Arm will release two new Neoverse compute subsystems (CSS) built on a brand new third-generation of its Neoverse IP.
By bringing compute subsystems to the product line through its new N3 and V3 products, Arm says it will be able to unlock technologies such as chiplets.
First launched in October 2018, the Neoverse cloud-to-Edge microarchitecture consists of 64-bit ARM processor cores with the N-Series intended for data center use, the E-Series for Edge computing, and the N-Series for high-performance computing.
In 2023, Arm introduced Neoverse CSS to simplify and accelerate the adoption of Arm Neoverse-based technology into new compute solutions by enabling its partners to build specialized silicon more affordably and quickly than previous discrete IP solutions.
Unlike rivals such as Intel and AMD, Arm does not sell its own chips – it licenses out core designs, which other companies build upon, adding elements like memory, storage, and PCIe controllers.
The new Neoverse CSS N3 is focused on power efficiency and offers a 20 percent per core uplift in performance-per-watt over CSS N2, with the first instantiation of CSS N3 offering 32 cores in as low as 40-watt thermal design power (TDP).
CSS N3 is built on Arm’s newest Neoverse N3 IP platform, bringing Arm v9.2 features to the N-Series. It also includes a two-megabyte per core private L2 and supports the latest in PCIE and CXL IO standards, as well as UCIE chiplet standards.
Arm is also extending the CSS program to its V-Series cores through the introduction of Neoverse CSS V3. The company says V3 delivers an over 50 percent socket performance boost when compared to CSS N2, and can scale up to 128 cores per socket and supports the latest high-speed memory and IO standards.
The V3 offers hardware support for the Arm Confidential Computing architecture and both the V3 and N3 cores deliver industry-leading private L3 cache sizes to boost performance.
Speaking ahead of the launch, Dermot O’Driscoll, VP of product solutions, infrastructure line of business, at Arm, said that when it comes to performance, the most significant improvement is the gain for AI data analytics workloads, with the V3 having an 84 percent performance uplift over the V2, and the N3 having a 196 percent performance uplift when measured on simulation.
Arm extends its ecosystem partnerships
Arm is also working with its ecosystem partners to solve the physical layer compatibility of chiplets, a challenge that occurs when designs from different teams are paired together. O’Driscoll said that through its recently launched chiplet system architecture, Arm is hoping to establish an interoperable ecosystem, working with 15 partners to “unlock the potential of chiplet technology,” with CSS making it easier for customers to co-design and build a compute chiplet to sit alongside their AI accelerators.
Meanwhile, the Arm Total Design ecosystem has grown to 20 members since it was launched four months ago to include EDA and complimentary IP providers in addition to silicon design partners. Broadcom, Faraday, Intel Foundry Services, Samsung, and TSMC were all among the partners named by Arm.
“The platforms we have introduced today are the best generation of Arm Neoverse, yet they'll be the foundation of the next-generation products and services partners build as more AI workloads are adopted and permeate across the entire continuum of compute,” said Mohamed Awad, SVP and GM, Infrastructure Line of Business, Arm.
“From the largest data centers to networks, servers, and all aspects of the computing infrastructure, Neoverse V3 and N3 provide both the computing performance and efficiency needed to underpin this transformation, and our compute subsystems supported by the Arm Total Design ecosystem allow for faster time to market and more investment in innovation than ever before.”