The CPU Architecture for the Future of AI
The Armv9-A architecture gives AI developers the programming tools and environment necessary to innovate at pace for the rapidly expanding AI market. AI applications process vast datasets, perform complex calculations, and interpret data collected in real-time to benefit both commercial and everyday tasks. As these programs compute more data, the need for both performance and security becomes paramount. Armv9-A is built with both needs in mind. It offers faster compute for high-performance use cases with a suite of security features that help ensure both data and operational integrity. These benefits are delivered on the device or server without compromising power consumption.
Faster Algorithms and Power Efficient Compute, Built with Security in Mind
Armv9-A accelerates the processing of large datasets, which is critical for cutting edge scenarios like genomics and computer vision. All are made possible without the need to recompile software and remain power efficient through improved memory use.
Developers can be assured that data and proprietary programs are built with security in mind, through Arm’s latest confidential compute features. Combined with the latest optimization tools, Armv9-A offers performance improvements without compromising development time or software integrity.
Key Features for Unleashing CPU Performance at Scale
Scalable Vector Extensions optimize the amount of data processed at once, based on available silicon. With the latest iteration, SVE2, algorithms can process more data on more powerful processors, unlocking more high-performance compute scenarios without re-compiling software.
AI applications process a lot of data, making them power hungry. SME uses memory efficiently, improving bandwidth. This increases compute performance without impacting power consumption. Complex compute can take place locally for applications such as audio/visual processing and digital assistants.
BRBE gives a history of recent execution so developers can prioritize optimization tasks, improving the software experience for users. Combining BRBE with other tools enables profiling-based optimization, which recompiles and runs the most optimal code based on real-time usage.
Key Features for Building-in Security from the Ground Up
Realms are separate computational worlds that run a specific application. The use of a realm prevents attacks from software that runs at higher privilege levels. The contents of a realm, or its processes, cannot be accessed and remain encrypted. Realms can interact with an accelerator and maintain their integrity.
Attacks frequently attempt to subvert software control flow. PAC and BTI help ensure control flow integrity by providing defences against return-orientated programming (ROP) and jump-orientated programming (JOP) attacks. Through effective authentication, hackers cannot modify code, or manipulate the control flow of an application.
Spatial and temporal memory safety issues account for 70% of all serious security bugs. MTE helps detect memory safety issues, such as use-after-free or overrun. Both the pointer and data they point to are tagged, so the processor can check permissions at run time.
Inference on Arm CPUs
Our relentless innovation enables the most pervasive CPU architecture for the future of AI, whether in the datacenter or at the edge. We enable the creation of custom silicon to match software strategies, deliver the CPU architecture to create performant software, and provide a comprehensive suite of security features to protect sensitive data, with software libraries that facilitate development cycles. Find out more about the benefits of using Arm CPUs for your AI workloads and why our platform is the most pervasive.
Get Building with Armv9-A Architecture Today
Everything you need to get started with Armv9-A. Includes technical documentation, how-to guides and tutorials.
Latest Armv9-A News and Resources
Webinar:
Blogs:
- Introducing Armv9 Scalable Matrix Extension for AI Innovation on the Arm CPU
- New Armv9 CPUs for Accelerating AI on Mobile and Beyond
- Part 1: Arm Scalable Matrix Extension (SME) Introduction
- Part 2: Arm Scalable Matrix Extension (SME) Instructions
- Part 3: Matrix-matrix multiplication. Neon, SVE, and SME compared