A few months after first introducing the N2 and V1 Neoverse platforms, Arm is highlighting more of the performance and the features the platforms deliver. The chip designer is also touting the growing list of processors using the Neoverse family of products to bring specialized processing to cloud, HPC and elsewhere.
The Neoverse-based product news includes:
- The next generation of Marvell’s Octeon family of networking processors will be based on Neoverse N2, delivering 3x performance over the prior generation. They’ll begin sampling by the end of 2021.
- India’s Ministry of Electronics and Information Technology (MeitY) will license Neoverse V1 for its national exascale HPC project.
- Tencent is investing in both hardware testing and on software enablement in order to adopt Neoverse technology for cloud applications.
- Alibaba Cloud has tested the upcoming Alibaba Cloud ECS Arm instances, showing a 50 percent performance improvement of the DragonWell JDK on Neoverse N1
- Oracle is adopting Ampere Altra CPUs, with up to 160 Neoverse N1 cores, for Oracle Cloud Infrastructure.
- Amazon Web Services’ Arm-based Graviton2 is expanding Arm’s footprint in the cloud.
When Amazon launched Graviton2, “one of the things that people were surprised about wasn’t just that Arm could make a compelling core,” Arm’s Chris Bergey said to ZDNet, “but the fact that AWS could put 64 of those cores and have consistent performance across scale-out workloads.” The consistent performance, he said, was in large part attributable to Arm’s CMN-600 mesh interconnect.
Arm is now introducing the CMN-700, a key element for constructing higher-performance Neoverse V1 and Neoverse N2-based SoCs.
Meanwhile, the Neoverse V1, as previously announced, delivers a 50 percent uplift over Neoverse N1 and a 4x improvement for machine learning workloads. The Neoverse V1 is the widest microarchitecture Arm has ever produced, and it’s focused at HPC and exascale computing.
The Neoverse N2 platform, meanwhile, is the first based on the Armv9 architecture. It delivers 40 percent higher single-threaded performance compared to N1 while retaining the same level of power and area efficiency as N1. For high-throughput computing, such as in hyperscale cloud, it delivers 1.3x improvement on NGINX over N1. For use cases with power and space constraints, such as edge and 5G use cases, it delivers 1.2x faster DPDK packet processing over N1.