On Wednesday, September 14, Arm held a press conference to announce the latest update to its server CPU roadmap, which the company calls the “Neoverse” core. The key revelation at the press conference was the V2 core (codenamed Demeter), which will power Nvidia’s “Grace” CPUs for high-performance computing (HPC) and artificial intelligence (AI) workloads.
Arm Neoverse V, N, and E series cores target three different performance and power tradeoffs. V-cores provide higher performance per core with higher power and larger die area. The E-core is designed to significantly improve power and chip space efficiency. N cores are in between the other two types of cores, balancing performance, scale, and power. Arm has updated the roadmap for three Neoverse CPU cores to support the latest platform technologies: PCIe (Gen 5 and Gen 6), CXL (2.0 and 3.0), and DDR5 memory.
Ampere computing leads the way
Arm discussed several new key customer design wins — including Microsoft Azure, Google Cloud — but these companies all use Ampere Altra products. Ampere Computing is the only CPU company currently offering commercial (standalone) Arm server chips to cloud service providers.
For those unfamiliar with the company, it’s a privately held startup led by founder, chairman and CEO Renée J. James. She founded the company in 2017 after serving as Intel’s president. She was also the first female general manager of Intel’s business unit and the company’s first female executive vice president. She founded the company with a mission to fill a void in cloud computing and has assembled a team of veterans from Intel, Sun Micro and other leading companies.
Ampere Computing has launched two products in the past two years: the 80-core Ampere Altra in 2020 and the 128-core Ampere Altra Max in 2021. The company already sampled its latest product, the AmpereOne, to customers earlier this year. While Ampere Computing uses Arm’s Neoverse N1 core for its Altra series to get to market faster, the company’s real goal is to develop its own CPU cores using only the Arm ISA license.
The Ampere Altra family has proven that more CPU cores can be instantiated per socket using leading 7nm process technology and still achieve lower power consumption than competing server processor cores. This is a very attractive value proposition for cloud service providers because they can run more instances per rack and per watt. The Ampere Altra Max contains 128 physical cores on a single chip, and the performance of these cores scales linearly, as Ampere’s server chip design is optimized for cloud scaling, using an intelligent mesh network-on-chip (NOC) and extensive I/O and memory Bandwidth. Each CPU core is designed with sizable caches and consistently fast clock speeds. The chip’s power efficiency allows it to maintain consistent performance under heavy workloads.
Modern cloud infrastructure needs to be scalable and must extend seamlessly beyond the main data center. A hybrid multicloud architecture requires resources distributed from hyperscale data centers, on-premises private clouds, edge resources, and all points in between. Ampere “cloud-native” processing is optimized for this distributed environment, with consistent performance and low power consumption.
Cloud services (and others) running Ampere Altra (Max)
Arm-based cloud instances have achieved significant market share gains based largely on Ampere CPU chips. While early AWS Graviton instances released in 2018 first felt interest in Arm-based cloud instances, the real growth of late has been driven by companies using Ampere Altra processors. Oracle Cloud was one of the first to adopt Ampere chips, but more recently, in September, Microsoft released new Ampere-based Azure virtual machines, and Google Cloud released a preview instance of the same Ampere CPU-based T2A in August. HPE announced Ampere-based ProLiant systems at HPE Discover in July, and European web hosting company Hetzner announced Ampere-based instances in August.
Low-Power Computing Applications Based on Ampere Server Chips Extend to Autonomous Driving – Cruise’s autonomous vehicle development platform uses Altra processors to reduce power consumption while still achieving the required computing performance. Arm-based servers are also great for supporting Android cloud gaming. For example, Nvidia and Ampere are collaborating on a project called AICAN (Android-in-Cloud-with-Ampere-and-NVIDIA). The AICAN server platform uses Ampere Altra processors and NVIDIA GPUs to natively run Arm-compatible Android mobile games without modification or emulation. Another example: the ability to reduce operating costs and provide more cores enables Red Bull Racing to use Oracle Cloud and Ampere Altra for better, faster race simulations.
Arm-based server market share is growing, especially in cloud instances. While AWS has designed its own Arm-based Graviton processors, many other cloud service providers have relied on Ampere Altra products to pack more CPU cores per rack and reduce operating costs without investing in server CPU design teams. Ampere’s server chips give them all the benefits without the risk.
Arm’s new Neoverse roadmap, Alibaba’s Yitian 710, AWS Graviton 3, Nvidia’s Grace, and Ampere’s AmpereOne are leading the strong Arm server ecosystem. Alibaba and AWS chips are designed strictly for internal use, Nvidia Grace chips target HPC and AI workloads, and Ampere Computing chips focus on bringing predictable high performance and efficiency to the cloud and network edge.
Ampere Computing solutions should become even more attractive with AmpereOne, which uses a new custom CPU core based on the Arm instruction set, but is designed entirely by the company’s experienced design team and manufactured using advanced 5nm process technology. Details of AmpereOne are still being announced. Meanwhile, Ampere Computing’s plans to lead the development and deployment of cloud servers are just beginning.
Tirias Research provides tracking and consulting services to companies across the electronics ecosystem, from semiconductors to systems, from sensors to the cloud. Members of the Tirias research team advise AMD, Ampere, Arm, Intel, NVIDIA, Qualcomm, Synopsys, and other companies across the cloud and IP ecosystem.