Arm’s latest Lumex chip designs aim to supercharge AI on smartphones, wearables, and even PCs without needing the cloud.
Key Takeaways
- Arm launched its Lumex Compute Subsystem, a new generation of chip designs optimized for on-device AI, spanning smartphones, wearables, and PCs
- Lumex supports four core types, from ultra-performance to ultra-efficient, enabling diverse AI and gaming workloads
- SME2 instructions and KleidiAI integration make Lumex chips AI-ready without requiring developers to change existing code
- New Mali GPUs under Lumex boost mobile gaming, delivering console-quality visuals and up to 20% better performance
What Happened?
Arm Holdings has announced Lumex, its latest chip design platform tailored to bring AI power directly onto devices. Unlike traditional models that rely on cloud processing, Lumex is optimized to handle AI workloads entirely on smartphones, smartwatches, and even PCs. This means reduced latency and improved efficiency.
Lumex Powers On-Device AI Without Cloud Reliance
Arm’s Lumex Compute Subsystem (CSS) introduces four scalable core types designed to match various device needs, from flagship phones to tiny wearables. These designs allow manufacturers to integrate AI capabilities directly into devices without needing cloud support.
- C1-Ultra: Delivers top-tier performance, offering a 25% single-thread boost over Arm’s previous generation. Ideal for high-end smartphones handling large AI models, advanced photography, or generative AI tasks
- C1-Premium: Packs near-Ultra performance in 35% less space, balancing performance with efficiency for mid-range devices
- C1-Pro: Efficiency-focused, with a 16% gain in sustained performance, ideal for media playback and streaming AI tasks
- C1-Nano: The most power-efficient core, using 26% less energy, built for ultra-small devices like smart rings and fitness trackers
The designs are optimized for 3-nanometer manufacturing nodes, such as TSMC’s process, which is also used by Apple’s latest iPhone chips.
SME2 and KleidiAI Make AI Integration Seamless
To boost AI capabilities further, Lumex CPUs include Scalable Matrix Extension v2 (SME2) instructions. These enable CPUs to efficiently process the matrix operations at the heart of modern AI, essentially turning the CPU itself into an AI coprocessor.
Arm also unveiled KleidiAI, a software layer that works with major AI frameworks like PyTorch’s ExecuTorch, Google’s LiteRT, Microsoft’s ONNX Runtime, and Alibaba’s MNN. This allows app developers to benefit from performance boosts without modifying any code.
According to Arm, this tech leads to:
- Up to 5X AI performance increases
- 3X more efficient AI processing on battery-saving cores
- 4.7X faster speech-based workload handling
- 2.8X quicker audio generation
Major companies like Samsung, Apple, and MediaTek are already exploring how Lumex and SME2 can improve AI performance and responsiveness across their devices.
Mali G1 GPU Brings Console-Quality Graphics to Mobile
In addition to AI, Lumex also powers Arm’s new Mali G1-Ultra GPU, promising a 20% graphics performance gain and 9% lower power use per frame over the previous Immortalis G925.
This GPU supports:
- Ray Tracing Unit v2 (RTUv2): 2X better lighting, shadows, and reflections
- Enhanced performance for top games like Fortnite, Genshin Impact, and Honkai Star Rail
- Upgraded performance on sub-flagship devices with Mali G1-Premium and G1-Pro GPUs
Arm is betting that mobile gamers will soon expect the same graphics quality found on PCs and consoles and Mali G1 is a major step in that direction.
Strategic Market Push
The Lumex platform is part of Arm’s broader Compute Subsystems business, designed to help manufacturers get AI-powered devices to market faster with pre-packaged solutions. While Arm does not yet manufacture its own chips, the company has hired key personnel and is investing heavily in that direction.
To capture interest from leading handset makers beyond Apple and Samsung, Arm is also hosting a dedicated launch event in China, recognizing the region’s dominant role in global smartphone production.
SQ Magazine’s Takeaway
In my experience, on-device AI is the next major leap for mobile computing. What Arm is doing with Lumex is more than just a chip update. It is about empowering real-time, private, and power-efficient AI for everyone. No more waiting for cloud responses or draining your battery to run a smart assistant. This is the kind of foundational tech shift that often goes unnoticed by average users, but it’s what makes your next phone feel faster, smarter, and just more seamless. I found the KleidiAI integration particularly clever. Letting developers enjoy better performance without rewriting a line of code is a win-win.
