Amazon’s AWS News Highlights Shifts In Semis, Hybrid Cloud And 5G Edge

It’s rare to see multiple big picture tech issues addressed in a single vendor’s special event presentation; then again, it’s also rare to see 3-hour-long keynotes, especially in the virtual event COVID-19 era. At the kickoff to Amazon Web Service’s re:Invent conference, however, the company managed to do both. It covered an impressively wide range of critical cutting-edge computing concepts mixed in with a staggering number of new announcements in a surprisingly long speech by AWS CEO Andy Jassy.

Frankly, it was a bit much to fully take in during a single sitting, but in the end, it served the purpose of showing how much trailblazing work Amazon’s cloud computing business is doing, how broadly its scope has expanded, and how many industries it has begun to dramatically impact. From custom semiconductors to new methods of building and managing cloud-native applications to machine-learning platforms, storage architectures, database migration tools, IoT sensors and analytics, and of course 5G, there was literally something for nearly everyone in the keynote speech.

In particular, the company made some critical announcements around semiconductors and compute instances, new methods of creating containerized applications for public and private clouds, and the spread of computing resources across more physical locations that reflect the new realities of modern computing and point towards a more distributed computing future.

On the computing instances front—which refers to a set of specific hardware resources that AWS provides access to so that customers can leverage them for various application and development efforts—the company kicked things off the night before the official event start with the surprising news that it was going to offer its first-ever Mac instance. Leveraging Intel-based Mac Minis (not the new Arm-based M1 units—though apparently that will be coming sometime next year), the company’s new EC2 (Elastic Compute Cloud) Mac instance allows Mac developers, or those curious to try Mac application development, to gain access to the combination of Apple hardware along with Amazon’s collection of storage, networking, and other cloud-computing services. Like Amazon’s other flavors of compute instances, access is billed on a second-by-second usage basis, making it relatively inexpensive for companies to work on Mac-specific versions of their applications and services. Once the M1-based version of this becomes available, it could prove to be very attractive, given the strong initial interest in the Arm-based Macs.

Even more interesting was the announcement of new Habana Gaudi-based EC2 instances, which leverage the new dedicated AI accelerator chip technology that Intel purchased last year. The Gaudi chips are specifically designed to do training for deep learning models that do tasks like natural language processing, object detection, personalization and more. Up until now most of this work has been done with GPU-based instances, but Amazon claimed the new Habana-powered offerings enable up to 40% better price performance on machine learning workloads than their GPU-based EC2 offerings.

As if that wasn’t enough of a challenge to GPUs in the cloud, Amazon threw down the gauntlet even more strongly with the announcement of its own new AWS Trainium custom AI acceleration chip that, as the name suggests, is also designed for training purposes. As with the Gaudi-based offerings, the company is claiming that Trainium offers speed and cost benefits over existing GPU-based instances for machine learning workloads. Specifically, Trainium is expected to offer 30% higher throughput and 45% lower cost per inferences versus their current GPU offerings. In addition, Trainium—like the Inferentia custom inferencing chip that Amazon introduced last year—will work with the company’s SageMaker machine learning platform. In addition, Trainium offers support for the TensorFlow, PyTorch, and MXNet AI development frameworks, while the Gaudi-based offering only supports TensorFlow and PyTorch.

In the traditional computing space, Amazon also talked about its Arm-based CPU efforts with its latest generation AWS Graviton2 custom chip and new instances such as C6gn designed around it. While the company discussed the importance of its ongoing partnerships with both Intel and AMD, it also made clear its intention to focus more efforts on Arm-based offerings—particularly because of the price performance benefits (up to 40% versus existing x86 CPU-based instances) offered to Amazon’s customers.

What’s fascinating about all of these developments is that they reflect a sea change in how people are thinking about traditional computing architectures in cloud computing environments—particularly with regard to Intel for CPUs and Nvidia for GPUs. In addition, they highlight both how new data center-focused architectures like Arm, as well as custom silicon offerings, and even non-mainstream data center options like the Mac are starting to have a dramatic impact on how leading-edge compute environments are thinking about the future.

Speaking of which, Amazon also clearly thought about the future from a software and cloud architecture perspective as well and made several announcements related to the new reality of hybrid cloud. Though the company used to argue strongly against the concept, it’s clear based on many of the new offerings it unveiled at re:Invent that it’s chosen, in Microsoft-like fashion, to “embrace and extend” the concept of hybrid cloud to better match their own wishes. Specifically, the company talked about redefining hybrid to mean the process of distributing AWS hardware, software, and services to more physical locations—some of which are not under the company’s direct control.

Regardless of the semantics, new offerings like ECS (Elastic Container Service) Anywhere and EKS (Elastic Kubernetes Service) Anywhere allow companies to create, manage, and run cloud-native, container-based applications within their own data centers, which is a critical capability for enabling hybrid cloud environments. Unlike some of its cloud computing competitors, however, AWS did not offer much in the way of tools to ease the process of moving workloads to different providers, as would be necessary for simpler multi-cloud deployments. Amazon did, however, announce its new Proton service, which provides an easier and more consistent way to provision the necessary compute instances, storage options, and other necessary infrastructure components for both container-based and serverless applications in the cloud and on premise.

In addition to these software capabilities, the company also discussed new types of locations and hardware options for running AWS in different types of environments. Building on previous announcements around AWS Outposts, which is a full rack hardware solution managed by Amazon that’s designed to run AWS services on premise, the company debuted two much smaller one-rack and two-rack “appliance”-style options for running AWS in a wider variety of locations, such as at an individual restaurant, small factory, cell tower and more. The company also discussed ongoing efforts with its previously announced Wavelength offering, which is specifically designed for telco carriers to integrate AWS services via dedicated hardware boxes closer to the edge of their 5G networks. In particular, Amazon announced a new Wavelength Zone in Las Vegas in conjunction with its partner Verizon. Both the Wavelength Zones and the new AWS Local Zones (which are expected to reach a total of 16 cities across the US in 2021) are specifically designed to bring local computing resources to the edge for latency-sensitive applications, such as cloud-based gaming and others that 5G proponents have been discussing for some time.

The net-net of all the announcements is that Amazon is continuing to distribute its ever-expanding variety of computing resources and cloud-based services to a significantly wider audience, with the ultimate goal of making its offerings faster and more efficient. It’s an audacious plan but based on the company’s announcements and efforts to date, it’s clearly something that’s directly in their sights.

Disclosure: TECHnalysis Research is a tech industry market research and consulting firm and, like all companies in that field, works with many technology vendors as clients, some of whom may be listed in this article.