Power consumption challenge for edge AI boom

April 12, 2019 //By Nick Flaherty
Power consumption challenge for edge AI boom
Power consumption in artificial intelligence at the edge of the Internet of Things is key issue for a booming market says a new report.

The Edge AI hardware market is expected to ship 610 million units this year and reach 1559.3 million units by 2024, at a CAGR of 20.64 per cent, according to the Edge AI Hardware Market report from MarketsandMarkets. 

It sees cameras and AI as major drivers for the growth in edge computing in IoT and dedicated AI processors for on-device image analytics but cites power as a major restraint. This is especially true for standalone smart devices with vision processing units (VPU) that help deliver power-efficient solutions for vision and artificial intelligence that are now entering the market. 

Such devices are designed to run deep neural networks at high speed and low power without compromising on accuracy, which enables devices to see, understand, and respond to their environment in real time. For example the Myriad X VPU introduced by Intel can achieve roughly 10X better performance that the previous version with multiple neural networks running simultaneously, to offer extended autonomous capabilities across wide range of applications such as drones, robotics, VR, and smart cameras. 

A majority of AI processors have an additional inbuilt Neural Processing Unit (NPU) that can handle significant amounts of parallel processing, uses low power, and is capable of cognitive tasks. With AI processors built in a phone and providing edge AI capabilitites, these tasks could be carried out right on the device, even without any connectivity. This would not only improve performance of the device, but also reduce strain on the battery, says the report.


Related stories:

Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.