Meta reveals that it will introduce its own designed AI processor 'Artemis' to its data center in the second half of 2024

Meta, which operates Instagram and Facebook, also develops large-scale language models such as

Llama . It has been revealed that Meta plans to introduce its uniquely designed AI processor ' Artemis ' to its data center in 2024.

Exclusive: Meta to deploy in-house custom chips this year to power AI drive - memo | Reuters

Meta to deploy custom-designed Artemis AI processor alongside commercial GPUs | Tom's Hardware

Meta plans to deploy its Artemis AI chip this year to reduce reliance on Nvidia GPUs

In addition to introducing many AI functions into major services such as Facebook and Instagram, Meta is also developing hardware that utilizes AI, such as announcing the Ray-Ban Meta smart glasses that can use AI assistants. . According to Reuters, the Artemis processor developed by Meta specializes in ``inference processing'' that is essential when running AI software.

Meta is building a large-scale data center consisting of 350,000 NVIDIA high-performance AI chips 'H100' for AI learning and inference. However, purchasing and deploying the H100 comes at a significant cost. Dylan Patel, founder of semiconductor research group SemiAnalysis, said, ``If Meta's implementation of Artemis is successful, Meta will save hundreds of millions of dollars in annual energy costs and reduce the cost of third-party AI processors. 'This could potentially save billions of dollars (hundreds of billions of yen) in purchasing costs.'

Meta CEO Mark Zuckerberg announces that he aims to develop and open source artificial general intelligence (AGI), and is also building computing infrastructure including 350,000 H100s - GIGAZINE

A Meta spokesperson revealed that the Artemis processor is scheduled to go into mass production in the second half of 2024, and reports that the development of AI models will be progressed in conjunction with Meta's hundreds of thousands of existing AI processors. doing. 'We believe that Meta's proprietary accelerators can be highly complementary to commercially available GPUs in achieving the optimal combination of performance and efficiency for Meta-specific workloads.'

in Hardware, Posted by log1r_ut