Google Develops Video-Transcoding Chip For YouTube To Improve Computing Efficiency

Popular video-sharing platform YouTube wants to make processing videos on its website much more efficient. The company has developed a custom chip for video transcoding. The chip is called Argos. YouTube is a Google subsidiary. The search engine giant bought the platform in 2006. Argos will make processing videos easy and deliver good results. The Argos chip will improve in computing efficiency. The new chip for video-transcoding is a part of efforts in doubling down in custom compute. It will bring up to 20-33x improvement than the previous optimized system. The system used software on conventional servers. Google had earlier roped in a former Intel top executive to head server chip design.

Google has experience in developing a similar chip in the past. The video coding unit package is a full-length PCI-E card. It simply looks like a graphics card. A board comprises two chips. The chips are buried under a cooled aluminum heat sink. There is an 8-pin power connector. The connector will ensure a power supply to PCI-E. According to Google, each encoder core in real-time will encode 2160p. This will make videos look sharper as there will be less compression. Google said that the cards are designed specifically to slot into the warehouse-scale computing system. YouTube’s each compute cluster will have a section of VCU machines. The machines will be loaded with new cards.

Notably, videos are compressed before being sent to users. This is done so that YouTube sends the smallest amount of data with the highest quality. This transcoding is slow and costly. Besides, the system is inefficient on existing CPUs. The new chip comes in the backdrop of growing demand for higher quality video. Google said that YouTube’s new chip project began in 2015. Meanwhile, YouTube has started offering more control to its app users over video resolution. Google has added options to allow users to prioritize higher quality. The option will change the video quality depending on the network and data connection.