The first question is what video configuration can your viewer play. 1080p video looks great on my 12-core HP Z800, but what about on your client’s computer? Obviously, that depends. If you’re distributing to a single or a very limited group of viewers, you can ask about their CPUs and graphics cards, and then customize your video as necessary.
As a general rule, 720p video will play smoothly on most Core 2 Duo-based computers, particularly computers equipped with NVIDIA graphics cards, which typically offer GPU acceleration for Flash playback (and yes, I’m assuming Flash distribution). Core 2 Duo-based computers first became available in 2006, so if your end user’s computer is newer than that, it’s likely a Core 2 Duo computer or faster. To be safe with 1080p video, you need a very fast two- or four-core system. If the graphics are embedded or come from a vendor other than NVIDIA, expect lower performance.
If you’re distributing to a wider audience that includes pre- Core 2 Duo-based models, you’re going to have to be more conservative, or you’ll have to offer multiple streams. For example, in my tests, on a 2003-vintage Pentium 4 computer with NVIDIA graphics, a 720p stream consumed 78% of CPU, with some dropped frames. Video produced at 848x480 consumed 43% and played smoothly. As we’ll see in a moment, sites such as CNN.com and other three-letter networks seem to have adopted 640x360 for the faux-HD widescreen look; that consumed only 31% on the Pentium 4 and 35% on my Acer Aspire One netbook. If you need a single, lowest-common denominator stream, 640x480 is a good choice.
The second question is what do your viewers expect. Often, those expectations are dictated (or at least influenced) by the kind of videos they watch online everyday and how those videos are delivered. I surveyed more than 25 media sites, looking at news, sports, excerpts, and full episodes, and I was surprised at a seeming pullback from HD content. The average news configuration (seven samples) was 577x352, and the average resolution for full episodes was 627x367 (eight samples). The big exception was MTV, which offered up to 720p streams and reported that more than 72% of its audience was watching at resolutions of 768x432 and higher.
Looking at corporate sites, I found that the most aggressive B2C companies (Nike, Coca-Cola, Burberry, etc.) averaged 949x497 resolution, while B2B companies averaged 727x444, with Cisco stepping out at 902x507 and GE feeling it with 934x525.
Still, about the only market conspicuously offering HD content were showing movie trailers. The lesson here is that you should offer multiple video configurations and let the viewer decide which to watch. That includes some streams that offer awesome quality but are too large to stream in real time. Which leads us to the next question: What data rate can your viewers stream?
To assess this, I looked at the combined audio and video data rates used by media and prominent corporate sites. The eight full-episode videos averaged 866Kbps video and 128Kbps audio, for a combined data rate of about 1Mbps. At the high end of the media sites that I measured, my contact at MTV shared that 72% of the site’s viewers were able to watch full episodes at 1.7Mbps or higher.
In contrast, the 1080p movie trailer files that I sampled averaged about 10Mbps, which few viewers could view in real time. Instead, sites distribute these videos via progressive download, so the files are stored on your hard disk, at least temporarily, and will ultimately play at full speed once downloaded, if your computer is fast enough. Of course, trailers are short, and viewers may not mind waiting 5 minutes to watch a 2-minute trailer in pristine quality, though few viewers will wait 3 hours to watch a 45-minute episode of a prime time TV show.
This begs the obvious question of what data rate is needed to provide the necessary quality. Again, that depends. The metric I find most useful is bits per pixel, which is calculated by dividing the data rate by the number of pixels per second in the video (height x width x frame rate). Or you can just download the free media analysis tool MediaInfo, which tells you the bits per pixel of any file that you analyze. Bits-perpixel information is useful because it lets you compare files encoded at different resolutions and data rates.
For example, our eight prime time episodes had an average bits per pixel of about .12. In comparison, the 1080p movie trailers averaged about .26, more than twice as much, while the 720p videos were .35. Clearly, the value proposition is different: Networks are distributing episodes that we want to see but missed in prime time. The advantage here is convenience; the quality doesn’t have to be pristine. In contrast, movie companies are trying to convince us to go see the movie, so quality is paramount. Back to our wedding videos. It seems like viewers will be willing to wait a few minutes to watch a 2- to 3-minute video, which you can produce in trailer-like fashion. If you’re delivering a full production of 25 minutes or more, perhaps the prime time paradigm is more apt, and you should provide a smaller, lower-bit rate stream that they can view in more or less real time. Offering multiple options to suit the various connection bandwidths and playback platforms is always a good idea.
Jan Ozer (jan at doceo.com) is a frequent contributor to industry magazines and websites on digital video-related topics. He is chief instructor at StreamingLearningCenter.com.