Neoverse CSS enables high-performance custom silicon - Infrastructure Solutions...

This blog is co-authored by Mohit Taneja, Sr. Product Manager, Arm Infrastructure Line of Business, and Saptarshi Mondal, Product Director, Arm Infrastructure Line of Business

Enabling High-Performance, Custom Silicon

Arm Neoverse is engineered to deliver performance with power efficiency to the full spectrum of infrastructure use cases, from cloud to edge. For workloads and use cases needing maximum performance, Arm offers the Neoverse V-series. Our latest Neoverse V2 cores are adopted and deployed by the world's leaders in cloud, HPC and AI. Announced at AWS Re:Invent 2023, the AWS Graviton4 CPU provides higher core counts and larger memory configurations than previous Graviton CPU-based EC2 instances. And NVIDIA is shipping the Grace CPU Superchip and Grace-Hopper Superchip to bring leadership performance and power efficiency to HPC and AI/ML workloads. Today we are announcing our next-generation of Neoverse V-series, the Neoverse V3 CPU and Neoverse CSS V3.

Arm Neoverse Compute Subsystem (CSS) products are designed to help Arm partners deliver the latest Neoverse technology to market more quickly, with less risk, and with more customer-defined differentiation than they could do so previously. Based on a CSS configured, validated, and optimized by Arm engineers, Arm partners can use Neoverse CSS as an advanced starting point for implementing their own custom silicon. This lets our partners utilize their engineering talent efficiently, optimizing for maximum performance and TCO for their unique workloads and operating environments. Recently announced by Microsoft Azure, the Cobalt 100 CPU is an example of customer collaboration with Arm to bring specialized silicon to market.

Today we extend the time-to-market and customization benefits of Neoverse CSS to V-series, with the Neoverse CSS V3. Neoverse CSS V3 offers the fastest path from concept to deployment of next generation cloud compute for our global partners, offering significant performance, efficiency, and differentiated features. CSS V3 also offers a +50% performance per socket improvement over CSS N2, which will enable our partners to address a range of new workloads and use models.

Let’s explore Neoverse V3 and CSS V3 and how they can deliver value for cloud infrastructure.

Neoverse CSS V3 Overview

Figure 1. Neoverse CSS V3 overview

Neoverse CSS V3 is a customizable compute subsystem that is configured, verified, validated, and PPA-optimized by Arm. Arm performs the undifferentiated heavy lifting, enabling Arm partners to build specialized silicon at a lower cost, with less risk and a faster time-to-market compared to traditional development methods. At the heart of this CSS is the Neoverse V3 CPU.

The Neoverse V3 CPU, along with system level innovations, allows CSS V3 to offer a compelling +50% performance per socket for SPECint2017 (estimated) over CSS N2. CSS V3 also brings similarly impressive socket level uplifts across a range of infrastructure workloads including machine learning, database and crypto.

Figure 2. Neoverse V3 performance uplift (%) over Neoverse V2 for cloud computing workloads

Neoverse CSS V3 gives our partners the fastest path to production silicon with the features and capabilities that next-generation infrastructure platforms require, while managing risk and enabling an extremely fast time to deployment.

Neoverse CSS V3 also brings to market many new systems level capabilities:

• V-series performance in a CSS product
• Neoverse S3 System IP capabilities
• Arm Confidential Compute Architecture
• AMBA CHI C2C – enabling multi-chiplet and disaggregated chiplets, including heterogeneous acceleration integration
• New Power management capabilities, including per-core DVFS

CSS V3 forms the key building block for specialized computing systems. It reduces design and validation cost for the general-purpose compute subsystem, allowing our partners to focus on their specialization and acceleration, while reducing risk and offering accelerated time to deployment. CSS V3 is offered in various configurations with up to 128 Neoverse V3 cores in a socket, and 2-socket configurations supported. It can also be scaled down to support smaller configurations (ex, 32-cores). CSS V3 is targeted at 3nm class process and comes with supporting floorplans and implementation flows to minimize physical design risk.

Neoverse V3 CPU Overview

The Neoverse V3 CPU represents a significant architectural improvement over Neoverse V2.

Figure 3. Arm Neoverse V3 CPU

Hardened platform: Security and reliability of the cloud infrastructure is essential in the modern world.

• Security of data during use, transit, and rest in muti-tenant & multi-use systems against external/internal threat actors is of utmost importance. Neoverse V3 delivers confidential computing at scale, to tackle issues around data security head on.
• Decreasing mean time between failure (MTBF) and improving overall reliability of the infrastructure goes hand in hand. Neoverse V3, by providing cutting edge RAS and Telemetry technology such as branch record buffer Extension (BRBE) helps improve MTBF.
• Software running on cloud infrastructure runs the world. Hence, it is of utmost importance that cloud infrastructure providers choose underlying technologies that are day-1 ready for the large ecosystem of critical software.  Neoverse V3 meets the day-1 ready challenge and delivers.

Through these technological innovations Neoverse V3 brings the hardened platform that our partner needs, and their customer demands.

TCO-optimized compute: Cloud native workloads require TCO optimized customization. The inherent flexibility of the Neoverse V3 CPU delivers significant improvements on real world workload performance, in both standard and purpose build configurations. These best-in-class server CPUs can be deployed to fit not only the most demanding AI and HPC applications but also more traditional enterprise grade infrastructure software. Case in point, Neoverse V3 delivers performance improvements of +96% on Machine learning, +16% on RDBMS, +9% on crypto and +12% on integer workloads over Neoverse V2.

Summary

Neoverse V3 and CSS V3 are designed to meet the most demanding hyperscale cloud infrastructure needs of the future. It delivers TCO optimized, hardened and time-to-market accelerated infrastructure building blocks for the next generation of AI, HPC and cloud native workloads. Arm is making significant investments to help our partners to bring Neoverse V3 based products to market cheaper, faster, and ready to go on day one. Please reach out to the Arm team and let’s build the future together on Arm.

Learn About CSS V3