AWS re:Invent is a week-long event filled with a lot of news – year in and year out. 2021 was no different as the company made several big announcements spanning storage to networking to compute. On the compute front, Moor Insights & Strategy Principal Analyst Patrick Moorhead gave a rundown on the company’s Graviton 3, Trainium, Inferentia, and Nitro SSD reveals. You can check that coverage out here. Paul Smith-Goodson weighed in on AI/ML and Will Townsend covered all things networking and 5G.
To not be repetitive in my Patrick’s coverage – let me give you a quick synopsis. Amazon is investing heavily in silicon design and development to support workloads across the enterprise natively. It is worth reading his (Patrick’s) piece to understand just how much of a silicon design house Amazon is.
I will spend the following few paragraphs explaining why I believe AWS innovations position the company inthe enterprise and what this could mean for the industry. I’ll also dig deeper into what significance this has for the larger Arm server play.
First – what I heard – the Cliffs Notes version
As mentioned, Arm made several compute-related announcements. As part of the company’s briefing, it oriented its latest round of innovations in three categories – support for every workload; building and scaling your global network; compute where you need it.
Each of these categories is worth digging into a little deeper. AWS enhanced its services portfolio to enable enterprise organizations to deliver workload-optimized compute platforms to customers – wherever needed and connected through secure, high bandwidth networking.
Announcements made on supporting every workload are precisely what you would think. Instance offerings that span the compute landscape. General purpose, compute intensive, memory-intensive, and storage-intensive.
- Want to utilize AWS for your machine learning (ML) needs? No problem – AWS claims the highest number of ML instances available in the world with five unique training instances and six unique inference instances. These instances are aided by AWS-designed silicon – newly introduced Trainium for deep learning acceleration and Inferentia for inferencing acceleration.
- How about EDA needs? AWS X2ezn instance delivers Intel Ice Lake-based high frequency (4.5GHz) to support the most demanding needs.
- And general purpose and compute-intensive workloads can benefit from the company’s new Arm-based Graviton3 chip. With Graviton3, the company claims 25% better price-performance v Graviton2.
The company’s networking announcements were focused on connecting endpoints quicker, more securely, and more easily, with monitoring and management tools to simplify the life of IT organizations.
AWS’s focus on “compute where you need it” is, to me, maybe the most significant area for the market to take note. This is where we stop talking about AWS as a cloud provider and start talking about AWS as an IT Solutions provider, competing with Dell Technologies, HPE, Lenovo, Cisco, and others.
If one takes a step back, it is easy to see how AWS now competes across the entire solutions stack – from silicon to hardware to software. And not in a “dipping its toes in the water” kind of way. Instead, the company is proving capable of delivering entire solution stacks to its customers – in the cloud, on the edge, or in the core datacenter.
And while it may have been completely obvious to every reader, re:Invent 2021 shows just how much of a player AWS is across the IT solutions value chain.
Second – consumption-based computing is the new norm
We have been talking about the commoditization of compute for a very long time. But maybe in the wrong context. While software architectures, frameworks, and delivery methods have rendered underlying architectures (e.g., Intel v AMD, x86 v Arm v POWER) less relevant – workload-specific performance needs have still driven architecture and CPU affinity and preference. In fact, Patrick Moorhead wrote about how AWS has differentiated on the compute front in his coverage of last year’s re:Invent
The packaging and delivery of optimized solutions via AWS Outposts, Azure Stack, HPE GreenLake, etc.… has begun to drive real velocity for consumption-based computing. The delivery of such services enables cloud economics and simplicity, on-prem.
Speak to any IT administrator supporting a somewhat modernized business – they aren’t buying servers. They are buying analytics platforms. Or hybrid cloud environments. Or AI solutions. Meaning, all-up solution stacks, prepackaged and optimized for performance. And in many cases – installed, monitored, and managed by the likes of AWS, HPE, Dell, or an integrator using these platforms.
So, the real question is this – as consumption-based computing continues to take root and expand – who wins? Ultimately, the customer. On the vendor side of the equation, The IT solutions provider that can deliver best performance at the lowest cost will be in a strong position. And AWS should play a prominent role as it can deliver solutions designed, developed, and optimized entirely on AWS technology. From silicon to hardware to software. That’s pretty compelling…and unique.
Third – I told you about Arm
I have been covering Arm since it first announced Neoverse – and have been “accused” of being overly optimistic on my assessments of Arm-based servers in the datacenter. I am not in a position to say “told you so,” but it does seem like the Arm Neoverse ecosystem is cranking. AWS is on a 12-18 month cadence of delivering Graviton (Arm-based) platforms that span the price and performance continuum. Ampere is finding a lot of success in the cloud, including Oracle Cloud Infrastructure (which can’t deploy its Arm instances fast enough to keep up with demand). Multiple cloud providers in China have built and expanded Arm in their respective services catalogs.
On the high-performance computing (HPC) and AI front, Arm has been equally successful. Its architecture stands atop the TOP500 for supercomputers and is well represented throughout the list. Further, Arm-based servers have achieved performance parity with x86 in the high-performance computing market.
One could easily argue that AWS paved the way for Arm’s success with its acquisition of Annapurna Labs and the release of Graviton. While the company continues to expand its Arm-based services, it brings Arm into the physical datacenter via Outposts. And as consumption-based computing continues to take root, look for Arm to gain share as vendors will be able to pass on price-performance advantages as a way to stay competitive.
Lastly, consider this. AWS’s Annapurna team can take its Graviton offering and flip bits to optimize for specific workloads – enabling features and instructions that favor a workload and turning off those that don’t. Performance goes up, and the price goes down. This is the beauty of Arm – an openness that allows for agility.
Finally – what does this mean for the enterprise?
The announcements made at AWS re:Invent 2021 reflect what AWS sees in enterprise IT. Organizations are in the throes of digitization efforts, and they require a new set of workloads and analytics tools to drive that transformation. Workloads and tools unfamiliar to IT on platforms that are too complex and too expensive.
The appetite for enterprises to simply consume these offerings as services, whether on-prem or in the cloud, is enormous. Just look at the growth of HPE GreenLake, Dell APEX, or Lenovo TruScale. Or, for that matter, AWS Outposts. Pre-configured solutions deployed and maintained for IT. Solutions where I only pay for what I consume. Nirvana for enterprise IT.
AWS continues to make the right moves to be a significant player in the modern, digitized business. Graviton3, Trainium, and Inferentia are examples of how the company is innovating in the silicon space. But in many ways, the significance of this silicon is less about the technology and more about how AWS owns the value chain.
It’s going to be a fun few years as we see how the IT solutions market evolves.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.