TLDR: Arm Holdings has launched its latest mobile CPU and GPU designs, the Cortex-X925 and Immortalis-G925, respectively, with significant performance boosts, particularly for AI workloads. These chips, manufactured on a three-nanometer process, aim to enhance smartphone and PC performance, and are part of Arm’s broader strategy, backed by SoftBank, to enter the AI chip market with its own designs by 2025.
Arm Holdings plc, a dominant force in mobile chip architecture, has officially unveiled its new flagship mobile CPU, the Cortex-X925, and a corresponding graphics processing unit (GPU), the Immortalis-G925. These next-generation designs are engineered to deliver substantial performance improvements, with a particular emphasis on accelerating artificial intelligence (AI) workloads in smartphones and personal computers.
The Cortex-X925 CPU core marks the largest speed increase in the Cortex product line’s history, boasting up to 36% faster performance for single-threaded applications compared to its predecessor. More notably, its AI performance is reported to be 25% higher. This enhancement is further amplified by a new CPU cluster design that offers a 46% better AI performance compared to previous generation Cortex-X4 clusters, promising more responsive performance and sustained throughput. Saurabh Pradhan, director of product management for Arm’s client CPU business, highlighted these improvements, also noting a 30% improvement in key user experience indicators that combine performance and power.
Accompanying the CPU is the Immortalis-G925 GPU, Arm’s new flagship for smartphones. This graphics card is capable of running some mobile games 37% faster than its predecessor and delivers a 34% performance improvement for applications utilizing AI models for data processing. A key innovation in the Immortalis-G925 is a feature called ‘fragment prepass,’ which reduces the need for data sorting, a computationally intensive operation. Anand Patel, senior director of GPU product management for Arm’s client GPUs, stated that this feature has led to a 43% reduction in render thread cycles, improving performance, power efficiency, and reducing CPU load.
Both the Cortex-X925 and Immortalis-G925 are designed to be manufactured using a three-nanometer process, representing a leap in semiconductor technology. In addition to these high-end offerings, Arm is also rolling out lower-end core designs: the midrange Cortex-A725, which is 25% more power-efficient, and the entry-level Cortex-A520, offering a 15% efficiency improvement.
Beyond these core designs, Arm is embarking on a significant strategic shift to develop and launch its own complete AI chips by 2025. This initiative is heavily backed by its parent company, SoftBank Group, which plans to invest up to JPY 10 trillion ($64 billion) to become a leading player in the AI space. Arm aims to have a prototype ready by Spring 2025 and commence mass production by Autumn of the same year. This move positions Arm in direct competition with industry heavyweights like Nvidia.
To bolster this ambitious plan, Arm Holdings has hired Rami Sinno, formerly Amazon.com’s artificial intelligence chip director, who led the development of Amazon’s AI chips Trainium and Inferentia. Arm’s CEO, Rene Haas, has publicly stated the company’s exploration beyond its traditional design business, including the development of smaller, function-specific, and modular chips. SoftBank is actively negotiating with contract chip manufacturers, including Taiwan Semiconductor Manufacturing Company (TSMC), Intel, and Samsung Foundry, to secure manufacturing capacity for these new AI chips.
Also Read:
- Gartner Predicts Generative AI Smartphones to Capture 20% of Total AI End-User Spending by 2025
- NVIDIA Unveils Rubin CPX GPU, Revolutionizing AI Inference for Million-Token Contexts
These upcoming AI chips are expected to transform AI computing, democratize access to AI technology, and promote inclusivity in the tech industry. While some reports suggest these chips might initially be used for SoftBank’s own data centers in the EU, U.S., Asia, and the Middle East, the broader impact is anticipated to be significant for the UK semiconductor industry and the global AI hardware landscape.
