Arm in the cloud definitely a trend now with Google Cloud’s embrace
Six of the world’s largest cloud service providers head on out with Arm arch
Comment It's been a rocky year for Arm. First, the British chip designer lost a financial boost with its sale to Nvidia killed by regulator scrutiny. Then Arm laid off staff as it made plans for an initial public offering, and now market conditions aren't looking great for that IPO.
The good news for Arm is that the cloud world has been increasingly warming up to the alternative instruction set architecture. The most recent sign: Google Cloud's introduction on Wednesday of its first Arm-based cloud instance, which the cloud service provider said will "deliver exceptional single-threaded performance at a compelling price."
Meant for "scale-out, cloud-native workloads," Google Cloud's Tau T2A virtual machines are powered by Ampere Computing's Arm-based Altra CPUs. This means the Arm-compatible VMs, available now for preview in the US and Europe, are meant to provide a strong performance-cost ratio for things like web servers, containerized microservices, media transcoding, and large-scale Java applications.
Google Cloud seems pretty stoked about what Arm can bring to the cloud world, given that it plans to let customers and partners try the T2A VMs for free under a trial period to "help jumpstart development." Even when T2A becomes generally available later this year, Google Cloud said it will "continue to offer a generous trial program that offers up to 8 vCPUs and 32 GB of RAM at no cost."
The T2A instance is part of Google Cloud's Tau VM family that debuted last year with instances running on AMD's third-gen Epyc Milan CPUs. The Arm-based instance type supports up to 48 virtual CPUs per VM and 4GB of memory per vCPU, and the networking bandwidth can go up to 32 Gbps. It also comes with a "wide range of network-attached storage options."
There are, however, some limitations for T2A, which also exist for the AMD-based T2D instance: no support for extended memory, sole tenancy, nested virtualization, nor custom VM shapes.
While Google Cloud didn't provide any performance comparisons to x86-based instances, Ampere leapt up and said a T2A instance with 32 of its vCPUs was up to 31 percent faster than Google's N2 instance using Intel's Ice Lake silicon with the same number of vCPUs. This was based on an estimated score for the standard SPEC CPU 2017 Integer Rate benchmark.
Using the cloud provider's VM pricing guide, Ampere said a T2A instance provides up to 65 percent better price-performance than the Intel-based N2 instance for on-demand pricing.
What about the software support?
As the cloud world has been largely rooted in x86 chips for most of the time, it's right to wonder how Ampere's Arm-based Altra CPUs can handle a wide range of software.
To that end, Ampere is doing its best to give people confidence that its processors are up to various cloud tasks. In a Wednesday blog post, the company noted how "the Arm-based server ecosystem has rapidly matured over the last few years with open-source cloud native software stacks extensively tested and deployed on Ampere Altra-based servers."
TrendForce: AWS to give Arm a leg up to 22% of datacenter servers by 2025READ MORE
"For example, Ampere runs over 135 popular applications across 5 different cloud native infrastructures to ensure that our customers have confidence in the Ampere software environment across the marketplace," wrote Jeff Wittich, Ampere's chief product officer.
The startup's server chips also supports several versions of Linux, including Ubuntu, Red Hat Enterprise Linux, and CentOS Stream.
Wittich pointed out that Ampere has a section on its website with a large list of applications, programming languages, and other kinds of software that have been tested on its Arm CPU cores.
Google Cloud did manage to get testimonials from a few independent software developers who said porting their code to T2A has been easy.
"We were pleasantly surprised with the ease of portability to Arm instance from day one. The maturity of the T2A platform gives us the confidence to start using these VMs in production," said Khawaja Shams, CEO of Momento, a startup providing serverless caching services.
T2A also got the nod of approval from the world of academia, with Harvard University Research Associate Christoph Gorgulla saying the "improved price-performance" of the instance helped his team "screen more compounds and therefore discover more promising drug candidates."
Several major cloud providers now on the Arm bandwagon
With the latest introduction of Arm-based cloud instances, the British chip designer's ISA is now supported by six of the world's largest cloud service providers: Amazon Web Services, Microsoft Azure, Google Cloud, Alibaba Cloud, Tencent Cloud, and Oracle Cloud. Other cloud providers are getting behind Arm too, such as JD Cloud, UCloud, and Equinix Metal.
All of this means it's very safe to say that cloud providers adopting Arm is definitely a trend now.
This is a development that would have been unthinkable to some people a decade ago, as GitHub engineer Jaana Dogan put it on Twitter.
If you asked me 10 years ago that ARM may potentially be the new go-to choice for servers, I'd assume you are from a parallel universe. It happened, quickly and smoothly. Graviton's success paid a huge role making this happen. https://t.co/WTmFEDBsu7
— Jaana Dogan ヤナ ドガン (@rakyll) July 13, 2022
Getting Arm chips into server-grade environments, running operating systems such a Linux, has taken a large amount of cooperation between software and hardware worlds primarily to agree on and stick to a standard base of features and expectations in these computers. This has made building and running software on Arm systems, particularly server boxes, relatively boring: it should just work like x86 just works, and it seems to do so.
AWS also helped paved the way for Arm's rise in the cloud with its decision to design an Arm-based server CPU in house using the talent it gained from Amazon's 2015 acquisition of chip designer Annapurna Labs. The cloud giant is now on the third generation of its Graviton chip, which is available in Elastic Compute Cloud instances now and for which it continues to make big price-performance claims against x86 chips.
That said, when considering all the other major cloud providers introducing Arm-based instances, plus some of the smaller ones, there's one common element linking them: Ampere Computing.
Founded by former Intel executive Renee James, the Silicon Valley-based startup recently said growing support for its Altra processors by a variety of businesses and cloud providers shows that the chips are better suited for cloud applications than Intel's or AMD's.
Like Arm, Ampere is also planning an IPO at some point, assuming that market conditions eventually get better. If you're curious about some of the ways Ampere's chip designs are a good fit for cloud applications, we suggest you read our recent interview with Ampere exec Jeff Wittich.
While the cloud world's growing embrace of Arm is a welcome sign for anyone tired of Intel's dominance over the space, the question now is how long Arm and silicon partners like Ampere and AWS can keep this momentum going. After all, Intel and AMD both have plans to introduce specialized cloud chips in the near future, and who knows, maybe RISC-V can shake things up even further. ®