Gemini Google Logo

TL;DR

  • MediaTek has announced that its Dimensity 8300 chip now supports Google’s Gemini Nano model.
  • This means mid-range phones can now run generative AI features on-device.

Gemini Nano is Google’s smallest AI model, designed to run locally on smartphones and other low-power devices. The model is already used to power features on flagship phones, but it turns out that mid-range phones can get in on the action too.

MediaTek has announced that it worked with Google to integrate and optimize Gemini Nano for the Dimensity 9300 flagship processor and the mid-range Dimensity 8300 chipset.

“MediaTek and Google plan to unveil an APK running on Dimensity 9300 and 8300 to aid developers and OEMs in deploying Gemini Nano applications,” the Taiwanese chip designer added.

It’s no surprise to hear that Gemini Nano runs locally on the Dimensity 9300. Google previously noted that the latest flagship chipsets from Qualcomm, MediaTek, and Samsung all support Gemini Nano and the accompanying AICore suite.

Opening the door for generative AI on mid-rangers?

What is noteworthy, however, is the fact that the AI model now supports (and is optimized for) a mid-range smartphone processor. This suggests you won’t need to buy the latest high-end phone to have advanced on-device AI features.

Gemini Nano powers several features on flagship Google and/or Samsung phones today. This includes Magic Compose for adjusting the tone of messages, summaries in the Google Recorder app, text summaries, and Smart Reply in Gboard.

It’s worth noting that the only globally launched phone with the Dimensity 8300 is the POCO X6 Pro. But we hope this chipset and Nano-powered AI features come to more devices.

MediaTek says the Dimensity 8300 is capable of supporting large language models with up to 10 billion parameters, but the company was mum on the chipset’s supported token rate (i.e. processing time for AI models). So it’s theoretically possible we’ll see slower performance on Dimensity 8300 phones when running various on-device AI features.

Source link