Arm releases new Neoverse V3 and N3 designs for cloud computing, HPC and AI acceleration

Arm has announced the introduction of two new designs within its Arm Neoverse ecosystem, aimed at cloud-to-edge infrastructure: the Neoverse V3 and Neoverse N3, marking the third generation of products for both platforms. These designs provide reference platforms for the development of efficient Arm chips. Major cloud service providers such as AWS with its Graviton 4 and Trainium 2, Microsoft with Cobalt 100 and Maia 100, and even NVIDIA with its Grace CPU and Bluefield DPU, have already implemented custom Arm server CPU and accelerator designs based on the Neoverse platform within their respective data centers.

Arm states that the Neoverse Compute Subsystem (CSS) enables hyperscale manufacturers to optimize Arm processor designs for their specific workloads, focusing on efficiency over sheer performance. The newly released Neoverse V3 and Neoverse N3 are designed to enhance efficiency and artificial intelligence (AI) inference capabilities. The Neoverse N3 is capable of delivering up to 32 high-efficiency cores per chip, featuring improved branch prediction and larger caches, which boosts AI performance by 196%. The Neoverse V3 can scale up to 64 cores, offering a 50% overall speed improvement over previous generations.

The Neoverse V3 and N3 support advanced features such as DDR5, PCIe 5.0, CXL 3.0, and chiplet architecture, which are becoming standards for future data centers and cloud architectures. The chiplet architecture allows customers to connect accelerators and other chips to Arm cores via the UCIe interface, reducing costs and accelerating time to market.

Arm has outlined a clear roadmap for the Neoverse platform, with forthcoming designs codenamed “Adonis” for the Neoverse V4 and “Dionysus” for the Neoverse N4. These will build upon the foundations laid by the V3 and N3, continuing to advance chiplet architecture optimization to further enhance efficiency and performance.