AWS Unveils Graviton4 and Trainium2 for Faster, Affordable AI Model Building
At re:Invent in Las Vegas, Amazon Web Services (AWS) announced two new AI chips, Graviton4 and Trainium2. These chips bring advancements in price performance and energy efficiency for various workloads, including machine learning training and generative AI applications.
- Up to 30% better compute performance
- 50% more cores
- 75% more memory bandwidth than Graviton3
- Up to 4x faster training than the first generation
- Deployment capability in EC2 UltraClusters of up to 100,000 chips
David Brown, VP of Compute and Networking at AWS, emphasized that Graviton4 is the most powerful and energy-efficient chip ever built, underscoring the critical role of silicon innovation in customer workloads.
- Over 50,000 customers for Graviton
- Microsoft recently launched Azure Maia 100 AI Accelerator at Ignire 2023
- Naveen Rao, VP of generative AI at Databricks, praises AWS Trainium for scale and high performance at a low cost.
- Roman Visintine, lead cloud engineer at Epic Games, commends Graviton4 instances as the fastest EC2 instances tested.
Juergen Mueller, CTO of SAP SE, shares that during the migration of SAP HANA Cloud to AWS Graviton-based Amazon EC2 instances, they observed up to 35% better price performance for analytical workloads.
- Graviton4-powered R8g instances are available in preview today, with general availability planned in the coming months.
- Trainium2 is expected to be available in Amazon EC2 Trn2 instances.
In essence, these new AWS chips signify a leap forward in AI model building, offering faster and more affordable solutions for a wide range of applications.