AWS takes on ML training and generative AI applications with custom silicon chips
Amazon Web Services has designed its own silicon chip to optimize performance for machine learning inference and generative AI applications with energy efficiency. The AWS Graviton4 is a new iteration of its own processor line, which is tailored for a wide range of cloud workloads with improved power performance and capabilities to handle complex compute-intensive tasks.
AWS highlights that companies like Discovery and Formula 1 are using Graviton-based instances. The company says that Graviton addresses the needs of handling larger in-memory databases and analytics workloads by offering improved compute, memory, and network capabilities. Moreover, Graviton4 will be integrated into the Amazon EC2 R8g instances, which are memory-optimized instances, for more virtual CPUs and memory than current generation R7g instances.
According to AWS, the Graviton4 offers up to 30 percent better compute performance compared to its predecessor. Additionally, it has a 50 percent increase in the number of processor core, allowing better parallel processing, making the silicon efficient at handling multiple tasks simultaneously. The AWS Graviton4 also has a 75 percent increase in memory bandwidth to handle large data sets.
“Graviton4 marks the fourth generation we’ve delivered in just five years, and is the most powerful and energy efficient chip we have ever built for a broad range of workloads,” says David Brown, vice president of Compute and Networking at AWS.
On the other hand, AWS Trainium2 is intended to provie high performance computing capabilities to accelerate the training of large models, including NLP, computer vision and other types of neural networks.
“And with the surge of interest in generative AI, Trainium2 will help customers train their ML models faster, at a lower cost, and with better energy efficiency,” Brown adds.
Anthropic, a company that launched AI assistant Claude, is collaborating with AWS to develop future foundation models using Trainium chips. It is anticipated that Trainium2 will be used for building and training these models.
Telco Systems to transform the edge with the launch of Edgility r6
Article Topics
AWS | generative AI | machine learning | ML | silicon chips
Comments