Arm-based chips are ubiquitous today, but the company isn’t resting on its laurels. After announcing its ambitions for powering more high-end devices like laptops a few months ago, the company today discussed its roadmap for chips that are dedicated to internet infrastructure and that will power everything from high-performance servers to edge computing platforms, storage systems, gateways, 5G base stations and routers. The new brand name for these products is ‘Neoverse’ and the first products based on this IP will ship next year.
Arm-based chips have, of course, long been used in this space. What Neoverse is, is a new focus area where Arm itself will now invest in developing the technologies to tailor these chips to the specific workloads that come with running infrastructure services. “We’ve had a lot of success in this area,” Drew Henry, Arms’ SVP and GM for Infrastructure, told me. “And we decided to build off that and to invest more heavily in our R&D from ourselves and our ecosystem.”
As with all Arm architectures, the actual chip manufacturers can then adapt these to their own needs. That may be a high core-count system for high-end servers, for example, or a system that includes a custom accelerator for networking and security workloads. The Neoverse chips themselves have also been optimized for the ever-changing data patterns and scalability requirements that come with powering a modern internet infrastructure.
The company has already lined up a large number of partners that include large cloud computing providers like Microsoft, silicon partners like Samsung and software partners that range from RedHat, Canonical, Suse and Oracle on the operating system side to container and virtualization players like Docker and the Kubernetes team.
Come 2019, Arm expects that Neoverse systems will feature 7nm CPUs. By 2020, it expects that will shrink to 5nm. What’s more important, though, is that every new generation of these chips, which will arrive at an annual cadence, will be 30 percent faster.