- ARM LUMEX fleas promise enormous improvements to AI on devices
- Its processors offer up to 5x best IA performance
- CPUs are considered the Central for AI
Arm has removed the envelopes from its new generation Lumex flea conceptions, optimized to execute certain local AI workloads on mobile devices.
Its architecture allows four different types of design, ranging from energy efficient nuclei for portable devices to high -performance nuclei for flagship smartphones.
Slacing of accelerated product cycles, which cause stricter time scales and a reduced margin for error, ARM says that its integrated platforms combine CPU, GPU and software batteries to accelerate the marketing time.
Arm lumex could be used in your next smartphone
ARM described Lumex as its “new calculation subsystem platform (CSS) to meet the growing requirements of AI experiences on devices”.
The CPU Armv9.3 C1 cluster includes integrated SME2 units for accelerated AI, promising 5x performance of better AI and 3X more efficiency compared to the previous generation.
Standard references see performance increase by 30%, with an acceleration of 15% in applications and energy use less than 12% in daily workloads compared to the previous generation.
The four proposed CPUs are C1-ULTRA for large model inference, C1-Primium for multitasking, C1-Pro for video reading and C1-Nano for portable devices.
The Mali G1-ULTRA GPU also allows 20% faster AI / ML inference than the Immortalis-G295, as well as improvements through games like 2x of better ray tracing performance.
LUMEX also offers G1-Premium and G1-Pro options-but no G1-Nano.
Interestingly, the arm positions the processors as an universal AI engine given the lack of normalization in the NPUs, even if the UBNs start to gain their place in the PC chips.
The launch with LUMEX is a comprehensive-Android 16 software stack, KLEIDIAI LIBRAISQUES COMPATIBLE SME2 and telemetry to analyze performance and identify the bottlenecks, allowing developers to adapt LUMEX to each model.
“Mobile IT is entering a new era which is defined by the way in which intelligence is built, scaling and delivered,” said the main director Kinjal Dave.
For the future, Arm notes that many popular Google applications are already compatible with SME2, which means that they are ready to benefit from IA features improved on AI devices when new generation equipment is available.
