|
Apple's decision to rely on Google's cloud infrastructure is
notable because Nvidia produces the most sought-after AI
processors.
Including the chips made by Google, Amazon.com and other cloud
computing companies, Nvidia commands roughly 80% of the market.
In the research paper, Apple did not explicitly say that it used
no Nvidia chips, but its description of the hardware and
software infrastructure of its AI tools and features lacked any
mention of Nvidia hardware.
Apple did not comment on Monday.
The iPhone maker said that to train its AI models, it used two
flavors of Google's tensor processing unit (TPU) that are
organized in large clusters of chips.
To build the AI model that will operate on iPhones and other
devices, Apple used 2,048 of the TPUv5p chips. For its server AI
model, Apple deployed 8,192 TPUv4 processors.
Nvidia does not design TPUs but rather focuses its efforts on
so-called graphics processing units (GPUs) that are widely used
for AI efforts.
Unlike Nvidia, which sells its chips and systems as standalone
products, Google sells access to TPUs through its Google Cloud
Platform. Customers interested in buying access must build
software through Google's cloud platform in order to use the
chips.
Apple is rolling out portions Apple Intelligence to its beta
users this week.
Reuters reported the use of the TPU chips in June, but Apple did
not disclose the full extent of its reliance on Google hardware
until Monday's research paper.
Google did not return a request for comment, while Nvidia
declined to comment.
Apple's engineers said in the paper it would be possible to make
even larger, more sophisticated models with Google's chips, than
the two models it discussed in the paper.
Apple unveiled a slew of new AI features at its June developer
conference, including integrating OpenAI's ChatGPT technology
into its software.
The Cupertino, California-based company's stock ticked down 0.1%
to $218.24 in regular trading on Monday.
(Reporting by Max A. Cherney in San Francisco; Editing by
Matthew Lewis and Varun H K)
[© 2024 Thomson Reuters. All rights
reserved.]
This material may not be published,
broadcast, rewritten or redistributed.
Thompson Reuters is solely responsible for this content.
|
|
