancing computational resources.

Taiwan Semiconductor Manufacturing Co. (TSMC) will manufacture the new chip using its advanced "5nm" process, promising a substantial performance improvement compared to its predecessor.

The deployment of the chip in data centers marks a significant milestone, with Meta already employing it to power various AI applications. Furthermore, the company has outlined plans to expand MTIA's capabilities, including support for generative AI workloads.

As Meta continues to advance its AI technology, the introduction of the new chip signifies a promising step towards achieving greater computational efficiency and performance across its platforms.

Also Read: Israel Sees Surge in Cybersecurity Exits, but Funding Drops in 2023