アーム、インテル、AMDと格闘
Intelに打撃を与えたAppleは今週、製品ライン全体で自社のArmプロセッサへの移行を開始したことを発表した。しかし、Appleが今週のニュースを取り上げた一方で、Armはしばらくの間x86の拠点に侵入し、データセンター、クラウド、さらにはハイパフォーマンスコンピューティング(HPC)にまで進出しています。In a blow to Intel, Apple this week announced it had begun the transition to its own Arm processors across its product line. But while Apple grabbed the headlines this week, Arm has been encroaching on the x86 strongholds for some time, making major inroads in the data center, cloud, and even in high-performance computing (HPC).
The most powerful computer in the world is now powered by Arm. Meanwhile, cloud providers like Amazon and Microsoft, desperate for higher core counts and lower power consumption, have begun adopting Arm architectures and in some cases developing their own.
In addition to this week’s Apple news, Ampere today announced a 128-core version of its Altra data center chip that it’s calling Altra Max.
The Altra Max will share much in common with its 80-core sibling when it starts sampling in the fourth quarter of 2020. The new platform is based on the same Arm Neoverse N1 architecture and will be produced using Taiwan Semiconductor Manufacturing Co.’s (TSMC) 7-nanometer process, and will be available in both single- and dual-socket configurations.
But while Altra Max will push the core count to 128 and the thermal design performance (TDP) to 250 watts, the chip is designed to work in the same hardware, providing an easy upgrade path for customers.
According to Jeff Wittich, SVP of products at Ampere, despite the leap in core count and TDP, the new chips are expected to manage core-clocks consistent with Altra. Ampere’s 80-core, 210-watt Altra chips max out at around 3 GHz.
All the technologies are ready for Arm to rise to prominence, explained Wittich. “At this point, you’ve got the right architecture from an Arm perspective, you’ve got the focus on high-performance cores based on the Arm [instruction set architecture], you’ve got TSMC with 5 nanometer out there already,” he said.
However, the technology is only part of it. Slowing innovation coupled with growing demand for high-performance and low-power consumption compute is also driving Arm adoption, Wittich said.
“Last week [Intel] announced their third-generation of 28-core Xeon CPUs,” he said, referring to the rival chipmaker’s recent Xeon Scalable announcement. “They’ve had first, second, third generation of Xeon SP; they’re all 28 core. That’s not the kind of scaling that the cloud providers are looking for.”
Across the industry, in data centers, public clouds, and even in networking infrastructure like 5G base stations, Arm’s foothold is growing. Oracle, Microsoft, and more recently Equinix Packet, Cloudflare, and Scaleway have all announced plans to deploy Ampere’s Altra chips.
Earlier this month, Amazon announced general availability of its Graviton2 chips in the company’s Elastic Compute Cloud (EC2). Graviton2 is Amazon’s second in-house processor based on Arm’s 64-bit Neoverse core design and replaces the outgoing Graviton chip. However, the Graviton2 isn’t simply a rebrand or direct license of an Arm design: the Graviton2 features custom silicon designed by AWS for its role in the company’s cloud data centers.
Microsoft has also embraced Arm architectures within its Azure public cloud offering. Last year, the company announced it would offer instances running on Marvell’s ThunderX 2 platform.
Meanwhile, Marvell has steadily won support for its Octeon TX2 and Octeon Fusion infrastructure chips. The Arm-based chips are designed to power appliances like switches, routers, secure gateways, firewalls, and smartNICs, and in the case of the Octeon Fusion, 5G base stations.
Nokia and Samsung have been among the first to publicly partner with Marvell on the chips. Both companies plan to deploy the chips in their 5G base station hardware.
Arm’s reach goes well beyond the cloud and into truly high-performance territory as well.
Fujitsu and Riken’s Arm-based Fugaku supercomputer took the top spot in the latest supercomputing TOP500, and the cluster led the Green500 in 2019.
Fugaku will help to discover new drugs, analyze and predict weather patterns, simulate new design and production processes, and wage war against COVID-19. The supercomputer is expected to be fully available later next year.