Apple Asserts Using Google TPU For AI Training
Apple released a technical report stating that two of Apple Intelligence's models were pre-trained in the cloud using a Google-designed chip.
On July 29, Apple released a technical report stating that two models of its artificial intelligence system, Apple Intelligence, were pre-trained on Google-designed chips in the cloud.
The report detailed how Apple utilized Google's Tensor Processing Units (TPUs) to train its AI models and simultaneously launched a preview version of Apple Intelligence.
Tensor Processing Units (TPUs) are specialized chips designed by Google to accelerate machine learning tasks, especially the training and inference of deep learning models. TPUs are widely used on the Google Cloud platform, available for developers to rent for large-scale machine learning projects, known for their efficiency and low power consumption.
According to the report, Apple used 2,048 TPU v5p chips for training the AFM on-device model and 8,192 TPU v4 chips for the AFM-server model. Google’s TPUs, first used internally in 2015 and made available publicly in 2017, have become one of the most mature AI custom chips, with an hourly cost of less than $2 three years after reservation.
Apple first unveiled Apple Intelligence at the WWDC event on June 10, 2024, and released the first version of the system on July 29. The update includes a new look for Siri, improved natural language processing capabilities, and AI-generated text summaries. Over the next year, Apple plans to introduce new features based on generative AI, including image generation, emoji creation, and an upgraded version of Siri.
Although Nvidia's GPUs currently dominate the high-end AI training chip market, demand is high and supply is limited. Companies such as OpenAI, Microsoft, and Anthropic rely on Nvidia's GPUs, while Google, Meta, Oracle, and Tesla are also purchasing these chips in large quantities to build AI systems.
Meta CEO Mark Zuckerberg and Alphabet CEO Sundar Pichai recently noted that while investment in AI infrastructure might seem excessive, the risk of not investing is even greater.
Apple stated that some pre-trained AI model inference will use its own chips in its data centers. This is the second technical report released by Apple regarding its AI system, and the company is expected to announce its quarterly financial report after the market closes on August 1.
Disclaimer: The views in this article are from the original author and do not represent the views or position of Hawk Insight. The content of the article is for reference, communication and learning only, and does not constitute investment advice. If it involves copyright issues, please contact us for deletion.
