Arm CEO Rene Haas delivered a keynote speech on June 3 at the Grand Halai Hotel in Nangang, Taipei, emphasizing the importance of fostering a comprehensive ecosystem to enhance AI acceleration through collaboration with hardware and software partners.
At the event, Haas unveiled Arm KleidiAI, a groundbreaking compute library designed to optimize AI performance for developers. He then turned it over to Arm Senior Vice President and General Manager Chris Bergey to introduce brand-new Arm Compute Subsystems (CSS) for Client, a terminal product tailored for generative AI applications. This innovation, utilizing TSMC’s 3nm technology, offers optimized CPU and GPU compositions in physical form, diverging from traditional IP licensing models to better cater to the AI flagship terminal market.
KleidiAI aims to streamline developer access to Arm CPU’s powerful AI capabilities. It offers a wide range of software programs tailored for diverse AI inference workloads, from cloud to edge computing. Leveraging over two decades of Arm architectural innovation, it allows developers to harness enhanced AI performance and capabilities, from the introduction of Advanced SIMD Extensions in Armv7 architecture to the latest Armv9 architecture, optimized for advanced generative AI tasks.
By focusing on accelerating AI capabilities on the CPU, where most AI tasks initially run, KleidiAI facilitates a seamless development process for developers targeting AI workloads. This approach ensures optimal performance and efficiency, especially as AI-based workloads continue to proliferate, necessitating efficient processing.
Haas reflected on the evolution of AI applications over time. While voice and fingerprint recognition have been prevalent for a while, he said what’s truly fascinating is the pace of adoption among users and across PCs, mobile, applications, and operating systems. “The web took a decade to develop, Twitter took five years, Instagram took two-and-a-half years, TikTok took nine months, and even ChatGPT took only two months to gain traction,” he said.
AI is not just about training a model and waiting for others to use it; it is about continuously evolving and expanding its capabilities. Haas emphasized that AI had yet to reach its full potential, and there is still much room for growth to leverage its full capabilities.
Haas also underscored growing support for the Windows on Arm (WoA) ecosystem by industry giants like Samsung, Apple, and Microsoft, indicating a shift toward Arm-based applications including Chrome, Office, and Galaxy. He positioned these developments as crucial steps toward fortifying the Arm ecosystem and advancing WoA AI PC initiatives.
Arm’s reach extends beyond mobile devices, with 70% of global products powered by Arm-based semiconductor chips. Highlighting Arm’s pivotal role in establishing the world’s AI infrastructure, Haas touted Arm’s complete AI compute platform, which merges accelerated computing and AI technologies to revolutionize visualization of AI and big data.
