Apple recently released information, as noted CNBCto a certain extent confirms that the degree of dependence of developers of artificial intelligence systems on NVIDIA components is greatly exaggerated. At least, Apple indicates that it pre-trains its artificial intelligence system Apple Intelligence using Google cloud clusters built on the basis of tensor processors of the same company. As noted earlier, in the future, Apple plans to build its own computing infrastructure for AI systems using adapted processors of its own design. NVIDIA does not particularly fit into these plans yet.
Image source: Reuters, Apple
Currently, when training large language models that will work as part of Apple Intelligence, the company of the same name uses two Google clusters. One of them, responsible for language models for end devices, is built on 2048 Google tensor processors of the latest generation v5p. The second, which is responsible for language models for the server part of the infrastructure, is built on 8192 tensor processors of the v4 series. Apple does not mention NVIDIA products when describing the device of this infrastructure, although the use of Google components can only be judged by indirect signs such as the characteristic abbreviation TPU.
