PARIS, Sept. 6 (Benin News / EP) –
Google updated their service Meet so that it consumes less resource of a computer when using its graphics processing unit (GPUs) instead of the central processing unit (CPU).
The Mountain View company highlighted the growing importance of video conferencing in recent years in a statement posted on its blog at Artificial intelligence (AI).
As a result, Google has included new features in its Meet service, such as implementing cloud processing to improve background effects and reduce device power during video calls.
These effects are made possible by the machine learning models that Google Meet works directly from the user’s browser without the need to install any additional “software”.
Until now, a computer’s processor was responsible for processing these Google machine learning models. With the latest Meet update, the GPU will take care of this task.
To achieve this, the company has implemented in this update a new segmentation model based on high-resolution (HD) input images, instead of low-resolution images, as before. A change that doubles the resolution compared to the previous model, according to Google.
However, their calculations show that doubling the input resolution quadruples the computational cost. A toll that the CPU of many devices cannot pay because it is limited by the size of the device. HD segmentation which must operate in parallel.
In this context, Google Meet will start relying on the GPU in mid-range and high-end devices, which have more powerful components. Google believes that the low-performance cores contained in this unit are ideal for processing high-resolution convolutional models, which in turn will reduce the use of computer resources.