Netflix Lukewarm on HEVC (to put it mildly)

“The bottom line is that in the great struggle to find the true signal among the noise, you should ignore the claims of those who create and sell the codec, and prioritize those who actually have to put it to use. But you knew that already, didn’t you?”

In a story reported by Streaming Media magazine, Netflix’s David Ronca, manager of encoding technology, stated, “We’re not seeing efficiency gains being claimed by HEVC encoding vendors,” though in two years, he expects “about a 20-30% efficiency versus x264.” Ronca was speaking at Streaming Media East, in a session that I didn’t make but will definitely watch when it comes on line. 

Here’s the relevant slide from Ronca’s presentation, the proverbial picture worth a thousand words (you can download the presentation here). 

netflixhevc copy.png

It’s an interesting take. At a high level, codecs go through three levels of review. The first is from the engineers and math geeks who create the codec; who have pretty much claimed that HEVC will deliver the same quality as H.264 at 50% the data rate.

Then you hear from the vendors who have to sell the codec. Here’s a quote from a white paper from Elemental Technologies, “Without sacrificing video quality, HEVC can reduce the size of a video file or bit stream by as much as 50% compared to AVC/H.264 or as much as 75% compared to MPEG-2 standards. Telestream echoed this sentiment in its own white paper, as shown in this chart. 

telestreamhevc.png

Then comes the findings of those who have to actually use the codec. Enter Netflix’s Ronca, who sees “no efficiency gain” as compared to x264.

What’s my take? As I explained in my Streaming Media session on HEVC, still-frame comparisons between HEVC and H264 do tend to support the 50% efficiency claims, particularly at 1080p resolutions and above. However, I’ve looked at multiple HEVC encodings of my standard test file, and during playback, HEVC consistently exhibits more motion artifacts than H264, even when encoded at 75% the rate of H.264. I warned that you should ignore peak signal-to-noise (PSNR) and similar analysis because they’re still-image only, and don’t measure quality during real time playback. I point this out clearly in my presentation, which you can download along with sample files here; when the Streaming Media video of the session becomes available, I’ll embed that as well.  

Obviously, Netflix is distributing movies, not still images, and Ronca’s findings are even less encouraging than mine. The bottom line is that in the great struggle to find the true signal among the noise, you should ignore the claims of those who create and sell the codec, and prioritize those who actually have to put it to use. But you knew that already, didn’t you?

About Jan Ozer

I help companies train new technical hires in streaming media-related positions; I also help companies optimize their codec selections and encoding stacks, and evaluate new encoders and codecs.

Check Also

Streaming Media 101: Training for App & Player Development/Testing Professionals