It’s been several months since Arm released information on both the “Neoverse N2” and “Neoverse V1” platforms for the first time. And on April 27, local time, Arm highlighted the performance and features of these platforms and adopted the “Neoverse” product family to perform specific operations in the cloud, high performance computing (HPC), and more. We introduced our partner companies and appealed for the expansion of their ecosystem. Below is new news about Neoverse-based products. The new generation of Marvell’s “OCTEON” family of network processors will be based on the Neoverse N2, which will deliver three times the performance gains of the current generation of OCTEON. Sample shipments will begin by the end of 2021. India’s Ministry of Electronics and Information Technology (MeitY) plans to license Neoverse V1 for a state-led exascale HPC project. Tencent is investing in hardware testing and software enhancements to enable Neoverse technology in cloud applications. When we tested the “Alibaba Cloud ECS Arm” instance that Alibaba Cloud plans to offer in the future, the performance of “Alibaba DragonWell JDK” on “Neoverse N1” improved by 50%. Oracle plans to use an “Ampere Altra” CPU with up to 160 Neoverse N1 cores in its “Oracle Cloud Infrastructure.” Arm-based “Graviton 2” on Amazon Web Services (AWS) is rapidly expanding the footprint of “Amazon Elastic Compute Cloud” (Amazon EC2). Arm’s Chris Bergey told ZDNet, “What surprised people (when Amazon launched Graviton 2) wasn’t just the fact that Arm (architecture) allowed them to develop notable cores, but AWS. Is the fact that it has 64 such cores and consistent performance across scale-out workloads. ” This consistent performance is largely due to Arm’s “Arm CoreLink CMN-600 Coherent Mesh Network” mesh-type interconnect. Also on the same day, Arm announced the “Arm CoreLink CMN-700 Coherent Mesh Network” as a core product for building SoCs that achieve higher performance based on Neoverse V1 and Neoverse N2-.
On the other hand, Neoverse V1, as previously announced, offers a 50% performance improvement over the Neoverse N1 and a 4x performance improvement over machine learning (ML) workloads. This product focuses on HPC and exascale computing and is developed based on the design philosophy of performance first. This also extends the breadth and depth of pipeline design, making it the widest architecture the company has ever developed.
Neoverse N2 is also the first platform based on the “Arm v9” architecture. The product delivers 40% performance gain over N1 in single-threaded performance while maintaining the same level of power and area efficiency as the Neoverse N1. In computing that requires high throughput such as hyperscale cloud, it has achieved a performance improvement of 1.3 times compared to NGINX on N1. In use cases where power and installation space are limited, such as edges and 5G, the speed is 1.2 times faster than DPDK packet processing on N1.