In a groundbreaking announcement at AWS re:Invent, Amazon Web Services, Inc. (AWS) introduced the next generation of its chip families—Graviton4 and Trainium2. These chips, designed by AWS, promise significant advancements in price performance and energy efficiency, catering to a wide array of customer workloads, including machine learning (ML) training and generative artificial intelligence (AI) applications.
Graviton4: A Leap Forward in Performance and Efficiency
Graviton4, the latest iteration in AWS’s chip design evolution, brings a host of improvements over its predecessor. Offering up to 30% better compute performance, 50% more cores, and a remarkable 75% increase in memory bandwidth compared to Graviton3, Graviton4 stands as the most powerful and energy-efficient chip in AWS’s arsenal. This innovation is set to empower customers with enhanced options to run virtually any application or workload on Amazon Elastic Compute Cloud (Amazon EC2).
The new chip boasts heightened security measures, encrypting all high-speed physical hardware interfaces. Graviton4 is tailored for a range of workloads, making it ideal for applications such as databases, analytics, web servers, batch processing, ad serving, application servers, and microservices. It will be available in memory-optimized Amazon EC2 R8g instances, offering larger instance sizes and significantly improving the execution of high-performance databases, in-memory caches, and big data analytics workloads.
Trainium2: Accelerating ML Model Training
Designed specifically for ML model training, Trainium2 takes a leap forward with up to 4x faster training performance compared to its predecessor. It is engineered to handle foundation models (FMs) and large language models (LLMs) with trillions of parameters. Trainium2’s capabilities are set to revolutionize the landscape, allowing deployment in EC2 UltraClusters of up to 100,000 chips. This massive scale enables the training of complex models in a fraction of the time, with a simultaneous 2x improvement in energy efficiency.
AWS Vice President of Compute and Networking, David Brown, highlighted the significance of these chip designs, emphasizing AWS’s commitment to real-world workloads that matter to customers. Graviton4 and Trainium2 showcase AWS’s dedication to providing the most advanced cloud infrastructure to its users.
Customer Impact and Collaboration:
Several prominent organizations, including Datadog, Epic, Honeycomb, Databricks, and SAP, have already embraced Graviton-based instances for their diverse workloads. AWS’s commitment to collaborating with companies like Anthropic, known for their responsible deployment of generative AI, further underscores the potential of Trainium2 in shaping the future of AI applications.
The unveiling of Graviton4 and Trainium2 at AWS re:Invent marks a significant milestone in chip design innovation. With heightened performance, improved energy efficiency, and a focus on real-world applications, these chips are poised to redefine the landscape of cloud computing. AWS continues to lead the way in providing customers with cutting-edge solutions, unlocking new possibilities in the realms of machine learning, artificial intelligence, and beyond.
