ARM has unveiled its next-generation CPU and GPU designs, alongside the introduction of the Lumex compute subsystem (CSS), a turnkey solution designed for 3nm semiconductor nodes. This new offering aims to streamline chipset development for ARM’s partners, allowing them to focus on CPU and GPU cluster differentiation.
The Lumex CSS emphasizes customization, with the C1-DSU enabling CPU configurations from 1 to 14 cores, incorporating up to three core types selected from C1-Ultra, C1-Premium, C1-Pro, and C1-Nano options. The Mali-G1 GPU offers scalability from 1 to 24 shaders. This level of customization allows partners to tailor their chipsets to specific use cases and applications.
ARM highlights the System Interconnect L1 within Lumex, which houses a system-level cache (SLC) and achieves a 71% reduction in leakage compared to standard RAM, leading to lower idle power consumption. The Memory Management Unit L1 facilitates secure and cost-efficient virtualization, allowing multiple operating systems to run simultaneously on a single device. This feature is particularly important for applications that require multiple operating systems, such as in automotive and industrial control systems.
According to ARM, a C1 CPU compute cluster delivers an average of 30% higher performance across six industry benchmarks. Gaming and video streaming experience approximately 15% faster performance, while workloads such as video playback, web browsing, and social media exhibit around 12% greater efficiency compared to ARM’s prior designs. These performance improvements are expected to enhance the overall user experience in a wide range of applications.
The high-end C1-Ultra CPU provides double-digit instructions-per-cycle (IPC) improvements over the Cortex-X925. The Mali-G1 Ultra GPU is reported to be 20% faster in rasterization and twice as fast in ray-tracing tasks compared to the Immortalis-G925. These advancements are likely to be particularly beneficial for demanding applications such as gaming and professional graphics rendering.
The new Scalable Matrix Extension 2 (SME2) enhances on-device AI performance. The new CPUs are up to 5x faster and up to 3x more efficient than earlier designs, while the G1 GPU achieves a 20% performance increase in inference compared to the previous generation. SME2 is expected to play a key role in enabling more complex AI applications on devices, such as enhanced image and speech recognition, and more sophisticated machine learning models.
