Meta to Deploy Custom AI Chips Alongside Nvidia GPUs for Specialized Workloads
-
Meta will deploy its own custom AI chips, called MTIA, starting with an inference chip called Artemis in 2024. This will complement GPUs from Nvidia and AMD.
-
The Artemis chip is based on Meta's first-gen MTIA chip announced last year, but with improved cores and high-bandwidth memory packaging.
-
Meta has become one of the biggest customers for Nvidia and AMD GPUs as it scales its AI workloads. Custom chips will help optimize specific workloads.
-
Meta plans to deploy 350,000 Nvidia H100 GPUs this year alone as it focuses more on developing artificial general intelligence.
-
Details are light on the exact capabilities of Meta's custom chips. More updates are expected later this year.