Android is about to make another Apple jump on Apple

Despite its opponents, the characteristics of AI remain the cornerstone of modern smartphone innovation. But quickly and effectively feed the most recent and best processors’ AI tools, whether it is a dedicated AI accelerator or a processor equipped with special instructions to speed up automatic learning workloads.

In a blogArm has more detailed what to expect from its CPU extension of Matrix Extension 2 (SME2) and has teased that the trick will arrive on future Android smartphones each time it abandons its new generation mobile processors. Based on the previous announcements, this will probably happen in the coming months and will be renamed under LUMEX Rather than the old Cortex brand.

For a quick overview, the SME, which is initially part of the Armv9 architecture, is a set of optional CPU architecture extensions and explicitly advanced to speed up matrix matrix operations – the type of complex multiplication on automatic learning algorithms. SME2, ​​which made its debut towards the end of 2022, is based on these features, which, according to Arm, allow mobile inference tasks in real time, from image processing and natural language to the generation of voice.

Arm also shares some figures, which means that SME2 looks like the real matter. According to ARM, the Google Gemma 3 model offers 6x AI faster responses with equipment improved by SME2 than without. He can perform a text summary of 800 words in less than a second on a single processor nucleus, although it is not clear to which CPU arm refers here.

All of this seems super promising for the future text summary and intelligent response features, which might seem much more reactive than today’s implementations. In all cases, the key part of the advertisement for consumers is confirmation that the material capabilities SME2 “arrive soon” on Android smartphones.

Until now, the SME has been mainly reserved for servers and work class processors because of its demanding nature. However, SME2 is designed to be more evolving, allowing deployment in low power levels such as laptops, tablets and soon high -end smartphones. While Android chipsets have mainly ignored the original generation of SMEs, Apple’s M4 chip – currently used in iPads – is the nearest mobile device with SME2 support. That said, Apple has not yet brought this functionality to its iPhone line. This opens the door to new generation Android devices to obtain an important AI performance advantage on their competitors.

Above all, Android is already configured for the SME2 support. SME2 is activated in Google’s XNNPACK Library for Android and is supported on several frameworks like Llama.CPP, MNN from Alibaba and Onnx de Microsoft. Similarly, developers already using the Kleidii de Arm software library (which fits into these frameworks) will automatically benefit from SME2 equipment once it is available in Android smartphones while retaining compatibility behind with SME and neon extensions. It is a big boost for easy adoption.

In justice, we already knew that SMEs came to future Mobile Arm processors Gracious from Chris Bergey from Arm to Computex. A presentation slide revealed that the Travis processor of the new ARM generation will feature SME2 (probably SME2, ​​based on this last announcement). According to rumors, this nucleus feeds the dimensity of Mediatek 9500 and potentially other new generation mobiles such as the Exynos de Samsung line. Google’s tensor is often lagging behind several generations, but it could adopt the capacities of Arm SMEs by the 2026 Tensor G6.

However, with Qualcomm down the personalized CPU road, the next Snapdragon 8 Elite 2 may not share the same identification information, which means that a large part of the new generation Android flagship products will not immediately jump to SME2. According to the first rumors, the 8 Elite 2 will support the SME1 and SVE2 extensions, which will always be a significant upgrade on the current configuration but will not include the same SME2 capacities as the internal ARM Lumex nuclei. It remains to be seen if Apple Silicon will catch up with its own implementation of SMEs in the next generation. It is possible, because the A18 chips are already on ARMV9.2, but it will have to bring the SME2 equipment adopted from the M4 in its personalized telephone processor cores.

In any case, faster automatic learning on the processor of your smartphone is a big problem. Better performance for a range of tasks, such as text summary, translation on devices and image recognition, is the obvious advantage for end users, but it is also a positive decision for the AI ​​development ecosystem as a whole. Despite the availability of Android NN, the specifically coding of tailor -made automatic learning equipment remains a chore for small developers, which are more likely to use the CPU race to reach as many devices as possible.

The handsets with more powerful instructions available directly on the CPU will be a boon for these use cases. If this turns out to be the case, this is another big victory for new generation Android smartphones, while its biggest rival, Apple, is already stuck in the starting blocks of the AI.