Meta Announced New Chips Specifically Designed for AI Models

A novel innovation has been revealed, Meta announced new chips specifically designed for AI models

With the creation of a new processor particularly tailored for executing AI models, Meta is making major advances in artificial intelligence (AI). The business, known for its app family and metaverse vision, is beginning an ambitious mission to construct the next generation of AI infrastructure. Santosh Janardhan, Meta's VP & Head of Infrastructure, has blogged on their success. Meta is also working on an AI-optimized data center architecture and the second phase of a gigantic 16,000 GPU supercomputer dedicated to AI research, in addition to the special silicon chip for AI models. These programs seek to make it easier to design and deploy larger and more powerful AI models at scale.

According to Meta, AI already plays a critical role in Meta's products, leading to greater personalization, safer and fairer goods, and improved user experiences. With the launch of Code Compose, a generative AI-based coding helper designed to enhance developer productivity throughout the software development lifecycle, the firm is even reinventing how coding is done.

The MTIA (Meta Training and Inference Accelerator), an in-house bespoke accelerator chip family, lies at the heart of Meta's infrastructure improvements. According to Meta, MTIA is optimized for inference workloads and outperforms CPUs in terms of computational power and efficiency. Combining MTIA chips with GPUs improves speed, lowers latency, and increases efficiency for each application.

Meta is building a data center to serve current products while also accommodating future generations of AI gear. This AI-optimized architecture will include liquid-cooled AI hardware as well as a high-performance AI network that will connect thousands of AI chips to build data center-scale AI training clusters.