Apple Intelligence AI Models Trained on Google's TPUs, Not Nvidia's

Jul 29, 2024, 08:33 PM

Apple has released a detailed paper outlining its AI models, revealing that both the on-device and server versions were fully trained on Google’s Tensor Processing Units (TPUs). Notably, Nvidia's GPUs were not used in any part of the training process. The technical paper, which delves deeply into the specifics of the AI models, does not specify the size of the larger model, though it is speculated to be 70 billion parameters. This shift to using Google's custom chips marks a significant move for Apple Intelligence, which has traditionally relied on Nvidia's hardware for such tasks. The information was reported by CNBC.