The new Arm Lumex CSS platform accelerates development cycles, enabling scalable, high-performance edge AI
The new Arm Lumex CSS platform accelerates development cycles, enabling scalable, high-performance edge AI
September 16 , 2025
Posted by KUKE
0
The new platform tightly integrates hardware and software, not only accelerating time-to-market but also meeting the growing demand for edge AI experiences across the entire ecosystem.
Mobile devices are progressively evolving into AI-driven tools capable of adapting, predicting and enhancing how people interact with the world. As on-device AI becomes more sophisticated and mature, the pressure on mobile chips continues to intensify.
Accelerated product cycles – where each new generation of flagship mobile devices launches faster than its predecessor – compel chip suppliers and OEMs to innovate within tighter timeframes, leaving little margin for error. Implementing the advanced packaging technologies required to sustain AI performance within the area and thermal constraints of mobile form factors presents a formidable challenge. Concurrently, the progression to smaller process nodes, such as 3 nanometres, introduces complex design challenges.
This is precisely why Arm introduced its integrated platform. By combining Arm CPU and GPU IP with a physically realised, deployment-ready software stack, it accelerates time-to-market while delivering industry-leading performance on the latest advanced process nodes. Arm Lumex represents Arm's new dedicated Compute Subsystem (CSS) platform, designed to meet the ever-growing demand for edge AI experiences on flagship mobile devices and PCs.
Redesigned for the AI-first era
Lumex delivers the latest co-designed, co-optimised Arm computing IP and advanced features for modular and highly configurable platforms:
New Armv9.3 C1 CPU cluster: delivering leading instructions per clock (IPC) performance and featuring integrated second-generation Arm Scalable Matrix Extension (SME2) units, enabling more responsive accelerated AI experiences on the CPU cluster.
The new Arm Mali G1-Ultra: delivering faster AI inference capabilities while offering desktop-level visual effects and a richer gaming experience through next-generation ray tracing capabilities.
The new system IP incorporates Arm SI L1 system interconnect and Arm MMU L1 system memory management unit: designed to eliminate system performance bottlenecks and reduce latency in inference-intensive and computation-intensive workloads.
3-nanometre-ready CPU and GPU physical implementation: engineered to deliver industry-leading power, performance and area (PPA) and accelerate the process of achieving flagship-level performance on-chip.
Accelerating AI performance across real-world scenarios through cross-CPU and GPU technology
In terms of the CPU, the Armv9.3 C1 CPU cluster featuring SME2 technology combines native support for mainstream frameworks and runtime libraries via Arm KleidiAI. This delivers significant speed improvements over the previous-generation CPU cluster under identical conditions across a broad spectrum of AI workloads, including typical machine learning (ML) inference, speech processing, and generative AI. This delivers a fivefold increase in AI performance alongside a threefold improvement in energy efficiency. These SME2-driven enhancements enable users to enjoy smoother AI interactions on their everyday consumer electronics while benefiting from extended battery life.
Moreover, owing to microarchitectural optimisations and tighter integration between cores, the Arm C1 CPU cluster has set new benchmarks in both performance and energy efficiency, as demonstrated by:
Across six industry-leading performance benchmarks, the average performance improvement over the previous-generation CPU cluster under identical conditions reached 30 per cent.
In mainstream applications (including gaming and video streaming), average speeds have increased by 15% compared to the previous generation CPU cluster under identical conditions.
In everyday mobile workloads such as video playback, social media and web browsing, power consumption is reduced by an average of 12% compared to the previous generation CPU cluster under identical conditions.
Compared to the previous-generation Arm Cortex-X925 CPU, the Arm C1-Ultra CPU delivers a double-digit improvement in instructions per cycle (IPC) performance.
The Mali G1-Ultra further enhances AI performance and efficiency, delivering a 20% increase in inference speed for AI and ML networks compared to the previous-generation Arm Immortalis-G925 GPU.
In gaming, the Mali G1-Ultra delivers threefold ray tracing performance thanks to the new Arm Ray Tracing Unit v2 (RTUv2), bringing high-end desktop-level visuals to mobile devices; In mainstream industry benchmarks and gaming applications—including Dark Zone Breakthrough, Fortnite, Genshin Impact, and Honkai: Star Rail—Mali G1-Ultra achieves a 20% increase in graphics performance.
Lumex's Scalable System Backbone
To support AI-first experiences, one cannot confine oneself to merely iterating on compute IP; mobile system-on-chip (SoC) must continuously evolve across the entire interconnect and memory architecture.
Consequently, Arm introduces a new scalable system interconnect, specifically optimised to meet the bandwidth and latency demands of high-intensity AI and other compute-intensive workloads. This ensures sustained performance leadership on Lumex without compromising system responsiveness. The new SI L1 System Interconnect features an industry-leading System-Level Cache (SLC) with exceptional area efficiency, reducing leakage power by 71% compared to standard compiled RAM and significantly lowering standby power consumption.
For partners, this interconnect technology delivers a highly flexible, scalable solution to optimise PPA requirements across diverse mobile and consumer electronics devices. The SI L1 system interconnect, designed for flagship mobile devices, features fully integrated optional SLC and supports Arm's Memory Tagging Extension (MTE) for best-in-class security. Meanwhile, the Arm NoC S3 on-chip network interconnect targets cost-conscious, non-coherent mobile systems.
In addition to new interconnect technologies, Arm has introduced a new generation of Arm MMU L1 system memory management units. This technology enables secure, cost-effective and scalable virtualisation across a wide range of mobile and consumer electronic devices.
Achieving industry-leading PPA through physical implementation
Lumex provides CPU and GPU implementations optimised for 3-nanometre process technology and ready for production, supported by multiple foundries. This enables Arm's chip partners and OEM manufacturers to:
Employ these implementations as flexible building blocks to focus on differentiated designs at the CPU and GPU cluster level;
Achieving outstanding frequency and PPA;
During the transition to the latest 3-nanometre process node, it helps ensure chips achieve first-time tape-out success.
Developers can immediately enjoy the benefits of Lumex
To fully unlock Lumex's performance potential, developers require early access to its capabilities before devices reach the market. To this end, Arm has launched a new suite of software and tools enabling developers to immediately commence prototyping, build AI workloads, and harness the full AI capabilities of the Lumex CSS platform. Specifically, this includes:
A complete Android 16-ready software stack, spanning from trusted firmware to the application layer;
The complete and free KleidiAI software library enabling SME2;
A brand-new top-down telemetry solution for analysing application performance, identifying bottlenecks and optimising algorithms.
KleidiAI achieved remarkable success in its inaugural year, having now been integrated into all major AI frameworks and widely deployed across diverse applications, devices, and system services such as Android. This lays a solid foundation for forthcoming initiatives, ensuring applications can immediately realise performance and efficiency gains for their AI workloads when Lumex-based devices launch in the coming months.
In graphics processing, with future Android releases supporting RenderDoc and unified observability tools like Vulkan counters, Streamline, and Perfetto via Lumex, developers can analyse workloads in real time, optimise latency, and precisely balance battery life against visual fidelity.
Laying the foundations for the next generation of mobile intelligence
Mobile computing is entering a new era, defined by how intelligence is built, scaled and delivered. As AI becomes the foundation of all experiences, platforms must possess the capability to predict, adapt, scale and accelerate future developments.
Lumex is engineered precisely for this future vision, delivering comprehensive benefits across the entire ecosystem. Whether for OEMs building and scaling innovative devices or developers crafting next-generation applications, Lumex empowers the Arm ecosystem to deliver differentiated, AI-first platforms and experiences more effortlessly – achieving smarter performance for faster, large-scale deployment.
English
