Apple selects Google Tensor chips to train AFM language models

Apple has announced the release of the beta version of Apple Intelligence, which uses artificial intelligence to improve iPhone functions. To train two key language models, AFM-on-device and AFM-server, the company used Google Tensor chips rather than Apple Silicon's own chips.

Here's What We Know

The AFM-on-device model used 2048 TPUv5p chips, while the larger AFM-server model required 8192 TPUv4 chips. This choice of Google chips rather than Nvidia is surprising, but could be a sign of a change in Apple's approach to developing its technology.

These innovations indicate that the iPhone will become even smarter through the integration of Apple Intelligence and Google technology.

Source: Apple