Facebook, Instagram, and WhatsApp’s parent company Meta unveiled its first Custom silicon AI chip called MTIA (Meta training and inference accelerator). This initiative is a component of Meta's larger plan to transform its AI infrastructure, speed up AI research, and influence the developing metaverse.
Software developer Joel Coburn from Meta stated during a presentation on the new processor that the company had first used graphics processing units, or GPUs, for inference tasks but had discovered that they were not the best option. "Their efficiency is low for real models, despite significant software optimizations. This makes them challenging and expensive to deploy in practice, this is why we need MTIA," said Joel.
In addition to offering more computational power and efficiency than conventional CPUs, this silicon chip will be specifically useful for Meta's internal workloads, providing optimal performance, decreased latency, and greater efficiency across a variety of jobs.
Ceo Mark Zuckerberg shared the picture of the chip on his social media. “MTIA is our first generation custom silicon chip that we designed to power our AI recommendation systems to help figure out the best content to show you even faster,” he wrote.
In addition, Meta unveiled CodeCompose, a new generative AI coding aid that will compete with Google's improved Colab and GitHub's Copilot.“We’re executing an ambitious plan to build the next generation of Meta’s AI infrastructure,” said Santosh Janardhan, head of infrastructure at Meta, in a blog post. He claimed that Meta custom designs most of its infrastructure, including data centres, server hardware, and mechanical systems, to optimise the end-to-end experience.
“This will be increasingly important in the years ahead,” said Santhosh.“Over the next decade, we’ll see increased specialisation and customization in chip design, purpose-built and workload-specific AI infrastructure, new systems and tooling for deployment at scale, and improved efficiency in product and design support.” he said.