To make YouTube work, where millions of videos are streamed in real time to hundreds of millions of users per day, tremendous power is needed in its data centers to deliver this service without any performance and speed issues of any. so. This includes specialized hardware developed by Google itself to speed up the quality of its streaming service.
Google receives 500 hours of content per minute per minute, which is enough to overwhelm even the most advanced processors. This requires the development of specialized equipment.
Argos, Google’s ASIC to speed up YouTube
This is why Google created a custom chip, an ASIC which they named Argos, and is responsible for transcoding video files uploaded by users, as well as real-time video streaming. Thanks to Argos, 4K files are quickly available at different resolutions and bit rates out of the box.
Argos can be found in thousands of units on YouTube servers in the form of a PCI Express card, performing a task that much more expensive hardware would be required if a server processor or even a GPU were used for it. Assuming a huge cost in infrastructure and energy. Demonstrate how accelerators for specialized tasks are the future of computing performance.
To design its custom chip, Google has needed a total of 100 engineers since 2015. They are currently deploying the second generation of Argos on YouTube servers. Remember that Google, although not competitive in the hardware market, has developments for its own use, as is also the case with its Tensor Processor Unit for artificial intelligence algorithms.
What is video transcoding?
To understand what Argos does, we need to understand what video transcoding is, which is taking a video file in a specific format, resolution, and bitrate in real time.
This is important for offering streaming videos for different connection types and resolutions. For example, when the resolution of a video on YouTube is changed in real time, transcoding occurs because the video from YouTube’s servers has undergone format and bit rate conversion.
Transcoding hardware is not only used on YouTube, but on all content streaming platforms, including services like NetFlix. As well as in Cloud Gaming. It is a market in which GPU designers orient their designs and in which they will compete with ASICs specializing in these tasks, such as Argos.