Arm Unveils New Lumex Mobile Platform with Enhanced CPU and GPU for AI-Optimized Smartphones
Arm has announced its latest mobile platform, Lumex, featuring revamped CPU and GPU architectures, along with reengineered interconnect and memory management systems, all tailored to support the upcoming wave of AI-enabled smartphones.
A Shift Toward Fully Integrated Solutions
Based in the UK, Arm has been steadily moving away from merely providing processor cores to offering more integrated solutions. The Lumex compute subsystem (CSS) exemplifies this evolution, integrating advanced system-level components to enhance performance and efficiency.
Performance Improvements and Power Savings
With each iteration, Arm continues to improve upon previous designs. The Lumex platform boasts an average 15% boost in CPU performance and a 20% increase in GPU capabilities, all while reducing power consumption by 15%. These enhancements prepare the platform for demanding AI workloads in mobile devices.
Focus on AI Acceleration: SME2 Extensions
Central to Lumex is the implementation of Arm's SME2 Scalable Matrix Extensions, designed explicitly for AI acceleration. Stefan Rosinger, Senior Director for CPUs at Arm, highlighted that SME2 enhances AI compute performance by an order of magnitude, offering lower power consumption and faster computation times—key benefits for mobile devices.
Expected Deployment Timeline and Manufacturing
Arm anticipates Lumex will be integrated into smartphones and other devices within the next year or so. Built with 3nm manufacturing processes, chips based on Lumex are expected to achieve clock speeds exceeding 4 GHz, promising significant performance improvements.
Multi-Core CPU Design: The C1 Cluster
Unlike traditional designs, all cores in the Lumex CPU cluster are labeled C1, with the top-tier performance core named C1-Ultra. The cluster supports four core types:
- C1-Ultra: Highest performance cores
- C1-Premium: Next tier
- C1-Pro: Balanced cores
- C1-Nano: Power-efficient cores
Depending on device class, configurations may include:
- Flagship phones: 2 C1-Ultra + 6 C1-Pro
- Sub-flagship: 2 C1-Premium + 6 C1-Pro
- Mainstream: 4 C1-Pro + 4 C1-Nano
Advanced GPU: Mali G1 Series
The Mali G1 GPU in Lumex spans various grades:
- Mali G1-Ultra: Equipped with Arm's redesigned Ray Tracing Unit (RTU), delivering 40% higher performance than the previous Immortalis-G925 and supporting high-quality graphics.
- Mali G1-Premium & Pro: Feature increasing numbers of shader cores—up to 9 for Premium and over 10 for Ultra—with the latter adding ray tracing capabilities.
The GPU also accelerates in-game AI features with support for FP16 matrix multiplication, enabling more efficient tensor processing while conserving memory bandwidth and power.
System-Level Optimizations
Lumex introduces new hardware components for system efficiency, including:
- System Interconnect (SI): Redesigned with channelized architecture to prioritize traffic and improve quality of service (QoS).
- System Memory Management Unit (SMMU): Optimized to reduce latency by up to 75%, enhancing AI workload performance.
Developer Support and Framework Integration
Arm emphasizes its collaboration with software frameworks by integrating KleidiAI libraries with popular platforms such as PyTorch, Llama, LiteRT, and ONNX—facilitating optimized AI workloads leveraging SME2 instructions.
The Role of AI Processing in Mobile Devices
Geraint North, an Arm Fellow, advocates for keeping AI processing within the CPU, citing its ubiquity across mobile phones. Unlike GPUs and NPUs, which vary across devices, the CPU remains a constant, enabling consistent performance.
However, industry analysts note that some, such as Gartner, recognize the importance of dedicated neural engines or NPUs in smartphones—especially for tasks like running small language models—predicting that premium and budget devices will increasingly incorporate such hardware.
Market Implications and Industry Outlook
Arm's strategic focus hints at its aim to capture significant market share in data centers and mobile AI. While the company faces competition from x86 architectures in servers and challenges in asserting dominance over Qualcomm's prevalent NPU-equipped chips, industry experts believe Arm's approach—integrating SME2 for AI workloads—may carve out a substantial segment.
Bob O'Donnell of TECHnalysis Research comments, "Arm's latest instructions could simplify AI acceleration, especially given the fragmentation and lack of standardization in NPU designs."
Conclusion
Arm's Lumex platform marks a significant step toward more integrated, AI-friendly mobile processors, combining high performance, power efficiency, and system-level optimizations to support the next generation of smart devices.