At chip level, Intel has predictably driven the development of common cloud and edge architectures. The rapid expansion of cloud servers, whether huge or tiny, are essential to the company’s growth, and it needs to defend this territory against the webscale giants’ inhouse chip developments; the open source community led by RISC-V; the possible revival of IBM Power; and of course ARM.
Of course, Intel has other battles to fight in the telco cloud environment, such as stealing greater market share in the Ethernet switch-chip market, from Broadcom, Barefoot, OEMs’ designs and others. But the x86 processor remains the core of its business, and so far it has done a good job of keeping ARM-based challengers at bay. Qualcomm has put its server processor on the back burner for now; Marvell, courtesy of its Cavium purchase, is the main player.
But Huawei recently launched an ARM-based server processor, the Kunpeng 920, heavily targeted at cloud applications, and now ARM has launched two new platforms in its Neoverse family, the N1 and E1, targeting high performance cloud and edge, as well as comms, applications.
If ARM designs are to succeed in the data center, the most likely market will be cloud applications, which need the lean, power-efficient hallmarks of the ARM platform. The ARM-based server chips which have achieved sales have mainly targeted sectors like Cloud-RAN or the offload or acceleration of web processes, as well as comms, rather than more traditional data center processors, where Intel Xeon still reigns supreme. To date, SoCs based on ARM cores have made their deepest inroads in comms gear, where Intel is strong too; and in storage servers.
While ARM’s initial server-oriented processor cores were fairly generic, the latest Neoverse offerings are more specifically geared to cloud, edge and 5G workloads.
Neoverse made its debut in October, as the overall brand for ARM’s server processors, differentiating this from the general Cortex brand for all CPU designs. It scored a big hit a few weeks later when Amazon AWS said it had used the architecture for some of its A1 Graviton processors, which support new EC2 computing instances for its customers – at a lower cost than instances using Intel architecture.
Neoverse was originally on the 16-nanometer Cosmos design, but ARM says it will be pushing forward with a new generation each year. The new designs, which were codenamed Ares, are on 7nm, and then it will move to Zeus (7nm+), and then Poseidon (5nm). Initially, ARM said Neoverse would power “a new class of cloud servers that manage the networking, storage, and security workloads”, and with the AWS deal, it is expanding to application processor workloads too.
The commercial names for Ares are Neoverse N1 and E1 and Drew Henry, ARM’s general manager for infrastructure, focused on the message of a unified platform throughout the network, saying: “This N1 platform is really about the core compute you need in the hyperscale data center, at the 5G base station, or at an Internet gateway.” For a data center server system-on-chip (SoC), ARM expects vendors to use 64 to 128 cores running beyond 150W. For storage, security, and large edge nodes, systems would typically run eight to 32 cores at 25W to 65W.
ARM says this chip delivers 2.5 times the performance for a cloud workload such as Nginx, as the previous generation. It can scale from four to 128 cores, giving chipmakers the ability to build more diverse products by adding accelerators or additional functions using their own on-chip custom silicon.
This hints at a more flexible approach than ARM has for its device processor IP (or than Intel has). Rather than its partners choosing between standard silicon, or the expense of an architectural licence which allows them to customize the designs, the strategy for Neoverse seems more nuanced, reflecting the complexity of the market, with a huge variety of feature combinations to be considered. But there is also the need to incorporate a wide array of functions in the core design to meet the hefty demands of the cloud computing platform, so Neoverse N1 has virtualization technologies, performance and management technologies, reliability and service support integrated.
Robert Dimond, a system architect at ARM, told EETimes that the carrier edge is the most uncertain area for Neoverse, because so far very few telecoms or cloud operators have built this infrastructure at any scale. He said the N1 is designed to be flexible enough to support many types and sizes of edge node, adding: “There has to be compute at the edge for services to scale … but the reality is that different uses cases will have different needs.”
There is a second new Neoverse design, the E1, which is optimized for throughput performance – 2.7 times better than in its predecessors, said ARM. It also promises 2.4 times better throughput efficiency and over twice the compute performance.
This is aimed particularly at telco networks and at smoothing the transition from 4G to high throughput, fully virtualized 5G infrastructure. It will target equipment from 16-core SoCs running at 15W for gateways or 35W for 5G base stations; going up to 32-core versions which could run the data plane for routers with multiple 100Gbps Ethernet ports.
In networking, embedded processor vendors such as NXP and Texas Instruments have largely migrated from proprietary cores to ARM.
However, while the N1 and E1 hit ARM’s self-imposed targets of 30% performance gains every year, they may still be over 30% below the performance of Intel Xeon equivalents, according to chip analysts Linley Group, as cited by EETimes. However, another analyst firm, Tirias Research, said the N1 had significantly closed the single-thread performance gap with Intel’s Core CPUs, while catching up with Xeon in other areas such as virtualization and security features.
Last fall, ARM said it expected one million Neoverse-based chips to have shipped in 2018. No confirmation of that yet, but the company is certainly starting to extend the reach of its server processor designs – after several years of uphill struggle – to a wider range of partners, and looking beyond its traditional licensee base. Building on the AWS relationship is critical of course – the webscale giants are the biggest purchasers of cloud and data center processors, but are also interested in designing their own chips, which could be a strong opportunity for an IP company like ARM.
Also from the cloud world, Red Hat now believes ARM is on a par with other architectures to support its software. Jon Masters, chief ARM architect at the Linux company, said it had been a 10-year journey to make ARM “a first-class architecture along with x86, IBM Power, and IBM Z in Red Hat Linux”.
Huawei could also be important – even more so if the USA follows through on threats to limit the Chinese company’s ability to buy components from American companies (like Intel). ARM is owned by a Japanese parent, Softbank, and has its Chinese IP and business carefully separated thanks to the transfer of some assets to an autonomous subsidiary, majority-owned by local investors. Huawei is the third largest buyer of semiconductors, of all kinds, in the world after Samsung and Apple. Many of these are device-related, but the firm aims to be a powerhouse in all kinds of infrastructure, building on its existing telecoms and enterprise business. Indeed, Huawei’s own core MWC message this year is a cloud-to-edge one, with chips, equipment and software that aims to span every size of server, including those which will host specialized functions like virtualized base stations and cores.
“Heterogeneous computing is really where the industry is moving,” Drew said. “In this post-Moore’s Law world you need to be able to provide a very flexible design that allows our customers to really build what they want to build, and that’s why we are seeing such remarkable things happening in the silicon ecosystem.”