• The Meta platforms announced details of the company’s upgrade version of AI chip on Wednesday. The chip helps train the models faster, reduce the dependence of meta-on Nvidia AI chips and reduces the whole energy costs.
  • Last year, Meta unveiled the first version of Meta training and Inference Accelerator(MTIA v1). MTIA is a long-term venture that provides the most efficient architecture for meta critical workloads.

The Meta platforms announced details of the company’s upgrade version of AI chip on Wednesday.

AI chip by Meta

Meta Platform presented its next generation of artificial intelligence accelerator chip on Wednesday, a meta-custom chip for Meta’s AI workload. Meta has been involved in this project named “Artemis” for a while and The advanced chip also targets the data centers like the former version.

Next-generation Meta is under construction of large-scale infrastructure such as next-generation AI (GenAI) products and the series of related services, recommendation systems, and advanced AI research support.

Also read: Open AI and Meta to launch AI models with ‘reasoning’ capabilities.

High-quality evolution

With this new version of MTIA, the calculation and memory bandwidth of the old solution can be doubled and at the same time, a close connection with the workload can be established. To give users high quality recommendations, it has been developed to rank lists and recommended models.

Meta’s new TIA design also features an improved chip structure (NC), which allows adjustments to the network of other PEs. These and other developments are essential to USA’s long-term road map, their long-term road map and broader challenges.

In order to increase computer power and catch up the rivals, many AI companies like Google, Microsoft, Amazon were already looking for their own chips.

Meta said in its post that “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.”

Also read: Meta to integrate AI into Ray-Ban smart glasses

Former version of MTIA

The initial version of the Meta training and Inference Accelerator(MTIA) debuted last year. Meta’s first-generation AI accelerator within the company got the deep learning recommendation models, the variety of the experience was expanded by the products, in mind.

MTIA v1 is a significant step in improving the efficiency of infrastructure computing and better supporting software developers.