Arm fleshes out details of Neoverse V2 processor core
Details of Arm’s Neoverse V2 processor core, codenamed Demeter, have been released by Arm. The processor is based on the Armv9 architecture which is optimised for data centre applications, including BERT (bi-directional encoder representations from transformers), Google’s machine learning framework for natural language processing. Arm has doubled the icache (instruction cache) of the earlier Neoverse to 2Mbits and increased vector performance with the Arm CMN-700 mesh interconnect for cloud processing workloads.
Neoverse V2 will deliver market-leading integer performance for cloud and HPC (high performance computing) workloads and introduces several Armv9 architectural security enhancements, said the company.
The next generation N-Series product is in development and will be available to partners in 2023. Arm has also applied the core to its E-Series CPUs for data plane processing, 5G RAN, edge networking and accelerators. The Neoverse E2 platform combines the Cortex-A510 CPU with the scalable Neoverse CMN-700 and N2 system backplane. It targets cloud technology, constrained applications with a scalable range of core counts, SystemReady compatibility and PCIe, CXL, IO and interfaces. Arm is already defining the next generation E-Series platforms.
Neoverse platforms are built to address the full spectrum of accelerated infrastructure with specialised processing to reduce workload power consumption and meeting the increased compute requirements of workloads.
Arm announced that Nvidia will use its Neoverse V2 cores for the 72-core Grace CPU which will be officially launched this week. Grace will combine the V2 with the power efficiency of LPDDR5X memory to deliver two times the performance per Watt over servers powered by traditional architectures, said Arm.