• Skip to main content
  • Skip to search
  • Skip to footer
Cadence Home
  • This search text may be transcribed, used, stored, or accessed by our third-party service providers per our Cookie Policy and Privacy Policy.

  1. Blogs
  2. Breakfast Bytes
  3. The Start of the Arm Era
Paul McLellan
Paul McLellan

Community Member

Blog Activity
Options
  • Subscribe by email
  • More
  • Cancel
systemready
arm devsummit
project cassini
neoverse
ARM

The Start of the Arm Era

19 Oct 2020 • 5 minute read

 breakfast bytes logoSometimes, you attend an event and it feels like you are present at the start of a new era that will change some aspect of the technology industry. Of course, things don't change overnight. One event I remember from the last decade were hearing Krste Asanovic laying out what was going on in RISC-V. I realized that RISC-V would be big. Another was seeing Yann LeCun pointing a little camera on his PC during a keynote at the Embedded Vision Summit in 2014 and a neural network identifying what it was the camera was looking at: a shoe, a mouse, a coffee cup. I realized that neural networks and deep learning were going to be significant.

At the Arm DevSummit recently, it felt like the start of the Arm era. It wasn't a single presentation, it was several facts:

  • AWS's Graviton 2 is 40-50% faster than x86 (see my post Xcelium Is 50% Faster on AWS's New Arm Server Chip for details).
  • Several EDA workloads at Arm, AWS, and other companies are running on Graviton 2 instances (see my post EDA on AWS Graviton).
  • The world's fastest supercomputer, Fugaku, is Arm-based (see my post Japanese Arm-Powered Supercomputer Takes the TOP500 Crown for details).
  • Netflix is running some workloads on Graviton 2. Here's part of a recent quote from Netflix's Ed Hunter: "We use Amazon EC2 M instance types for a number of workloads inclusive of our streaming, encoding, data processing, and monitoring applications."
  • Lurking in the wings is whatever 5nm chip Apple comes up with for its Arm-based Macs, which it has already said will be before the end of the year.

Richard Greenthwaite put it succinctly in a panel session on Thursday when he said what he was most looking forward to in the future was:

The fact that we have started to break out of our homeland of the mobile phone. We are being taken seriously in servers and supercomputers. This vision of being the architecture of the digital world. It's a big opportunity.

When Arm first started about getting into servers, the value proposition was lower power, smaller physical footprint, lower cost. In a sense, it was always that Arm servers would be better value than x86 servers, not that they would outperform them. For example, see my 2016 post How ARM Servers Can Take Over the World. For internet workloads, serving hundreds or thousands of users at the same time, single-thread performance is less important than having lots of threads that can run at the same time. But with Intel's well-known 10nm process problems, the top-of-the-line Arm processors seem to be faster even for single-thread performance. Of course, Intel still has something like 95+% market share in data centers, I don't want to over-exaggerate the situation.

Chris Bergey

 One of the keynotes on Thursday morning at Arm DevSummit was a presentation by Chris Bergey, GM of Infrastructure at Arm, titled A New Infrastructure for a New Era. He started with a medley of previous announcements: Neoverse, AWS Graviton, partnerships with VMWare, Marvell, NVIDIA's CUDA on Arm, AWS instances on Graviton 2, ORAN for 5G, Fugaku supercomputer, the expanding software ecosystem. "We're just getting started," the final screen said.

Chris pointed out all the usual trends—5G, AI, autonomous driving—and pointed out that it is the invisible fabric that underlies this that makes everything "just work". 

Arm is on a journey to bring together an ecosystem for building tomorrow's digital infrastructure...reinventing the digital infrastructure is one of the biggest opportunities in technology today. It's also one of the biggest challenges.

He then brought up some partners. Kingsum Chow of Alibaba, Mark Shan of Tencent Open Source and TARS Foundation, Priyanka Sharma of Cloud Native Computing Foundation.

Chris said that Arm thinks that there are three key pillars to building this next-generation infrastructure:

First, we need silicon that is targeted at specific performance goals and use cases. Second, we need to unify an ecosystem around standards. And third, we must work relentlessly to make software and product development as frictionless as possible.

He looked at each of these in turn. Internet traffic and data center workloads have gone up 8X and internet traffic 12X. But power has remained flat due to server consolidation, virtualization, and so on.

Looking forward to the next ten years, Chris had the graph above. Workloads and traffic will continue to double every two to three years, but power could also go up by an order of magnitude without new ideas and new technology.

But we think this is a future we can avoid. Neoverse is our platform for servers, networking, and infrastructure equipment for solving the power dilemma. Neoverse is designed to deliver maximum performance, while minimizing power, space, and other resources.

He moved onto the Neoverse roadmap (above). "Expect regular uplifts of 30% or more".

Chris moved onto the second pillar, standards. Earlier in the week Arm had announced SystemReady for building a wide range of solutions. SystemReady is part of Project Cassini, ensuring a cloud-native experience across the edge ecosystem. You should be able to write an application once, and have it run on infrastructure from a wide range of manufacturers from cloud data centers to the edge. By the way, Cadence was one of the companies on the logo slide of partners to SystemReady.

The third pillar is frictionless development. Arm has been working with a big ecosystem (see some of the logos) to ensure that languages and libraries are available. In this context, Cadence appears as a workload. See my earlier post EDA on AWS Graviton.

One difference about the cloud, compared to the older release model, is CI/CD which stands for "Continuous Integration, Continuous Development". 

The result is a big increase in performance across the board (54% for Xcelium, as Chris highlighted in the top center of the picture).

Chris thanked everyone for all the work that they have contributed to making this happen.

Learn More

Start at developer.arm.com/infrastructure

UPDATE: A "Designed by Cadence" video about the Fugaku Supercomputer:

 

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.