The end of Moore’s Law forced YouTube to make its own video chip

In the context: About seven years ago, Partha Ranganathan realized that Moore’s Law was dead. Google’s VP of engineering could no longer expect chip performance to double roughly every 18 months without a significant increase in costs, and that was a problem given that he helped Google annually budget infrastructure spending. Faced with the prospect of getting a chip twice as fast every four years, Ranganathan realized they needed to change something.

Ranganathan and other Google engineers looked at the big picture and implemented transcoding (for YouTube) consumed most of the computing cycles in their data centers.

The off-the-shelf chips that Google used to launch YouTube weren’t all that great for specialized tasks like transcoding. The YouTube infrastructure uses transcoding to compress videos to the smallest possible size for your device, presenting them in the best quality possible.

They needed an Application Specific Integrated Circuit, or ASIC, a chip designed to perform a very specific task as efficiently and effectively as possible. Bitcoin miners, for example, use ASIC hardware and are designed solely for this purpose.

“What we really want to do is take all the videos that get uploaded to YouTube and transcode them into every possible format to get the best possible experience,” said Scott Silver, VP of Engineering at YouTube.

It didn’t take long to sell the ASIC idea to senior management. After a 10-minute meeting with YouTube CEO Susan Wojcicki, the company’s first video chip project was approved.

After a 10-minute meeting with YouTube CEO Susan Wojcicki, the company’s first video chip project was approved.

Google began rolling out its Argos Video Encoding Units (VCUs) in 2018, but did not publicly announce the project until 2021. At the time, Google stated that the Argos VCUs offered 20 to 33 times the performance of traditional server hardware. well-tuned transcoding software works.

Since then, Google has switched the switch to thousands of second-generation Argos chips on servers around the world, and at least two follow-up solutions are already in development.

The obvious motive for building your own chip for a specific purpose is to save money, but this is not always the case. In many cases, big tech companies are simply looking to create a strategic advantage with off-the-shelf chips. Consolidation in the chip industry is also playing a role, as the category now has only a couple of chip manufacturers to choose from, producing general-purpose processors that can’t handle specialized tasks.

See also: Death of shared computing

Jonathan Goldberg, head of D2D Advisory, said control over semiconductor companies’ product plans is at stake. “And so they build their own, they control the roadmaps and thus gain a strategic advantage,” Goldberg added.

Argos isn’t the only custom chip Google has created. In 2016, the company announced its Tensor Processing Unit (TPU), a specialized ASIC for AI applications. Since then, Google has released more than four generations of TPU chips, giving it an edge over its AI competitors. Google has also created the Pixel 6 series of smartphones using a custom-designed Tensor SoC, bringing together hardware and software for its mobile lineup.

Image credit: Studio Glazetics

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button