Meta Introduces AI Processors and Supercomputer to Boost Artificial Intelligence Research

Meta Training and Inference Accelerator (MTIA)

Meta introduced its new chip, called the Meta Training and Inference Accelerator (MTIA), which is part of a “family” of processors designed to accelerate AI learning and inference tasks. The company developed the first generation of MTIA in 2020 and claims it significantly improves the company’s efficiency in terms of performance per watt when running recommendation tasks.

Custom AI processors are becoming increasingly common among technology giants. Google designed a processor, the TPU (short for “tensor processing unit”), for training large generative AI systems. Amazon offers proprietary chips for AWS customers, and Microsoft collaborates with AMD to develop an in-house AI processor called Athena.

Read also: Neuroscience Expert Gary Marcus Speaks at US Senate on AI

Research SuperCluster (RSC) Supercomputer

In addition to the MTIA, Meta also introduced its AI research-dedicated supercomputer, called the Research SuperCluster (RSC). Assembled in collaboration with Penguin Computing, Nvidia, and Pure Storage, the RSC comprises a total of 2,000 Nvidia DGX A100 systems equipped with 16,000 Nvidia A100 GPUs.

The RSC enables Meta researchers to train models using real-world examples from the company’s production systems, unlike the previous AI infrastructure that relied solely on open-source and publicly accessible datasets.

Meta Scalable Video Processor (MSVP)

Meta is also developing another processor to handle specific types of computing tasks, called the Meta Scalable Video Processor (MSVP). This processor is designed to meet the processing needs of on-demand video and live streaming. Meta plans to use the MSVP for the majority of its “stable and mature” video processing tasks, resorting to software video encoding only for tasks requiring specific customization and significantly higher quality.

In summary, Meta is working to catch up in the AI space, particularly generative AI. The company faces increasing pressure from investors who fear it is not positioning itself quickly enough to capture the potentially significant generative AI market. If forecasts are accurate, the total addressable market for generative AI software could reach $150 billion, and Goldman Sachs estimates it could increase GDP by 7%.