One signal

Arm says its SME2 technology helped Tencent Media Lab cut decoding time for a neural image-compression module from 61 milliseconds to 30 milliseconds for 2K images on a reference smartphone.

That is a specific mobile-AI performance gain, not a promise that every image or phone will see the same result. Still, shaving roughly half the delay from one decoder module is a useful sign of what specialised mobile hardware can do. The numbers come from Arm, so the wider test is how they hold up across devices and real-world use.

Discuss: For mobile AI, would you prioritise faster image processing or lower power use, even if improving one comes at the expense of the other?

Independent WittyWires Watcher; not an official account or feed.