Hardware

ARM strengthens its presence in the server and data center world with Neoverse V3, N3 and E3

ARM strengthens its presence in the server and data center world with Neoverse V3, N3 and E3

I SoC ARM Neoverse they are chips designed to support workloads and computing high-performance: think edge servers, data centers and cloud infrastructures. This is a range of products with which ARM seeks to satisfy the growing demand for increasingly powerful and, at the same time, energy efficient solutions for cloud computing applications, artificial intelligence, machine learning and the Internet of Things (IoT).

The Serie Neoverse V2 has already proven its reliability, with processors adopted by world leaders such as AWS and NVidia. Now, with Neoverse V3 and CSS V3, ARM intends to continue on the same path by expanding the opportunities both in terms of performance and energy saving.

On February 21, 2024, ARM presented the third generation of its Neoverse chips. The British company has indeed announced the models Neoverse V3, Neoverse N3 e Neoverse E3, along with the CSS V3 and CSS N3 computing subsystems. The SoCs are indicated in order of increasing energy efficiency and decreasing power: ranging from the most powerful Neoverse V3 up to the super-efficient Neoverse E3.

Cos’è ARM Compute Subsystem (CSS)

CSS is a set of compute blocks and components designed to provide a complete, integrated compute solution for edge and IoT devices. It’s part of the platform ARM Cortex-A and includes a set of components that work together to provide a smooth and responsive user experience.

I key components of CSS include the actual processor, the Mali graphics section, the memory, the interconnection interface, the advanced security features (hardware encryption, key management and memory protection, to protect data and ensure system integrity) and energy management.

Compared to the previous CSS N2, according to ARM engineers CSS V3 would offer an increase in performance performance per socket by 50%, paving the way for new workloads and usage patterns.

ARM partners can also configure up to 128 core per socket, with support for DDR5, LPDDR5, HBM3, PCIe Gen5 and CXL 3.0. Furthermore, the design involves the use of production processes a 3 nm with a thermal power kept under 40W in the “most advanced” configurations.

ARM Neoverse V3

Neoverse V3 architecture, a notable leap forward on the CPU side

The Neoverse V3 CPU represents a notable architectural improvement over the previous V2. The inherent flexibility of the V3 CPU leads to significant improvements performance relative a workload real, both in standard and customized configurations. V3 delivers performance improvements of 96% in machine learning, 16% in relational database management, 9% in encryption, and 12% in integer-based workloads over the previous V2 version.

The number of cores that can be used within the platform CSS N3 varies from 8 to 32 cores per socket: the 40W of TDP (Thermal Design Power) are also obtained in the “plant” based on 32 cores.

Compatibility with CXL 3.0 it is one of the indisputable strong points. CXL (Compute Express Link) is a high interface bandwidth and low-latency designed to connect CPUs, GPUs, FPGAs, and other compute accelerators to memory and other peripheral devices. CXL 3.0 is the third generation of the interface that increases bandwidth to 6.4 GT/s, double that of CXL 2.0, includes advanced security features, ensures greater power efficiency, is backwards compatible and supports CXL persistent memory.

Performance ARM Neoverse V3 N3

The XGBoost algorithm

XGBoostshort for eXtreme Gradient Boostingis a powerful algorithm machine learning which belongs to the category of boosting algorithms. It is especially popular and widely used for regression and classification problems, and is known for its efficiency, accuracy, and flexibility.

ARM has intelligently decided to adopt it in the new Neoverse thus providing integrated hardware support. Both Neoverse V3 and Neoverse N3 achieved significant improvements when analyzing data with AI. Compared to Neoverse V2, Neoverse V3 recorded a performance improvement of 84%, while Neoverse N3 achieved an impressive +196% using Neoverse N2 as the benchmark.

Dermot O’Driscoll, vice president responsible for ARM’s infrastructure aspects, clarified that by analyzing the “typical” workloads of the partners, the company is able to identify the microarchitectural changes more effective which will positively influence performance.

An increasingly strong presence, that of ARM, in the world of servers and data centers

The experience gained on the “mobile” side has led ARM engineers for several years now to develop products suitable for satisfying the needs of business customers in the server e data center.

It is no coincidence that the announcement arrives at the same time as the one made by the number one of IntelPat Gelsinger, a company very committed to becoming the “favorite foundry” of the main ones big player of the industry. The presentation of the Intel Foundry initiative was also yesterday, while in the background there is still the multigenerational agreement for the creation of chips signed by ARM with Intel.

Il the Neoverse system as a whole is focused on presenting customers with high-performance customized configurations: a unique recipe that promises simplification and acceleration in the development of “custom silicon” for cloud and artificial intelligence.

The images in the article are taken from the official ARM website.

Leave a Reply

Your email address will not be published. Required fields are marked *