Build a global manufacturer and supplier trusted trading platform.

Amazon released a new server chip Graviton3: machine learning performance is three times the original, energy consumption reduced by 60%

12/03/2021 Amazon, server chip, Graviton3, machine learning

Amazon released a new server chip Graviton3

  Recently, Amazon AWS (Amazon Web Services) in re:Invent 2021 conference, officially released its third generation of self-researched server chip Graviton3, the simultaneous launch of three Graviton3-based Amazon Elastic Compute Cloud (Amazon EC2) executive Individuals.

  As early as 2018, Amazon launched the first generation of self-researched Arm server chip Graviton, the launch of Graviton3 is already the third generation of AWS self-researched server chip.AWS said that Graviton3 compared to Graviton2, computing performance increased by 25%, floating point performance and encryption performance is twice as much as the original, machine learning performance Graviton3 also adds new pointer authentication capabilities to improve security.

  Although AWS did not detail the specific parameters of Graviton3, however, sources indicate that Graviton3 uses the Neoverse N2 core with ARMv9 architecture and supports DDR5 memory.

  Graviton3 will provide support for AWS' EC2 C7g instances for compute-intensive workloads such as HPC, EDA, distributed analytics and CPU-based machine learning inference. EC2 C7g instances offer 30 Gbps of network bandwidth and Elastic Fabric Adapter (EFA) support and are currently running in preview.

  AWS also introduced the latest Trn1 instances based on its 2020 in-house AI chip, Trainium, which is primarily used for machine learning model training and supports Trn1 instances that offer customers the best price/performance and fastest machine learning model training speeds for deep learning model training in Amazon EC2; compared to the current P4d implementation Compared to current P4d implementations, the cost of training deep learning models with Trn1 instances is reduced by up to 40 percent, competing with products such as Nvidia's chips.

  In addition, AWS's EC2 M6a instance uses AMD's Zen 3 architecture-based EPYC 7003 series processor. Compared to the previous generation of EC2 M5a instances, the price/performance ratio has improved by 35%, and the cost is 10% lower than other x86 processor-based EC2 instances. It is understood that AWS plans to launch more instances with AMD's third-generation EPYC series processors in the future.

Parts of materials of this site come from the internet, please contact if there is infringement

QR code