Beyond the Newsroom: 13 Breakthroughs at Arm in June 2024

The Arm Editorial Team is constantly on the lookout for the most intriguing and cutting-edge technology trends and stories. Going beyond the Arm Newsroom, we’ve listed 13 groundbreaking technology innovations at Arm in June 2024, such as amplifying the performance and integration of AI operations, transforming image processing, and revolutionizing silicon debugging.

Strengthening AI performance on Arm CPUs with KleidiAI

Unveiled at the end of May 2024, KleidiAI is a collection of highly optimized AI kernels for generative AI workloads. Principal Software Engineer Gian Marco Iodice explains that KleidiAI is packed with features designed to optimize AI operations, with a primary goal to turbocharge AI performance across Arm CPUs.

But KleidiAI is not just about performance, but integration benefits too. With its unique “three-file micro-kernel dependency”, it has now never been easier for AI developers to adopt KleidiAI into their AI frameworks. This focus on ease of integration ensures a smooth and seamless adoption process.

The impact of Arm KleidiCV on image processing

Arm KleidiCV, another component in the new Arm Kleidi Libraries, is set to revolutionize software-based image processing, offering greater speed and flexibility. Micheal Platings, Staff Software Engineer, highlights how KleidiCV uses Arm C Language Extensions and SIMD instructions to supercharge image processing operations and deliver a remarkable 75 percent performance boost in benchmarks on consumer devices, such as the Samsung Galaxy S22.

Boosting AI and Optimizing Cloud Workloads with Arm Neoverse

Arm Neoverse is a groundbreaking technology in the cloud data center landscape, delivering high-performance, power-efficient processors that can be optimized for specific workloads. Senior Product Manager Ashok Bhat offers additional details on how this technology can significantly improve AI capabilities, with advanced SIMD technology and new instructions, like SMMLA and FMMLA, resulting in an up to 3x performance boost on AWS Graviton3 compared to its predecessor.

Additionally, Arm is offering comprehensive developer support in collaboration with Hugging Face. This includes easy-to-use pipelines for inference and resources for performance tuning, making AI development more accessible and cost-effective.

Elaborating further, Software Product Manager Willen Yang and Tianyu Li, Senior Algorithm Engineer at Alibaba Group, discuss the use of Arm Neoverse N2-based Alibaba Yitian710 CPUs, for running large language models (LLMs) efficiently. Recent experiments show that optimizations have led to a 2.7x increase in prompt processing and a 1.9x throughput increase in token generation. The Yitian710 CPUs also offer a performance-effective solution for LLM interfacing, providing up to 3 times higher tokens compared to other server CPUs.

Meanwhile, Arm’s Streamline CLI Tools offer command-line profiling and analysis on Arm Neoverse-based servers, offering clear performance data to optimize application software. Technical Product Director and Distinguished Engineer Peter Harris explains how to optimize cloud workloads with Arm’s Streamline CLI Tools, which offer intuitive profiling and analysis that help improve software performance.

Revolutionizing silicon debugging with Scandump

Scandump is a cutting-edge technique that leverages Design for Testability scan chains to extract internal states from silicon. This allows it to offer crucial insights for debugging, especially when CPU deadlock or hardware becomes unresponsive.

Vincent Yang, Principal Application Engineer, explores how Scandump can significantly expedite the identification and resolution of complex issues, surpassing conventional software debugging tools. Moreover, Scandump offers a comprehensive view of system failures, making it an essential tool in the arsenal for comprehensive silicon issue resolution.

Arm and Samsung to pioneer next-gen communications

There has been an exponential growth in demand for data, primarily driven by AI and next-gen communication technologies, such as 5G and, in the future, 6G. This has resulted in the need for energy-efficient and scalable computing infrastructure.

To address these challenges, Arm and Samsung have entered a strategic partnership to develop a unified codebase optimized for multiple SIMD architectures, as explained by Arm’s Director Wireless vRAN Solutions Marketing Mo Jabbari. The alliance will focus on streamlining development and accelerating the deployment of next-gen communication networks and AI.

Transforming automotive E/E architectures with Arm

The automotive industry is moving from distributed Electronic Control Units (ECUs) to a central E/E architecture, simplifying software updates and reducing the complexity of wiring and hardware. Adam Lackorzynski, Security & System Architect at Kernkonzept, talks about how Arm’s Cortex-A and Cortex-R processors enable virtualization, allowing multiple functions to run on a single system while ensuring safety and real-time performance.

Efficient Matrix Operations with Arm’s SME

Following his part 1 blog on Arm’s Scalable Matrix Extension (SME), Xiu delves deeper into the capabilities of Arm’s SME in this second part blog. He illustrates its use in calculating the outer product of vectors and accumulating the results. Xiu also sheds light on the support for diverse data types and provides practical examples for each, as well as highlighting the prediction feature, which is a key tool for managing non-multiple matrix dimensions.

Enhancing SIMD Portability and Performance with SIMDe

Graduate Information Developer Khalid Saadi summarizes the findings of an academic study Open and Optimize VCC Implementations on Arm Architectures and discusses the use of SIMDe to port x86 SSE and AVX SIMD intrinsics to Arm Neon. While the current automatic translation of SIMD intrinsics from x86 to Arm Neon is effective, future optimizations could further enhance the performance of software like VVenC and VVdeC on Arm-based systems.

Maximizing Android Game Performance with Arm Tools

Featuring Technical Product Director and Distinguished Engineer Pete Harris, this video explores how to advance game development by leveraging Arm’s advanced performance optimization tools. The webinar offers detailed insights into maximizing Android games’ performance through expert monitoring, reporting, and optimization techniques.

Future of Computing with Windows on Arm

David Whaley, Director of Strategic Partnerships at Arm, and Jeremy Sinclair, Microsoft MVP and newest Arm Ambassador of the Arm Developer Program, take a leap into the world of seamless PC and laptop computing and reveal the transformative power of the Arm architecture. Both encourage developers to tap into the potential of Windows on Arm and be part of the ongoing revolution.

Arm at Gitex Africa 2024

Gitex Africa is the continent’s biggest tech expo and known as the pulse of Africa’s digital transformation, setting the scene for AI’s pivotal role in the future. Through this YouTube video below, watch how Arm enables AI and offers significant opportunities for African-based developers to build innovative AI computing solutions on Arm.

Simplifying application development on Arm with GitHub Actions

In the latest Innovation Coffee episode, Larissa Fortuna from GitHub and Diego Russo, Principal Software Engineer at Arm, discuss the introduction of the new Arm-hosted Runners for GitHub actions. The video explores the growing demand for Arm-based solutions in cloud infrastructure and how the partnership promises enhanced sustainability, cost-effectiveness, and support for the open-source community.

Arm Neoverse: Partner Stories

Arm Neoverse is transforming cloud computing with our partners, as highlighted in the latest Arm YouTube series. This not only has Google Cloud and Microsoft discussing their latest Arm Neoverse-based silicon solutions for the cloud, but also features companies including SAP, Modular AI and Rescale who are now adopting Arm Neoverse based-solutions for their own products.