5 Reasons Why Apple Ditched Nvidia Chips Over Google for Training AI Models

Although Apple did not specifically mention Google, they revealed that their Apple Foundation Model (AFM) and AFM server are trained on Cloud TPU clusters.

Introduction

Apple, the tech giant renowned for its hardware prowess and privacy focus, has made a surprising revelation: it has been using Google’s Tensor Processing Units (TPUs) to train its AI models.

This move, while unexpected, highlights the immense computational power required for developing cutting-edge AI systems and the collaborative nature of the tech industry.

Apple’s Foray into Generative AI

Apple’s generative AI models, known as Apple Foundation Models (AFMs), are designed to enhance user experiences across Apple’s ecosystem. These models are capable of performing tasks such as text summarization, suggested wording for messages, and more.

Unlike traditional AI models, generative AI models can create new content and responses, making them highly versatile for applications in natural language processing and other fields.

Apple AI Model

“Pre-training data set consists of … data we have licensed from publishers, curated publicly available or open-sourced datasets and publicly available information crawled by our web crawler, Applebot,” Apple wrote. “Given our focus on protecting user privacy, we note that no private Apple user data is included in the data mixture.”

Although Apple did not specifically mention Google, they revealed that their Apple Foundation Model (AFM) and AFM server are trained on Cloud TPU clusters. This setup enables efficient and scalable training of AFM models, including versions for devices, servers, and larger models,” the company said.

The paper also detailed how the use of TPUs facilitated the efficient training of large AI systems and highlighted the company’s ethical considerations in AI development.

Apple emphasized that it has not utilized private user data in training its AI models and has consistently adhered to responsible data practices.

TPU Utilization: Apple employed Google’s TPUv5p chips for building AI models and TPUv4 processors for server-based AI models.  

Apple Intelligence: These AI models power Apple Intelligence, a personal intelligence system integrated into iOS 18, iPadOS 18, and macOS Sequoia.  

Data Privacy: Apple emphasizes that no private user data was used in training the models. The data consisted of licensed content, public datasets, and information gathered by Applebot.  

Efficiency and Scalability: Google’s TPUs enabled Apple to train large AI models efficiently and at scale.  

Microsoft China Bans Android: Employees to Shift to iPhones by September 2024 – techovedas

Why Google TPUs Over Nvidia GPUs?

Apple’s choice to utilize Google’s Tensor Processing Units (TPUs) over Nvidia GPUs, despite the latter being an industry standard, is a significant decision in the realm of artificial intelligence (AI) hardware. This choice is influenced by a combination of technical, strategic, and possibly relational factors. Here’s a deeper dive into why Apple might have opted for TPUs over Nvidia GPUs:

1. Historical Tensions Between Apple and Nvidia

Apple and Nvidia have had a somewhat strained relationship over the years. This tension has its roots in past disagreements and issues, such as driver support and hardware compatibility.

For instance, Apple has faced challenges with Nvidia’s GPUs in Mac products, leading to a preference for other suppliers. This strained relationship likely made Apple cautious about relying heavily on Nvidia’s hardware for critical AI projects.

2. Technical Advantages of TPUs

Google designed its TPUs as specialized processors specifically for machine learning tasks, focusing on the neural network computations used in training and inference.

Here are some of the technical advantages TPUs might offer:

High Efficiency for Matrix Operations: TPUs are highly efficient at performing matrix multiplications, which are fundamental operations in neural network training. This efficiency can lead to faster training times and lower energy consumption compared to general-purpose GPUs.

Scalability: Engineers designed TPUs to scale efficiently across large clusters, making them suitable for training massive models like those Apple uses for its generative AI. This scalability is crucial for handling the vast computational needs of modern AI systems.

Optimization for Specific Workloads: TPUs are tailored for TensorFlow, a popular machine learning framework developed by Google. While Apple might use a variety of frameworks, the synergy between TPUs and TensorFlow can offer significant performance benefits, making TPUs an attractive option for specific AI workloads.

Follow us on Twitter here

3. Market Dynamics and Diversification

The AI hardware market is highly competitive, with Nvidia holding a dominant position. However, Google’s TPUs have gained significant traction, becoming a leading choice for many large-scale AI projects.

By choosing TPUs, Apple can avoid over-reliance on Nvidia and diversify its hardware ecosystem. This diversification is not only a strategic move to mitigate risk but also a way to leverage the strengths of multiple hardware providers.

4. Cost and Availability Considerations

As the demand for AI hardware skyrockets, securing the necessary resources becomes a challenge. Nvidia’s GPUs, particularly the high-end models like the H100, are in high demand and can be costly.

TPUs, being a product of Google Cloud, offer an alternative that might be more cost-effective or available in larger quantities, especially considering Google’s vast cloud infrastructure. This availability can be crucial for a company like Apple, which needs substantial computing power for training its AI models.

5. Strategic Partnerships and Future Directions

Apple’s decision to use TPUs might also reflect broader strategic considerations. By collaborating with Google on hardware, Apple could be positioning itself to benefit from future advancements in TPU technology.

Additionally, this partnership might pave the way for further collaboration in areas like cloud computing and AI research, where both companies have significant investments.

Read the Apple paper here

Conclusion

Apple’s use of Google’s custom chips to train its AI models marks a significant milestone in the tech industry.

This collaboration shows the changing landscape of AI chip development. Innovation is crucial to staying competitive.

With the launch of Apple Intelligence, users will enjoy enhanced capabilities. This advanced AI technology drives new features and improvements.

Kumar Priyadarshi
Kumar Priyadarshi

Kumar Joined IISER Pune after qualifying IIT-JEE in 2012. In his 5th year, he travelled to Singapore for his master’s thesis which yielded a Research Paper in ACS Nano. Kumar Joined Global Foundries as a process Engineer in Singapore working at 40 nm Process node. Working as a scientist at IIT Bombay as Senior Scientist, Kumar Led the team which built India’s 1st Memory Chip with Semiconductor Lab (SCL).

Articles: 2622