Arm Releases Details on 2 Neoverse Platforms and Mesh Interconnect for HPC, ML

Print Friendly, PDF & Email

Arm Holdings this morning released information on two new compute platforms and an interconnect for HPC, machine learning and other workloads introduced last September:

  • The Arm Neoverse V1 platform is a new computing tier for Arm and the first Arm-designed core to support Scalable Vector Extension (SVE), delivering 50 percent more performance for HPC and ML workloads, according to the company.
  • The Arm Neoverse N2 platform is the first based on Armv9, delivering 40 percent more performance for diverse workloads, the company said.
  • Arm Neoverse CMN-700 is a mesh interconnect designed to optimize the performance and performance/watt benefits of the two new platforms.

The company said Neoverse V1delivers a 50percent¹ uplift, 1.8x improvement for a range of vector workloads and 4x improvement for machine learning workloads over N1, and said Neoverse V1 is the first in a new performance-first computing tier for Arm.

“Neoverse V1 gives our silicon partners the flexibility to build compute for applications more  reliant on CPU performance and bandwidth while providing SoC design flexibility,” Chris Bergey, SVP/GM, Infrastructure Line of Business, Arm, said in blog. “With the performance-first mindset, the design philosophy behind Neoverse V1 was to build the widest microarchitecture Arm has ever produced to accommodate more instructions in flight in support of markets like high performance and exascale computing. The wide and deep architecture with the addition of scalable vector extensions (SVE), gives Neoverse V1 the lead in per-core performance, code  longevity with SVE, and provides SoC designers implementation flexibility. You can see the benefits of some of these design elements in SiPearl and ETRI’s HPC SoCs and we think this is the direction HPC compute is heading.”

Meanwhile, the Neoverse N2 platform is an infrastructure core, the first based on the Armv9 architecture with emphasis on security, power efficiency and performance.

“Delivering 40 percent¹ higher single-threaded performance compared to N1, Neoverse N2 still retains the same level of power and area efficiency as Neoverse N1,” Bergey said. “The scalability of Neoverse N2 extends from high throughput computing, such as in hyperscale cloud where we see 1.3x improvement on NGINX over N1 and down to power and space constrained edge and 5G use-cases which deliver 1.2x faster DPDK packet processing over N1.”

Bergey said the Neoverse N2 platform delivers better performance per-thread “and industry-leading performance per-watt driving a reduced TCO for users. Neoverse N2 is the first platform to feature SVE2 which drives a significant uplift in cloud-to-edge performance efficiency. For a broader set of use cases like machine learning, digital signal processing, multimedia and 5G, SVE2 not only brings performance but also the ease of programming and portability benefit of SVE.

Arm also unveiled the CMN-700 for building Neoverse V1 and Neoverse N2-based SoCs. Bergey said it provides a step function increase in performance on every vector over the CMN-600, from core counts and cache sizes to the number and types of memory and IO devices that can be attached.

“CMN-700 enables next-generation use cases for multi-chip, memory expansions and accelerators,” Bergey said. “Through our continued investment in CCIX and CXL, we’re providing more customization options and enabling partner solutions with fast fabric and high core count scalability. This will open doors to expand beyond traditional silicon limits and allow for more flexibility for tightly coupled heterogeneous compute.”