2021-04-23 06:26:31
To handle the huge amount of videos uploaded to YouTube is a terrible burden for any company. On average, there are about 500 hours of video uploaded by users to this platform per minute. Not only do they have to convert compressed formats for a huge amount of video per day, these videos also have to be displayed under many different resolutions.
If you try clicking on the wheel image in the lower right corner of each 8K clip on YouTube, you will see that this video can be viewed under 9 different resolutions, including 144p, 240p, 360p, 480p, 720p, 1080p. , 1440p, 2160p and 4320p – for each video individually.
Google VCU video transcoder with 2 Argos chips on board
This burden is so great that even datacenters with thousands of Google CPUs have a hard time handling them, especially when it’s multi-purpose CPUs like Intel Skylake chips or T4 Tensor Core GPUs. Nvidia.
That’s why Google decided to build its own server chip for this giant video platform. Recently in a blog post, Google has detailed description of the new chip “Argos” for YouTube.
According to Jeff Calow of Google, these Argos chips have helped improve the efficiency of the video transcoder (VCU – Video (trans) Coding Unit) on “20-33 times more than previously optimized systems that are running software on traditional servers. “

The drawing depicts the Google Argos chip diagram with 10 Encoder Core coding cores inside
Each VCU is shaped like a graphics card with 2 Argos chips mounted on it. According to Google’s design drawings, each of these chips contains 10 “encoding cores”, capable of encoding 2160p resolution video, with frame rates up to 60 FPS in real-time using 3 reference frame.
According to Google’s statement, thanks to the new Argos chip, the 4K videos uploaded by users to YouTube will be ready to watch in just a few hours, instead of having to wait until a few days as before.
Besides helping users save waiting time when they want to watch high quality videos as soon as possible, Argos also brings significant financial benefits to Google. The company said that even including the research and development costs of the chip, Argos saved the company a huge amount of money by not having to resort to external chips like those from Intel and Nvidia.

Table comparing the video conversion efficiency between Google VCU and systems using Intel and Nvidia chip (Perf / TCO stands for performance per unit cost of owner)
The chip development project for YouTube has been launched by Google since 2015, and Argos has begun to be rolled out in Google’s data centers. In fact, a second generation Argos chip has begun to be deployed with a focus on the AV1, the new video codec standard that delivers better picture quality but also consumes more computing power.
See Arstechnica
#process #hours #video #uploading #minute #Google #designed #chip #YouTube #increasing #efficiency #times