This is the first SVE (scalable vector extension) kernel designed by arm, which can greatly improve the performance of high performance computing and machine learning.
Arm n series is a platform focusing on balance and energy efficiency, while arm V series pursues extreme performance, with larger buffer, cache, window and queue design, and larger bandwidth, area and power consumption.
Based on the concept of performance first, V1 platform has created the widest micro architecture of arm so far, which can accommodate more running instructions, support high-performance and megabyte computing, plus SVE function, V1 has obvious advantages in single core performance, and can comprehensively crush N2 platform.
While strengthening the single core performance, V1 platform also supports powerful multi-core expansion, up to 96 cores and 96 threads. It can use 7Nm and 5nm in technology, and also supports ddr5 memory, PCI 5.0 bus, hbm2e high bandwidth memory, ccix 1.1 interconnection standard, etc.
Arm V1 platform is still based on armv8.4-a CPU instruction set, but the overall architecture and module composition are very similar to those based on armv9 instruction set, and there are many similarities. At the same time, there are all-round micro architecture improvements. For benchmark testing and server / high performance computing workload enhancement, of course, it will not drive high power consumption regardless of the cost, as well as in power supply, power consumption The management of fever was strengthened.
Compared with N1, as the first platform of arm's new computing series that emphasizes performance first, V1 brings 50% performance improvement, 1.8 times vector workload optimization and 4 times machine learning workload optimization. It can flexibly build powerful computing power for applications that highly depend on CPU performance and bandwidth, and SOC design is also very flexible.
In terms of ecology, sipearl, a French chip company, and ETRI are building high-performance SoC Based on V1 platform.
India's Ministry of electronic information technology (meity) has adopted V1 platform at night to build a national megabyte high performance computing project.
In addition, arm released arm neoverse cmn-700 synchronously today, which is the most advanced mesh grid interconnection technology in the industry, and can give full play to the performance and energy efficiency advantages of V1 and N2 platforms.
Based on the previous generation of cmn-600, cmn-700 further improves the performance of each vector, including the number of cores, cache size, additional memory and the number and type of IO devices.