Chip designer unveils next generation of compute architecture for cloud, edge, and HPC workloads
Chip designer Arm is rolling out the latest offerings in its Neoverse compute platform, which was first introduced two years ago to bring a more efficient architecture to systems spanning the cloud to the edge.
The Lowdown: The introduction Tuesday of the N2 and V1 platforms comes just more than a week after GPU maker Nvidia announced it’s planning to buy Arm for $40 billion in a deal that could take as long as 18 months to close.
The Details: Neoverse was a key step in Arm’s decade-long effort to bring its low-power processor designs – which can be found in most smartphones, tablets, and other mobile devices – to the data center to give enterprises a viable alternative to Intel’s x86-based processors. The goal was to create an architecture that could run everything from Internet of Things (IoT) devices to hyperscale and high-performance computing (HPC) servers. Arm promised at least a 30% year-over-year improvement in performance through 2021.
The initial platform, N1, delivered a 60% increase over the existing Cortex-A72 CPU, and the N2 (40%) and V1 (50%) also offer significant improvements over the N1. The V1 platform is designed for single-threaded workloads that rely on CPU performance and bandwidth and supports Scalable Vector Extensions (SVE), which is important for such markets as HPC, cloud, and machine learning.
The N2 – the next iteration of the N-series platform – addresses scale-out compute environments across a range of workloads, from the cloud to SmartNICs, enterprise networking, and power-constrained edge devices. The 40% performance improvement comes in the same level of power and area as the N1.
The Impact: Intel for almost two decades has been the dominant player in the data center processor market. During that time, there has been an ongoing demand among enterprise and channel partners for options to the Intel architecture to grow competition and drive down prices. A reinvigorated AMD is back in the data center with its Epyc processors, and vendors like Nvidia and Xilinx are providing GPUs and field-programmable gate arrays (FPGAs) to work with CPUs to speed up compute performance.
Arm – which designs chip architectures that it then licenses to manufacturers like Qualcomm and Ampere – is gaining traction in data centers as companies and cloud providers adopt Arm-based chips or, in the case of Amazon Web Services (AWS) and its Graviton processors, build their own chips using the designs. Arm-based chips also are used to power Fugaku, which sits atop the Top500 list of the world’s fastest supercomputers.
Background: Nvidia is eyeing Arm as a way to bring CPU capabilities into its portfolio and drive its efforts in such areas as artificial intelligence (AI), HPC, and data analytics. Arm’s current owner, SoftBank Group, will use money from the sale to alleviate some of the financial pressures it’s under. SoftBank bought Arm for $32 billion in 2016.
The Buzz: “Neoverse technologies are appearing in new server and SoC designs across the infrastructure world, and software and tools support has flourished,” Chris Bergey, senior vice president and general manager of Arm’s Infrastructure Line of Business, wrote in a blog post. “Developers see not only the performance and efficiency gains that Neoverse can deliver, but the broader design freedom and flexibility that comes with a new way of thinking about deploying infrastructure. Based on that collaboration and hard work, I’m excited about what our ecosystem partners will be able to achieve in the future. The velocity of Neoverse has increased the pace of innovation in the infrastructure. What amazing things can we achieve together in the coming years?”