Note: This blog contains contributions from Patrick Moorhead, President, and principal analyst, Moor Insights & Strategy.
One major trend we see at Moor Insights & Strategy is for cloud giants to use and sometimes develop their own, specialized silicon to give them a competitive advantage. We’ve seen this at Microsoft Azure, Google GCP, and Alibaba. The advantage can be flexibility, cost and even performance and comes at great expense and some risk.
Amazon announced the immediate availability of EC2 instances on its Arm-based servers last week at AWS re Invent, attended by Moor Insights & Strategy president, Patrick Moorhead. This is significant as it marks the first major cloud provider deploying Arm at scale. It’s also significant as the CPU (named Graviton) was developed by Amazon’s Annapurna Labs. What does this mean for the market? Is the market finally ready to adopt Arm-based servers for enterprise-grade workloads? Moreover, how does this impact the CPU market as a whole?
What was announced?
The Graviton based instances are known as EC2 A1. Five instances are available, ranging from a single virtual CPU (vCPU) to 16 vCPUs and from 2GB RAM to 32 GB RAM are available. The AWS team positions these instances as being well suited for scale-out workloads and applications such container based microservices, web sites, cache fleets and scripting language-based applications (e.g., Perl, Ruby, Python, etc.). If an application is written in a traditional development language, it will need to be recompiled for Arm. It’s also important to note that Linux support will be limited to Amazon Linux 2, Red Hat Enterprise Linux (RHEL) and Ubuntu initially. Amazon has committed to broadening Linux support over time.
- Graviton is built on The Arm Cortex- 72 64-bit architecture
- 16 cores single-threaded cores (no symmetrical multithreading)
- 4, four core complexes
- 2MB of L2 cache
- 32GB RAM
What this means for AWS customers
AWS has the widest variety of compute instances on the market today. These include GPU, CPU, FPGA, and ASIC (Inferentia)-based instances at varying sizes and deployment methods. Graviton is yet another proof point that supports AWS’ desire to offer the broadest array of compute.
Does this validate Arm in the public cloud datacenter?
In a word, yes.
To fully appreciate the significance of Amazon’s announcement, consider the recent attempts of companies to deliver Arm-based server CPUs into the market. The list is long and distinguished: Qualcomm, Samsung, Broadcom, AMD, Applied Micro to name a few.
The genesis of Arm for the server market can very clearly be traced to the Austin chip startup, Calxeda. Its EnergyCore System on a chip (SoC) showed a lot of promise and even secured server design wins. But as my Moor Insights & Strategy colleague (and former VP of Marketing at Calxeda) Karl Freund says, “the market just wasn’t ready.”
Five years later, things have changed. Moreover, Amazon just proved that these companies were right in their market assumptions. Perhaps just ahead of the game. Alternatively, missing on execution. Or not in it for the long haul as time to revenue was long.
Will others follow AWS?
Other cloud providers have been exploring Arm-based offerings for some time, so I believe there is no doubt that others will follow the lead of AWS. The bigger question becomes whether other cloud providers will choose to go down the path of developing a CPU “in-house” or rely on the Arm silicon ecosystem.
CPU development is a long, complex endeavor and requires a significant investment in people and resources. AWS was able to achieve success through its acquisition of Annapurna Labs for roughly $350 Million in 2015. Through this acquisition, AWS was able to build a CPU to meet its very specific needs. Moreover, AWS can rely on this team to design and deliver innovative CPUs that can deliver market differentiation at a fraction of the cost of a third party.
What role does Nitro play
AWS Nitro is a set of tools (hardware and software) that allow hosts to dedicated maximum host resources to instances (a1.medium, a1.large, etc..) that are running. All of the CPU resources that are utilized in a normal environment are off-loaded to chips and software that, in turn, assure greater CPU performance and flexibility. In the future, this should make all compute architectures easier to integrate into AWS.
Does this have any implications for the enterprise?
In the near term, the enterprise datacenter will remain x86 dominated, as enterprise. IT is a fairly conservative market driven by risk-averse professionals. However, AWS’ adoption should soften the enterprise beachhead for Arm and its ecosystem of silicon and software partners.
Consider AWS as an innovator in Geoffrey Moore’s technology adoption lifecycle. Other cloud providers will range from innovators to early adopters. Arm will eventually find a home in the enterprise IT datacenter as “early majority” organizations adopt. I expect these “early majority” adopters will be enterprise customers that are aggressively deploying edge environments, as Arm is a dominant force in IoT. Also interesting are AWS initiatives like Outposts that pull public cloud infrastructure, on-prem.
Over time, Arm’s Neoverse should find a footprint in the enterprise datacenter, as organizations can better understand workload affinity and the areas where it provides a significant price/performance advantage.
What should Arm do with this first Neoverse “win.”
Amazon’s rollout of Graviton based instances is a big deal. However, it is one “win,” and Arm still has a lot of work to do to build the market for its Neoverse lineup (I outline some of this in my initial coverage of the Neoverse launch, here).
Arm’s secret to success in this CPU market is no real secret- continue to develop the Neoverse ecosystem, make the necessary investments, and be patient. Without a robust ecosystem of hardware and software partners, the innovations in the Neoverse IP will never be realized. Moreover, without the investments in the market, those ecosystem partners will struggle to find success. If Arm executes on these, it will find success in time.
How does this impact the CPU market as a whole?
Longer term, I believe Arm’s Neoverse will be viewed as a viable option for all segments of the markets – from tier 1 cloud providers to smaller organizations deploying edge environments. Heterogeneity will describe the datacenter of the future as Arm (and perhaps other architectures) co-exist with x86 in supporting the new and emerging workloads that drive the business.
One more thing…
It should be noted that Arm has found some success in the datacenter. Cavium (recently acquired by Marvell) developed the Arm-based ThunderX2 CPU which is being used in high-performance computing (HPC) clusters and scale out deployments at bare metal cloud providers such as Packet.
These are exciting times. Moreover, AWS has once again differentiated itself by being the first major cloud provider to offer Arm based instances at a fraction of the cost (up to 45% less than x86 instances). I suspect Amazon’s competition will be looking at the adoption rate of these A1.xx instances to inform deployment decisions.
This is also an exciting time for Arm. This announcement has surely injected a renewed energy into the Neoverse go-to-market effort. Let’s see how this translates into wins in the new year.