Quote:
Originally Posted by hello_hello
The SSIM tuning was used for x264 but not for x265.
|
Yeah, that was a problem. We didn't realize that quality would be measured only with Y-SSIM, otherwise we would have added --tune SSIM to our command line for these tests, and x265 would have done much better.
Quote:
Originally Posted by hello_hello
According to those tests x265, which generally rated the best in respect to bitrate for a specific quality, still didn't do all that much better than x264. 74% the bitrate of x264 when encoding speed wasn't a factor and around 90% for more realistic encoding speeds. Wasn't the original promise somewhere in the vicinity of 50%?
|
You're right. These tests were forcing each encoder to run at a particular speed on a particular machine. At a given encoding speed, x264 can be run at relatively higher quality settings than x265. The promise of 2x the encoding efficiency of HEVC vs AVC is achievable, but not for free. You only get the full benefit of HEVC only if you pay for it with more computation (which means, more powerful hardware or slower encoding). The MSU test was based on x265 from early April. Since that time we've made a number of solid improvements in speed and quality, adding a massive amount of AVX2 assembly code optimizations, and improving many algorithms (--limit-refs, --limit-modes, --lookahead-slices). We believe a fair retest today would demonstrate much more impressive results for x265.