Apple is using Amazon's Trainium and Graviton chips to provide search services, and is also evaluating whether the Trainium2 chip can be used to pre-train AI models.
Amazon announced the general availability of its AI training chip,
Amazon EC2 Trn2 Instances and Trn2 UltraServers for AI/ML training and inference are now available | AWS News Blog
https://aws.amazon.com/jp/blogs/aws/amazon-ec2-trn2-instances-and-trn2-ultraservers-for-aiml-training-and-inference-is-now-available/
AWS-Trainium2-Instances-Now-Generally-Available - US Press Center
Apple says it uses Amazon's custom AI chips
https://www.cnbc.com/2024/12/03/apple-says-it-uses-amazons-custom-ai-chips-.html
Amazon AI chips may help train Apple Intelligence in the future
https://appleinsider.com/articles/24/12/03/apple-considers-using-amazon-chips-to-train-future-apple-intelligence-models
Exclusive | Amazon Announces Supercomputer, New Server Powered by Homegrown AI Chips - WSJ
https://www.wsj.com/articles/amazon-announces-supercomputer-new-server-powered-by-homegrown-ai-chips-18c196fc
During the event, Amazon announced that the AI training chip 'Trainium2', announced in November 2023, has become generally available. Trainium2 is the latest version of the machine learning chip 'Trainium', and is characterized by training up to four times faster than before. At this event, it was reported that in addition to being available on 'Amazon EC2 Trn2 instances' equipped with 16 Trainium2 chips in a single instance, 'Amazon EC2 Trn2 UltraServers' equipped with 64 Trainium2 chips will also be generally available.
Amazon announces AWS processor 'Graviton4' and AI training chip 'Trainium2' - GIGAZINE
'As the number of parameters in AI models grows to trillions, we understand that customers need new approaches to train and run these large-scale workloads,' said David Brown, vice president of Compute and Networking at AWS about Amazon EC2 Trn2 UltraServers. 'The new Trn2 UltraServers provide the fastest training and inference performance on AWS, helping organizations of all sizes train and deploy the world's largest models faster and at lower cost.'
Benoit Dupin, Apple's senior director of machine learning and AI, also took to the stage at the event and talked about how Apple uses AWS. According to Dupin, Apple has been using AWS for services such as Siri, Apple Maps, and Apple Music for over 10 years, and also uses Amazon's AI chips Inferentia and Graviton for search services. Dupin reported that the introduction of these AI chips has actually led to a 40% efficiency improvement.
Dupin also revealed that the company is evaluating whether Trainium2 can be used to pre-train its own AI models. 'In the early stages of our evaluation of Trainium2, we expect pre-training to improve efficiency by up to 50%,' Dupin said. In fact, AWS CEO Matt Gurman told CNBC that Apple is a beta tester for Trainium2, saying, 'Apple came to us previously and said, 'How can you help us with our generative AI capabilities? We need the right infrastructure to build AI.''
'We have a strong relationship with Amazon. Amazon's infrastructure is highly reliable, enabling us to provide high-quality service to customers around the world,' Dupin said of Amazon.
Amazon also revealed that it will offer 'Trainium3' built on a 3nm process node in the second half of 2025, and is building a cluster of Trn2 UltraServers called 'Project Rainier,' 'EC2 UltraCluster,' in collaboration with Anthropic, an AI company developing large-scale language models such as 'Claude.'
According to Amazon, EC2 UltraCluster uses hundreds of thousands of Trainium2 and has processing power more than five times that of the cluster used by Anthropic. Amazon said, 'When completed, EC2 UltraCluster will enable Anthropic to build and deploy future models as the world's largest AI computing cluster ever reported.'
According to the Wall Street Journal, AWS's efforts are being carried out at its Annapurna Labs in Austin, Texas.
What is 'Annapurna Labs', where Amazon is developing its own AI chips to reduce its reliance on NVIDIA? - GIGAZINE
Related Posts:
