Tech

Further report: The hardware and infrastructure behind Gemini-Siri – Google and Nvidia | News

Reports have been circulating for months that Apple will no longer run the next generation of Siri and certain models from Apple Intelligence locally or on its own servers, but will instead seek Google’s help. Just a few days ago, reports about Google Cloud and Nvidia hardware emerged again. The Information has now followed up again and clarified the statements made recently. Said cloud requests should actually run on Google’s infrastructure with “Nvidia Blackwell B200” servers. Apple wants to use Nvidia’s “Confidential Compute” to keep user data encrypted during processing.

Change of course is probably imminent
That would be a significant departure from Apple’s previous external image. Since introducing Apple Intelligence, Apple has emphasized that tasks run directly on the device and only more complex requests are processed via private cloud compute – i.e. on servers with Apple Silicon and an Apple-controlled data protection model. However, this external processing is likely to be the standard case for newer models in the future.

Apple infrastructure is not enough
The reason is apparently simply the required computing power. Apple tried internally to get a customized Gemini version running on its own private cloud compute infrastructure. But this solution was too slow. This is exactly where Google’s data centers with Nvidia Blackwell B200 came into play, because they are designed for large AI models and specifically for training and inference of large language models.

Emergency solution: An otherwise unpopular approach
This would mean that Apple would be using exactly the architecture in the background that the company otherwise prefers to avoid. Foreign models on a foreign cloud with foreign accelerators for a key function of the system. While this is technically understandable, it remains tricky when it comes to communication. Temporal aspects may also have played a role, as another one or two years late would look quite strange.

Data protection via Nvidia functions
Nvidia’s Confidential Compute could be an attempt to mitigate the contradiction between the 2024 representation (“local, for data protection”) and the rumored plans. Processing takes place in a protected environment so that the cloud operator does not have access to the unencrypted data. As it has also been said several times: In the long term, Apple would like to have everything in its own hands, so the Google solution is planned for a transition phase (probably several years).

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button
Close

Adblock Detected

kindly turn off ad blocker to browse freely