Argos custom server technology chips was developed by Google for their video streaming subsidiary, YouTube.
Higher than traditional infrastructure would yield 20-quotients can expect from using the custom Argos encoding chips will be around or slightly above the current year’s common TIFF encoding results
Google has two Argos ASICs running on a full-length PCIe card that are connected to an aluminum heat sink through the Video (Trans) Cunitooleramic Unit (VCU).
In the aforementioned paper, Google goes on to state that each chip comes with 10 “encoder cores”. 3 reference frames allow each core to encode up to 2160p (2,160 frames per second) in real time, and 60 FPS for each core
As the search giant Google has stated, designing video transcoding ASICs for the data center is a very different challenge than doing so for consumers’ gadgets. where the small devices are just thousands of square feet in size, there is an increased emphasis on cost-efficiency.
As the demand for power and area decreases in personal computers, so does the cost in the data center ASIC. Because of this, “multi-ASIC” deployments are able to cut the overall amortization costs in the data center while maintaining high throughput, but are disadvantaged compared to single-ASIC models.
according to the Argos, thousands of Argos, the number of chips are running in Google data centers at this time. In keeping with its tradition, Google, rather than running dedicated VCU machines in its facilities, will put them in all of YouTube’s data centers.
According to a recent blog post from the company’s lead engineer, the organization is currently working on a follow-up product which is better at compressing video, called AV1, which is said to use a more efficient coding method and put more computation demands on the encoding stage.
Also this year, Google has integrated their own TPU family of processors and Titan chip, and today they have also hired former Intel Chief AI researcher Uri Frank to build a System on Chip.