Arm has launched a new range of Armv9 CPUs optimized for 3nm SoC and updated GPUs based on 5th Gen GPU architecture as successors to last year's Arm Total Compute Solutions 2023 (TCS23).
The latest additions to the Armv9 CPU lineup comprise the Arm Cortex-X925 CPU for ultimate performance, the Arm Cortex-A725 CPU for superior sustained performance, and the new Arm Cortex-A520 designed for optimal energy efficiency in low-intensity workloads, all tailored for 3nm processing.
Additionally, ARM has enhanced the DynamIQ Shared Unit (DSU-120) to deliver reduced power consumption and footprint across Armv9.2 CPU cluster configurations. These advancements are integrated into the new Arm Compute Subsystems (CSS) for Client, representing ARM's most powerful compute platform for Android to date.
Arm Cortex-X925 -
The ARM Cortex-X925 succeeds the Arm Cortex-X4 CPU, boasting a 36% increase in single-threaded (peak) performance and a 46% enhancement in AI performance compared to its predecessor.
An optimized 3nm implementation of the Cortex-X925, coupled with a premium subsystem and packaging, promises a 30% boost in performance scores on next-generation consumer devices. Enhancements to the microarchitecture, including a potential 3MB private L2 cache, offer improved configurability for CPU cluster implementations across various consumer devices.
The tape-out ready physical implementation of the Cortex-X925 for 3nm enables ARM's partners to fully leverage the power, performance, and area (PPA) benefits of the 3nm process, thereby expediting silicon development and deployment timelines with a high-volume production-ready solution.
Arm Cortex-A725 -
The ARM Cortex-A725 succeeds the Cortex-A720, delivering a 35% improvement in performance efficiency and a 25% enhancement in power efficiency compared to its predecessor. It offers optimized implementation on 3nm SoC through ARM's advanced physical implementation.
Arm Cortex-A520 (2024) -
The new Cortex-A520 shares the name with last year's energy-efficient CPU for low-intensity workloads but utilizes advanced 3nm physical implementation. This results in a 15% efficiency boost compared to the Cortex-A520 in TCS23.
DSU-120 -
the new DSU-120 Core Cluster for 2024 incorporates fresh performance and efficiency features, alongside new low-power modes and enhancements tailored for mainstream consumer devices. Furthermore, it maintains the capability of scaling up to 14 cores for high-performance scenarios.
These advancements result in substantial power reductions of 50% for typical workloads and a 60% decrease in cache miss power across the CPU cluster, thereby mitigating leakage and enhancing the battery life of consumer devices.
ARM highlights the availability of low-power modes such as half slice power down and quick nap. Additionally, these enhancements provide support for a wide array of low and high-intensity AI-based tasks, spanning from biometrics and speech-to-text to AI-driven smart camera applications, content creation, and ML-powered AAA gaming.
Immortalis-G925 -
The new Immortalis-G925 boasts a 37% increase in performance (measured in fps) compared to last year's Immortalis-G720 while consuming 30% less power. ARM asserts that this equates to an average performance improvement of 46% over the Immortalis-G720. Specifically, Genshin Impact sees a 49% performance boost, while Roblox runs 46% faster than its predecessor.
Other popular mobile games experience performance enhancements ranging from 29% to 72%, including Call of Duty Mobile, Diablo Immortal, the Day After Tomorrow, Fortnite, and PUBG Mobile, as per ARM's claims. The Immortalis-G925 introduces a novel mechanism called fragment prepass, eliminating the need for applications to conduct object or primitive sorting. This enhances efficiency in reducing overdraws when processing geometry workloads.
Moreover, the Immortalis-G925 promises a remarkable 52% improvement in ray tracing performance for intricate objects. This encompasses a 27% increase in performance while maintaining the same level of accuracy in gaming scene transparency. Alternatively, developers have the option to slightly reduce transparency accuracy for a notable performance enhancement of up to 52%, along with a 57% reduction in memory traffic, according to ARM.
Additionally, the Immortalis-G925 now supports 50% more shader cores, with the largest configuration offering up to 24 cores compared to the previous generation's maximum of 16.
The Mali-G725 is scalable between 6 and 9 cores, targeting premium mobile handsets. While it offers the same API support as the Immortalis-G925, it delivers a captivating gaming experience for lower-tier devices not addressed by the Immortalis-G925. On the other hand, the Mali-G625 is scalable between 1 and 5 cores, catering to smartwatches and entry-level mobile devices.
AI Performance -
The Immortalis-G925 promises 34% faster AI and ML inference compared to the Immortalis-G720. While most AI processing occurs on the CPU, the GPU accelerates various AI tasks such as image classification, object detection, and natural language processing.
ARM aims to target a wide range of consumer devices across different tiers, from flagship to entry-level, with the new GPUs, delivering unprecedented levels of gaming and AI performance for faster, more immersive, and more intelligent visual experiences.
Arm’s new GPUs aim to deliver unprecedented gaming and AI performance for faster, more immersive, and more intelligent visual experiences. The first SoCs featuring these new CPUs and the Immortalis-G925 GPU are expected later this year, with the MediaTek Dimensity 9400 likely being the first to incorporate them.
