In October 2018, Arm introduced Neoverse, a dedicated brand and roadmap of processor IP for cloud, 5G network and edge infrastructure. The company pitched it as the foundation for infrastructure serving a world of a trillion connected devices.
The launch was led by Drew Henry, then Arm’s senior vice president and general manager of its Infrastructure Line of Business. He said Neoverse would let partners push compute to where it was needed, store data where it made most sense and connect networks securely.
The players and their pitch
Arm, whose designs are licensed to chipmakers rather than sold as finished chips, said Neoverse was aimed at hyperscale cloud data centres, storage and 5G networks. Its stated design principles were threefold. The IP would be built for cloud-native and networked workloads with security in mind. It would be tuned for leading-edge process nodes. And a broad ecosystem would share software, tools and silicon platforms.
Arm also argued from market position. It said it was already the largest architecture deployed in global internet infrastructure, with close to 30 percent unit share. Patrick Moorhead, principal analyst at Moor Insights & Strategy, said at the time that data centres were becoming distributed and connected, and that purpose-built IP should put Arm in a good position from hyperscale to edge access.
The roadmap as announced
The roadmap named three platforms: “Ares” on 7nm, “Zeus” on 7nm+ and “Poseidon” on 5nm. Ares was due in early 2019. Arm promised performance gains of about 30 percent per generation through 2021.
Arm had designed IP for servers, automotive and networking as separate focus markets over the previous year. Neoverse gathered the infrastructure work under one name.
What happened to the roadmap
Ares arrived on schedule. Architecting IT reports that it launched in February 2019 on a 7nm process and became the Neoverse N1. The Next Platform noted that Arm also released an edge-focused E1 design, code-named “Helios”, shortly afterwards.
The annual cadence then slipped. The Next Platform wrote in 2021 that the cadence had proved tricky and that the market had split into three families: N-series for core data centre, E-series for edge, and V-series for very high performance. In September 2020, Arm updated the roadmap. Zeus became the Neoverse V1, and a second-generation N-series design, N2 (code-named “Perseus”), was introduced alongside it. HPCwire reported that Arm claimed 50 percent more single-threaded performance for V1 and 40 percent more for N2, both measured against N1. ServeTheHome said Arm had achieved a 60 percent uplift with N1, against the 30 percent promised.
Fierce Electronics reported that Chris Bergey, then general manager of Arm’s infrastructure line, described V1 as supporting Scalable Vector Extensions, which matter for machine learning and HPC work. He positioned N2 for power-constrained edge devices, SmartNICs and enterprise networking. The Next Platform noted in 2021 that the design of N2 was finished and licensable.
Chips built on Neoverse
ServeTheHome listed Ampere Altra and AWS Graviton2 as chips based on N1. HPCwire reported that SiPearl chose the Zeus core for the first-generation server processor behind Europe’s exascale plans. Architecting IT said AWS Graviton3 used Neoverse V1, while Graviton4 uses Neoverse V2, which Arm announced in September 2022.
Poseidon, the 5nm generation named in 2018, later became associated with the Neoverse V3 name. Architecting IT said the third generation of the V-series was announced in February 2024.