For 2025, Arm has created an official brand for its mobile CSS, and is moving away from its historic Cortex brand for its CPUs.
www.eetimes.com, Sept. 10, 2025 –
This week’s launch of Arm’s Lumex, the new Compute Subsystem (CSS) for mobile devices brings a series of new brands from the company as it adds new CPUs, GPUs, system intellectual property (IP), and new Scalable Matrix Extensions version 2 (SME2) with accompanying hardware acceleration. In addition to achieving double-digit performance gains for the sixth generation, the Lumex platform is focused on AI performance as close to the CPU and data as possible.
An Arm CSS is a complete group of hardware on software for building a compute subsystem. A CSS offers all the available Arm compute and system IP that could serve as a complete product or the starting point for further design with additional functional IP blocks from Arm, other companies, or designed in-house. For a company developing a new mobile SoC, the CSS can reduce the design time by months.
For 2025, Arm chose to create an official brand for its mobile CSS – Lumex. Along with the brand name, Arm is moving away from its historic Cortex brand for the CPUs to C1-Ultra, C1-Premium, C1-Pro, and C1-Nano, as well as the DynamIQ Shared Unit (DSU) to the C1-DSU. Additionally, Arm is shifting from the Immortalis brand for GPUs back to Mali with the Mali G1-Ultra, G1-Premium, and G1-Pro.
The Performance
Built on the new Armv9.3 architecture and optimized for the 3nm process node, the latest CPU cores provide greater performance efficiency over the previous generation. According to Arm, the C1-Ultra performance cores and C1-Pro cores that are included in its mobile flagship reference design, deliver a 25% improvement in single-threaded performance and 16% improved gaming performance over the prior generation Cortex-X2 and Cortex-A720 CPU cores, respectively. Meanwhile, C1-Nano provides a 35% power reduction from the prior generation Cortex-A520 cores. The newest member of the product family is the C1-Premium, which is offered for sub-flagship products and offers a 35% smaller size than the C1-Ultra. Arm claims an average 30% performance uplift with this generation of CPUs.
With a focus on mobile gaming, the new GPU G1-ultra provides up to a 2x improvement in ray tracing performance plus a 20% uplift in graphics and inference performance over the previous generation Immortalis-G925. The Mali G-Ultra includes new significant enhancements to the shader cores, faster interconnects, support for MMUL.FP16 precision, and new RTUv2 ray tracing cores. The Mali G can also be scaled according to the desired performance requirements. Arm’s performance data was benchmarked on a 14-core GPU. Additionally, the C1-DSU is enhanced with better power management features to provide an average 11% reduction in power over the DSU-120.
Adding AI acceleration
The real change, however, comes in the support of SME2 enabled through Arm’s Kleidi AI software platform. In addition to the CPU architectural changes, the SME2 solution includes a matrix multiple-accumulate (MAC) hardware core that is embedded in the CPU complex. Think of it as a neural processing unit (NPU) without the overhead. In fact, customers can use one or two cores depending on the desired performance. The SME2 cores are designed to provide acceleration for small language models (SLMs) on device. According to Arm, the SME2 should handle average SLMs in the 7 billion parameter range with an average 3.7x performance improvement over just running the workloads on the CPUs. While the GPU or NPU could provide even more performance efficiency, the addition of the SME2 gives the AI developer another option for running AI workloads on a device without having to optimize to a specific GPU or NPU.
Final thoughts
Arm continues to innovate at the core and platform level. Achieving impressive performance with each generation. The CSS platform is also a huge benefit to customers looking to leverage as much Arm IP as possible to reduce production risk and time-to-market. However, not having an NPU block is still a significant gap in the Arm portfolio, particularly for new customers seeking a comprehensive solution, which Arm has been moving closer to with each generation. Tirias Research believes that Arm will need to offer an NPU at some point to further improve the competitiveness of its platform.
And while the Lumex platform is aimed at mobile devices, particularly smartphones, it has enormous potential in many other applications, particularly embedded/IoT applications. The C-Nano cores combined with SME2 support could be used to build some very compelling low-power solutions with AI capabilities.