Everything You Need to Know about Graviton on AWS

NEW! Want to save up to 42% on your data lake costs? Learn about Graviton on Qubole.

Let’s take a look at the knowledge and the leadership of Amazon Web Services in the space of integrated service with respect to the other players in the market such as Google and Oracle, etc. The AWS Graviton IC is the new trend and Amazon is leading the paradigm of this.

What is Graviton?

Graviton is a type of integrated circuit (IC), tiny carbon chipsets on a backplane, computing processor architecture, designed and developed by Amazon Web Services (AWS), exclusively tailored to their global cloud infrastructure focusing on Big Data, Artificial Intelligence, Machine Learning, High-Performance Computing, Scientific Computation Dynamic Analytics Internet of Things, CDF and a lot like solutions.

Amazon is leveraging 64-bit Arm Neoverse cores, enabling ARM ICs to the cloud. AWS Graviton processors are custom-built to deliver the optimal price and performance ratio for workloads running in AWS.

Graviton AWS Architecture

Graviton 1: The graviton 1 architecture was built for web services, and microservices, also called offload computing. The vase Graviton architecture for AWS EC2 is loaded with enhanced security capabilities. With the first generation of Graviton, AWS built a processor tailored to deliver compelling cost savings to the customers using general-purpose instance types of applications requiring a small CPU and memory footprint.

Graviton 2: The graviton 2 was the enhancement of graviton 1 in terms of expanding software on a circuit type of deployment as it focused on specifics. AWS significantly expanded its features to cover additional instance types such as compute-optimized (C6g, C6gd, C6gn), general-purpose burstable (T4g), and memory-optimized (R6g, R6gd, X2gd), including the variants with NVMe-based local SSD storage.

Graviton 1,2:  These instances are capable of delivering up to 40% better price-performance compared to the equivalent Intel x86-based instances. With the introduction of this wide range of instance types, Graviton is now capable of catering to different workload requirements such as high-performance computing, in-memory caching databases, artificial intelligence, learning, and living real-time machine learning inference.

Graviton 3: This newest announcement offers a great advantage to any AWS customer capable of running their workload using an ARM-based architecture, NeoVerse, bflot16, FPGAs, physical and logical advancements, and technologies. With graviton 3, AWS is slowly planning and designing its own chips focused on big data, AI, ML, high-performance computing, scientific computation dynamic analytics, and IoT, all focused on design. Some of these computing technologies have been around for some time and have evolved and developed, but now we have re-designed them for AI and ML. Amazon came up with graviton 3 which is the first complete circuit designed, manufactured, tested, and assembled by Amazon.

Graviton Architecture: Graviton 3 architecture also includes another component called the outpost.  Amazon calls it the EFA or Extended Fabric Adaptor. These are brand new technologies that are exclusively driven by graviton 3 and they help ML and AL in allowing amazon to change the paradigm of cloud computing. The AWS outpost and AWS EFA are special instances, designed, and manufactured by AWS to extend the paradigm of the cloud where the cloud will come to us now. With a lower hourly price compared with equivalent instance types, Graviton can deliver:

• 7x more performance
• 4x more compute cores
• 5x faster memory
• 2x larger cache

The graviton family and mostly Graviton 3 allow AWS to offer solutions like the AWS Outpost and the AWS EFA that were announced in December 2021.

History of Apache Big Data, Hadoop ECO Stack with the IC Intent

Hadoop, HDFS, Hive, and the Apache Big Data stack, have long been running on Java Virtual Machine Hypervisor layer technology, Java, JVM for example. Around 12 years ago, IBM had the Patent GPFS and didn’t allow other companies to have a truly distributed file system, and thus all the work was done in Java.

To overcome this hypervisor compute limitation, companies had to surpass this java compute level, with the volume of data to come, and frame plans to have machine learning, living, learning, features, and run sets under one of the new artifactual intelligent frameworks to come.

In the early days, both leading chip manufacturers, Intel and AMD invested over $1.5B into ownership of Hadoop, HDFS. Somewhat forecasting the data explosions/implosion ahead, the ultimate goal, was to design a distributed file system (like HDFS) on the IC chipset itself to meet the demands and exceed compute limitations of a JVM cluster, distributed computing platform.

As of today AWS, Intel and others are on the race IC chipset path, in IC design and manufacturing. Basically, companies are taking the concepts of Hadoop/Big data, supercomputing, and putting them at the IC, chipset CPU in global architecture.

Patents are big deal/big money items of things, for example, IBM has over 70,000 awarded patents. From this, we can see, that companies like AWS, Databricsk, Snowflake, and Qubole have the same thought.

Data can become intellectual property for a company. And to consume that data is the key, regardless of where it is stored. This new architecture is the beginning of a shift in the next level of cloud computing. The real paradigm of the cloud is nothing but a big supercomputer.

A look at AWS Graviton Family Lineage and Value

1. 1. G5g: Best price performance for Android game streaming
      Powered by: AWS Graviton2
      Built for: Graphics applications including Android game streaming and ML inference.
   2. M6g, M6gd: Best price performance for general-purpose workloads with balanced compute, memory, and networking
      Powered by: AWS Graviton2
      Built for: General-purpose workloads such as application servers, mid-size data stores, microservices, and cluster computing.
   3. T4g: Best price performance for burstable general-purpose workloads
      Powered by: AWS Graviton2
      Built for: Broad range of burstable general-purpose workloads such as large-scale microservices, small and medium databases, virtual desktops, and business-critical applications.
   4. C7g: Best price performance for compute-intensive workloads
      Powered by: AWS Graviton3
      Built for: Compute-intensive applications such as high-performance computing, video encoding, gaming, and CPU-based machine learning inference acceleration.
   5. C6g, C6gd, C6gn: Cost savings for computing and network-intensive workloads
      Powered by: AWS Graviton2
      Built for: Compute-intensive applications such as HPC, video encoding, gaming, and CPU-based ML inference.