by Ryan Seghers, November 12, 2013
This is an addendum to the original test. The purpose is to investigate the characteristics of the new Lookahead rate control method introduced by Intel in the Haswell generation chips and exposed by q264 version 0.3.3. See the original test analysis for setup, caveats, and background.
When using Lookahead, the encoder pre-analyzes (looks ahead) some number of frames to determine their complexity. Given the complexity of the future frames the encoder can more intelligently allocate bits to the current and frame to improve encoding efficiency (quality per bit).
q264 version 0.3.3 adds the -la argument. It takes an integer number of frames specifying how many frames to look ahead, with 0 meaning that the encoder can decide for itself (by some unknown method). For this test -la 0 was specified to let the encoder choose.
This test used Intel graphics driver version 188.8.131.5257.
q264 vs x264: Speed on has1
These compare q264 and x264 encoding speed across the different H.264 profiles on has1. (Note that there are several caveats about this comparison as described in the original article.) Lookahead entails a big drop in encoding speed, but in this test scenario Intel Quick Sync hardware is still significantly faster than x264.
The fastest x264 speed-quality setting approaches q264, but the quality sacrificed at that speed-quality setting (until very high bitrates) is quite large so that in reality one would probably not want to use that x264 preset. So the practical speed gap is larger than appears on the charts. The speed gap between medium speed-quality settings from x264 to q264 is generally more than 2X.
q264 vs x264: Lookahead 1080p Quality on has1
These compare q264 and x264 encoding quality across the different H.264 profiles.
Lookahead (abbreviated LA in the charts) makes a dramatic improvemnt in q264 video quality, especially for the Main and High profiles. With Lookahead q264 keeps up with x264 further into higher bitrates before falling off due to the trajectory issue noted in the original article.
q264 vs x264: 720p Quality on has1
To pursue the PSNR trajectory issue, here is the equivalent chart on a 720p source.
This doesn't have the trajectory problem of the 1080p case, so in this case q264 with Lookahead and middle usage (4) matches x264 on medium preset. This is a nice milestone for Quick Sync, to match x264 in quality at its medium preset.
The relatively new Lookahead rate control mechanism significantly improves Intel Quick Sync video quality, at an equally significant performance cost. Given that Quick Sync was already extremely fast but in general is behind on quality, Lookahead is a welcome addition.
Unfortunately Lookahead doesn't solve the 1080p quality-vs-bitrate trajectory problem noticed in the original article. Lookahead keeps q264 competitive longer, but the trajectory difference dominates at higher bitrates.
For 720p video the Lookahead rate control method enables Quick Sync to match x264's quality at medium preset, which is a notable achievement. x264 still has more headroom to increase quality with its slower presets, but for users that use x264's medium preset (or faster), q264 can finally match that quality.