In a surprising revelation, Apple has disclosed that it does not use Nvidia GPUs to train its artificial intelligence models. Instead, the tech giant has turned to Google's custom-designed Tensor Processing Units (TPUs) for this crucial task. This information comes from a technical paper released by Apple, shedding light on the company's approach to AI development.
The decision to use Google's TPUs marks a significant departure from industry norms, as Nvidia's GPUs have long been the go-to choice for AI training among major tech companies. Apple's move suggests a potential shift in the AI chip landscape and highlights the growing competition in the field of machine learning hardware.
Apple's AI strategy, dubbed “Apple Intelligence,” encompasses multiple generative models designed for various user tasks. These models include a 3-billion parameter on-device language model and a larger server-based model. The company emphasizes that these AI systems are built with a focus on user privacy and responsible AI development.
The use of Google TPUs for training extends to both Apple's on-device and server-based models. Specifically, the on-device model was trained using a single “slice” of 2048 TPU v5p chips, while the server model utilized 8192 TPU v4 chips organized into eight slices via a data center network.
Apple's decision to leverage Google's cloud infrastructure for AI training is particularly noteworthy given the company's usual preference for in-house solutions. This choice may be driven by the need for specialized hardware and scalability in AI model development.
Despite using Google TPUs for training, Apple maintains that the inference process – where trained models generate content or make predictions – will occur partly on its own chips in its data centers. This approach aligns with Apple's commitment to on-device machine learning and privacy-preserving AI technologies.
The company's focus on on-device AI is evident in its recent developer resources. At WWDC24, Apple introduced new tools and frameworks to help developers integrate machine learning and AI capabilities into apps for iPhone, iPad, Mac, and other Apple devices.
Apple's AI initiatives extend beyond just language models. The company is also developing a coding model for Xcode and a diffusion model for visual expression in apps like Messages. These developments showcase Apple's comprehensive approach to integrating AI across its ecosystem.
As the AI race heats up among tech giants, Apple's unique approach to model training and deployment sets it apart from competitors. By leveraging Google TPUs while maintaining control over inference and on-device processing, Apple is carving out its own path in the rapidly evolving landscape of artificial intelligence.
This strategic move not only highlights the shifting dynamics in AI hardware preferences but also underscores the importance of diverse approaches to machine learning development in the tech industry.
BONUS: Get our $200 “AI Mastery Toolkit” FREE when you sign up!
