ARM Cortex-A

2021: Integration with Nvidia Aerial A100 AI-on-5G

On June 28, 2021, NVIDIA announced the expansion of Arm-based CPU support for the NVIDIA Aerial A100 AI-on-5G platform, providing a wider choice of solutions in the 5G ecosystem.

According to the company, the NVIDIA Aerial A100 AI-on-5G computing platform uses the NVIDIA Aerial software development kit and will include 16 Arm Cortex-A78 processors in the NVIDIA BlueField-3 A100. This standalone converged card will provide access to enterprise AI peripheral applications via 5G vRAN with Watt optimized performance and shorter deployment times. More details here.

2018: Cortex-A76

On June 1, 2018, ARM announced the second generation of mobile processors based on the Cortex-A76 core.

According to the results of synthetic tests published by ARM, the Cortex-A76 processor compared to the A75 is able to demonstrate 35% higher performance when solving standard tasks and a quadruple increase when performing machine learning and artificial intelligence algorithms .

Cortex-A76 Performance vs. Predecessors

In addition, it is claimed that A76 exceeds its predecessor by 40% in terms of energy efficiency. This means that in the future, models based on the Cortex-A76 with a battery life of more than 20 hours may appear on the laptop market, according to ARM.

The Cortex-A76 chip is made according to a 7 nm process, operates at a frequency of 3 GHz and is based on DynamIQ technology, first used in the development of the A75 core, released in 2017. DynamIQ technology allowed developers of single-chip systems to create processors from several clusters containing various processor cores.

According to ARM estimates, processors manufactured by the company show an annual increase in performance of about 20%. The nutritional requirements, however, remain almost unchanged.

ARM also reported that Microsoft began porting Windows 10 to the ARM platform.^[1]

2017: Cortex-A75 и Cortex-A55

ARM, the developer of mobile processors, introduced in May 2017 the first two processor cores Cortex-A75 and Cortex-A55 based on the new multi-core processor microarchitecture DynamIQ, which in the coming years will become the basis of all new processors of the Cortex-A family.

Microarchitecture DynamIQ, first introduced by the company in March 2017, replaced the previous big technology. LITTLE, first introduced by ARM in 2011 and successfully established itself in more than five years. ARM notes that DynamIQ is a further evolutionary development of big ideas and technologies. LITTLE.

According to ARM representatives, the new Cortex-A75 and Cortex-A55 processor cores with DynamIQ technology will provide optimization for 50 times the performance of artificial intelligence computing over the next three to five years, and 10 times the performance using integrated hardware accelerators.

ARM emphasizes that unlike the previous positioning of its computing cores for certain narrow markets - for example, for mobile devices, with the release of microarchitecture DynamIQ the company begins the transition to a flexible universal platform with huge scaling capabilities that supports intelligent solutions for almost any application, from clouds and networks to gadgets.

Processor technology DynamIQ inherited from the architecture of the previous generation ARM big. LITTLE time-tested computing power management - when cost-effective processor cores are combined in a single chip with multiple high-performance cores. This allows you to design a mobile processor that can significantly increase performance if necessary, and save the battery of the mobile device the rest of the time.

Developing the idea of ​ ​ "the right processor for the right task," the DynamIQ architecture supports up to 8 processor cores per computing cluster, while there can be almost infinitely many clusters in the final chip.

Each computing cluster, in turn, provides a certain level of performance. Unlike big traditions. LITTLE, which used only pairwise acceleration of powerful and economical cores (2 + 2, 2 + 4, 4 + 4, etc.), the architecture DynamIQ can work with any combination of economical and powerful cores - from 1 + 3 or 1 + 7 to any others.

Features of ARM Cortex-A75

The powerful ARM Cortex-A75 processor core provides significant performance and energy efficiency gains over its predecessors Cortex-A72 and Cortex-A73. The chip has about 20% improved performance when working with integers, significant improvements when working with floating point numbers and tasks with a large load on the memory subsystem.

The Cortex-A75 processor is characterized by peak performance at single-stream loads due to the presence of a symmetrical three-way superscalar pipeline of varying length with a completely arbitrary (out-of-order) instruction sampling.

Cortex-A75 cores have a distributed L3 cache cluster, support for asynchronous frequencies and virtually independent power voltages for various cores within a multi-core processor or cluster. Cortex-A75 cores are also equipped with a separate L2 cache for each core with a halved latency compared to their predecessors.

In combination with the distributed DynamIQ module (DynamIQ Shared Unit, DSU), the Cortex-A75 processor allows to provide necessary performance level for a wide range of a system and the markets - from smartphones and smart houses to servers and automobile electronics.

With basic execution at the level of the ARMv8-A microarchitecture, the Cortex-A75 compute cores are fully backward compatible with the entire ecosystem of operating systems, tools and applications developed for this platform, while providing new capabilities for developers of systems with artificial intelligence.

Features of ARM Cortex-A55

According to the company, the economical processor core ARM Cortex-A55 provides the best in the series combination of energy efficiency and performance in its class. The chip is also based on DynamIQ technologies and ARMv8-A architecture with support for the latest machine learning enhancements.

The Cortex-A55 core provides 18% better performance when working with integers, 20% with floating point numbers, 40% when calculating NEON SIMD, 15% when working with JavaScript and 200% when calculating with a large memory load with a reduced 15% energy consumption compared to its predecessor Cortex-A53.

The Cortex-A55 core also has an integrated configurable L2 cache for each core with more than 50% reduced latency when accessing the memory subsystem. Cortex-A55 cores support L3 cache allocated to a cluster of 8 cores.

Notes