Lumex, the secret heart of devices: Arm ignites the spark of intelligence in the palm of your hand | Hardware near me | Types of computer hardware | Gpu vs cpu vs graphics card | Turtles AI
Lumex, the secret heart of devices: Arm ignites the spark of intelligence in the palm of your hand
From watch chips to gaming smartphones, Arm’s new 3nm architecture promises instant AI power, smooth graphics, and low power consumption, turning every device into a cloud-independent computing lab
Editorial Team10 September 2025
Arm unveiled Lumex, its new mobile AI platform, with four variants, from wearables to flagships, optimized for 3nm processors. It supports local AI, high performance, and new software libraries.
Key Points:
- Four Lumex variants: from ultra-efficient cores to the ultimate for high-end smartphones
- SME2 on Arm C1 CPU delivers up to 5x cloud-free AI performance
- Mali G1-Ultra graphics with double the ray tracing and 20% more efficiency
- Launch event in China to engage local manufacturers
Arm unveiled Lumex on September 10, 2025, a CSS (Compute Subsystem) platform designed to accelerate AI directly on-device: four chip types designed to cover applications ranging from watches to flagship phones, all optimized for the 3nm node (the same used in the latest iPhone chips).
At the heart of the innovation lies the new Arm C1 CPU cores (C1-Ultra, Premium, Pro, Nano), based on Armv9.3 and featuring SME2 Matrix Extensions: they enable up to a fivefold speedup in AI workloads on mobile devices, with reduced latency in voice and generative audio tasks. Lumex promises 12-15% power savings compared to the previous generation, while still delivering 15% better gaming and streaming performance.
On the graphics side, the Mali G1-Ultra GPU integrates a second-generation ray tracing unit (RTUv2) that doubles the performance of the Immortalis-G925 model, offering a 20% increase in power efficiency and higher frame rates in popular titles like Fortnite and Genshin Impact.
Lumex also includes a system interconnect network (SI L1) and an optimized MMU that reduces system latency by up to 75%, improves memory management, and provides QoS priority for the most demanding AI traffic. The approach is modular: partners can use Arm’s 3nm "off-the-shelf" hardware implementations or adapt the RTL code for specific customizations.
On the software side, KleidiAI libraries are integrated into the most popular frameworks PyTorch ExecuTorch, LiteRT, ONNX Runtime, and MNN to leverage the power of SME2 without source code modifications, improving portability and automatic acceleration of AI libraries.
Arm will host an event in China to showcase Lumex to major non-Apple and Samsung manufacturers, focusing its efforts on markets where demand for on-device AI is growing rapidly.
Overall, Lumex represents a significant step in the evolution of mobile chips: it combines high AI performance, console-quality graphics, and power efficiency, all without a cloud connection.
