Meta has unveiled details about the next generation of the company’s in-house AI chip as the company aims to drive its AI workloads while reducing its reliance on Nvidia silicon. The company’s latest version of its Meta Training and Inference Accelerator (MTIA) family of custom-made chips is designed for Meta’s AI workloads.
The company stated that these new-gen AI chips show significant performance improvements over MTIA v1, released in 2023. The chip is being used to help power Meta’s ranking and recommendation ads models.
MTIA is part of Meta’s growing investment in its AI infrastructure and will complement its existing and future AI infrastructure to deliver new and better experiences across its products and services.
The company stated that the latest MTIA chip is currently live in 16 of its data centre regions and delivering up to 3x overall better performance compared to MTIA v1.
“MTIA will be an important piece of our long-term roadmap to build and scale the most powerful and efficient infrastructure possible for Meta’s unique AI workloads,” the company said in a blog post.
“We’re designing our custom silicon to work in cooperation with our existing infrastructure as well as with new, more advanced hardware (including next-generation GPUs) that we may leverage in the future,” the company added.
“Meeting our ambitions for our custom silicon means investing not only in compute silicon but also in memory bandwidth, networking and capacity as well as other next-generation hardware systems.”
“We currently have several programs underway aimed at expanding the scope of MTIA, including support for GenAI workloads. And we’re only at the beginning of this journey.”
Meta has previously announced plans to build out massive compute infrastructure to help support its Gen AI ambitions, including the latest version of its open-source Llama LLM, Llama 3, which is set to release in 2024.
In a statement earlier, Meta CEO Mark Zuckerberg said that the company was bringing its AI research team ‘closer together’ and that it was building out its compute infrastructure to support its future roadmap, which includes a further push into AI and a move towards artificial general intelligence.
“Our long-term vision is to build general intelligence, open source it responsibly, and make it widely available so everyone can benefit,” Zuckerberg said.
“We’re currently training our next-gen model Llama 3, and we’re building massive compute infrastructure to support our future roadmap, including 350,000 H100s by the end of this year – and overall almost 600,000 H100s equivalents of compute if you include other GPUs.”
