YouTube, On2, and the Economics of Ultra HD: Tallying the Costs
The nice thing about cloud encoding is that it comes with an assigned a fixed cost. Combine that knowledge with a spreadsheet, and you can go in all sorts of interesting directions. Like how much it costs to convert your content to VP9/HEVC and how many views you need to recoup that cost. Just for fun, we can apply these numbers to YouTube and VP9 and see exactly why Google bought On2.
Of course, we have to make some assumptions, as dangerous as that can be. Let’s look at the cost side first. None of the cloud houses that produce HEVC would share their pricing, other than noting that it was more expensive than H.264, which make sense since it’s at least twice as CPU-intensive, as is VP9. But today, Amazon charges $0.015 per minute for SD encoding and twice that for HD ($0.03). Let’s bump that up another $0.015 for HEVC or VP9 and assume a blended rate of $0.045 per minute for VP9 or HEVC. For convenience, let’s work in hour chunks and assume that all adaptive groups have nine streams. This gives us a cost per hour of $24.30 for VP9 or HEVC.
Now let’s look at the savings side. YouTube has been deploying VP9 since September 2013. If you download 1080p streams of H.264 and compare them to 1080p VP9 streams, VP9 is about 43 percent more efficient, while 720p streams are about 35 percent more efficient. Obviously, that will drop for lower resolution streams, so let’s assume that both VP9 and HEVC will deliver, on average, a 25 percent bandwidth savings. Translating this to bitrates, let’s assume that 1 hour of H.264 video consumes 1GB, for a per second data rate of 2.083Mbps. With a 25 percent savings, the VP9/HEVC file would take 0.75GB, for a data rate of 1.562Mbps. Sound reasonable?
Amazon Cloudfront’s fourth tier of pricing is $0.04/GB. At this price, the H.264 hour of video would cost $0.04 to deliver, the VP9/HEVC $0.03, for a savings of $0.01 per hourly view. So it costs you $24.30 to encode the streams, which you recoup at $0.01 per view. Divide $24.30 by $0.01 and you get 2,430 hours of viewing to recoup the encoding cost.
Applying this logic to YouTube is challenging. We can guestimate the savings side but not the cost. Specifically, according to YouTube’s website, the site currently streams 6 billion hours per month. According to my contacts, YouTube is delivering 30 percent of all content using VP9, which means the company is saving 25 percent of delivery cost on 1.8 billion hours per month, or 21.6 billion hours of video per year. Amazon’s cheapest bandwidth cost is $0.02/GB; let’s assume YouTube is paying half that, or $0.01/GB. The same 25 percent bandwidth saving means YouTube is saving $54 million per year by using VP9.
What we can’t easily estimate is how much YouTube is spending on encoding those videos. According to my contacts at YouTube, the company doesn’t encode all content to VP9, just “the most popular content.” According to old numbers, only 0.33 percent of YouTube Videos achieve more than 1 million views, so 1 percent would be a good chunk of the most popular videos. YouTube receives 100 hours of video every minute; if we assumed that YouTube’s encoding cost is 50 percent of what we used above, or $0.0225, and nine streams of VP9 for every input stream, encoding 1 percent of the incoming videos to VP9 would cost about $6 million, for a yearly savings of around $48 million, which will grow as the percentage of VP9 delivered streams grows. These numbers are in the realm of wild guesses, but you can finally get a sniff of why Google bought On2 in the first place.
This article appears in the April 2015 issue of Streaming Media as "The Economics of Ultra HD."
Video gamers can immediately take advantage of the new offering, as well as updated video game streaming software.
Which delivers better quality, encoding time, and CPU performance—HEVC or VP9? We put them to the test to decide once and for all.
Overcoming obstacles that prevented it from making the switch previously, YouTube says HTML5 video is now ready for the spotlight.
HEVC might be getting all the attention, but Google's VP9 appears to match H.265 in quality and might play a bigger role in the transition from Flash to HTML5