Refactoring the cloud computing base again, Amazon Graviton3!

"Follow" and "star" us,

Receive the latest news about Amazon Cloud Technologies every day!

2021 Amazon Cloud Technology re:Invent global conference, the latest generation of cloud-native general-purpose processors, Amazon Graviton3, and related computing instances were released. From Amazon Graviton in 2019, Amazon Graviton2 in 2020 to today's Amazon Graviton3, every time from the chip, the computing service has been continuously improved. Compared with Amazon Graviton2, Amazon Graviton3 integrates 55 billion transistors, computing performance will be increased by 25%, floating point and encryption performance will be increased by two times.

On the machine learning side, Amazon Graviton3 includes support for bfloat 16 data and will be able to deliver up to 3x the performance. in performance, Amazon Graviton3's energy consumption has dropped by 60% compared to the previous generation.

To make the processor more powerful, the industry has different approaches.

first method is to increase the processor frequency. higher frequency of 161e94ec99e4ab means that the processor can run faster, but it also means that the chip will use more power, and higher power means higher cost, more heat, and lower efficiency. In this context, how did Amazon Cloud Technology choose Amazon Graviton3 when designing it? Our choice is to increase the performance of a single core, thereby increasing the number of workloads done in each cycle. so the vast majority of the workload can be promoted in a 25% performance boost on Amazon Graviton3.

second solution is to add more cores, but often more cores have lower performance in real workloads. When we did the test, we found that the memory bandwidth will greatly affect the computer type, so in Amazon Gravtion3, we do not use more cores, but improve the memory bandwidth.

Powered by the new Amazon Graviton3 processors, C7g instances are also the world's first instances to feature the latest DDR5 memory, delivering up to 25%

C7g instances powered by Amazon Graviton3 processors deliver up to 25% faster performance for compute-intensive workloads than current-generation C6g instances powered by Amazon Graviton2 processors. Amazon Graviton3 processor compared with Amazon Graviton2, as scientific computing, machine learning, and media encoding workloads floating point performance of up to 2 times, encryption workload up to 2 times faster up to 3 to the machine learning workload times the performance. Amazon processor Graviton3 energy efficiency is also higher, with the same type of comparison Amazon EC2 instance, can save up to 60% of energy consumption.

Compared to Amazon Graviton2-based instances, instances 161e94ec99e55d increase memory bandwidth by 50%, thereby improving the performance of memory-intensive applications such as scientific computing. also has 20% higher network bandwidth for C7g instances compared to Amazon Graviton2-based instances. C7g instances support the Elastic Fabric Adapter (EFA), which allows applications to communicate directly with network interface cards, providing lower and more consistent latency and improving the performance of applications requiring massively parallel processing such as HPC and video encoding. C7g instances are now in preview within a week of the Amazon Graviton3 release.

“Twitter hosts what’s happening and people are talking about right now. Twitter is working on a multi-year project to provide a Twitter timeline using Amazon EC2 instances based on Amazon Graviton processors. To further improve efficiency, we tested new C7g instance of Amazon Graviton3 processor.

In a series of Twitter can represent the workload performance benchmark tests, we found examples of Amazon Graviton3 processor based C7g compared with the example of Amazon Graviton2 C6g based processor, can be increased by 20% -80% performance , but also tail latency by up to 35% . We are very excited and look forward to better price/performance with Amazon Graviton3 processor-based instances. "

- Nick Tornow, Director of Platforms, Twitter

Founded in 1991, Epic Games is the creator of the Fortnite, Unreal, Gears of War, Shadow Complex and Infinity Blade series. Epic's Unreal Engine technology brings high-fidelity, interactive experiences to PC, console, mobile, AR, VR and the web.

(Image source: Fortnite official website)

"As we look to the future to build more engaging and immersive experiences for players, we're excited to use the Amazon Gravtion3 processor-based Amazon EC2 instances. Our tests show that they are suitable for even the most demanding and latency-sensitive workloads, while providing excellent price/performance, enhancing the experience of and any virtual engine-based game creation. 161e94ec99e675 ”

- Mark Imbriaco, Senior Director of Engineering, Epic Games

Amazon Cloud Technology's continuous investment and upgrade of self-developed chips has enabled many users to obtain huge cost-effective advantages in some key workloads. users expect us to keep pushing the boundaries with each new generation of Amazon EC2 instances. Cloud computing innovations starting at the semiconductor level are game-changing capabilities. The Amazon Graviton series of cloud-native processors have verified the entire process from chip design and server joint development to rapid deployment on the cloud to provide external services. Continued innovation at Amazon Cloud Technologies gives customers the opportunity to run their critical workloads with these new, game-changing instances for better price/performance.

leads the trend and reshapes the future

2021 Amazon Cloud Technology re:Invent Global Conference

One stop!

Learn about cutting-edge trends in cloud computing

Scan the QR code below

View full playback