Multimedia Engine Performance
As we’ve already noted above, the integrated graphics core of any modern hybrid processor incorporates a specialized functional subunit for video content processing, i.e. decoding or encoding. Intel calls this subunit Quick Sync. AMD’s subunits of this kind are referred to as UVD3 and VCE. Besides the different names, they vary in their operating principles. While AMD makes use of the shader processors available in the GPU, Intel adds specialized limited-functionality subunits which offer hardware acceleration for typical video encoding and decoding tasks. It is no wonder then that there are substantial differences in the way the Richland and Haswell-based processors handle multimedia loads.
We can even see them while just watching videos. In the majority of cases, modern processors have no problems at all playing HD video in various formats. Their hardware decoding works perfectly even if they need to reproduce a 1080p video stream at 60 fps and high bitrates. You may only encounter some playback defects when your video clip is encoded with unusual parameters. And even in this case, the problem is likely to be due to some flaw in the player software rather than to the hardware multimedia engine.
On the other hand, higher bitrates have been getting more common, so our hybrid processors may not have enough resource to decode them correctly. Since this is quite an important application, we want to check out how the Richland and Haswell-based processors cope with 4K videos. In this test we use a widescreen 4K video clip with a resolution of 3840x2160 pixels and a frame rate of 30 fps encoded with the H.264 codec with a bitrate of about 100 Mbps. Even with full use of their hardware capabilities, the multimedia engines may have problems with that clip as we checked out with the software player Media Player Classic – Home Cinema version 188.8.131.5213 with K-Lite Codec Pack 9.9.5 and with video decoding enabled via LAV Filters 0.58.0.
The next diagram shows the load on the processors’ x86 cores.
Surprisingly enough, none of the Intel processors, including the Ivy Bridge series, has any problems reproducing 4K video content whereas AMD’s Trinity and Richland-based APUs fail completely. It is clear that the UVD3 engine cannot decode the high-resolution video, therefore these APUs have to make heavy use of their x86 cores. Their x86 performance is not high enough to correctly decode the 4K video clip, so we have nearly 100% CPU load and dropped frames. Thus, Socket FM2 platforms seem to be useless when it comes to 4K video formats, which puts their perspectives in modern multimedia computers and HTPCs under question. Intel’s solutions seems to be a much better choice for HTPCs thanks to the higher performance of their integrated multimedia engine.
Video transcoding is yet another popular video-processing load. To test the related hardware technologies, we use CyberLink MediaEspresso 6.7 which supports both Intel’s Quick Sync and AMD’s VCE. The test consists of transcoding a 1.5GB 1080p video clip in H.264 format (a 20-minute episode of a popular TV series) into a lower-resolution version for watching on an iPhone 4S. So, the target format was H.264, 1280x768 pixels with a bitrate of about 6 Mbps.
Intel’s Quick Sync technology is way better than AMD’s VCE in this test. Any of the Haswell and Ivy Bridge series CPUs enjoys a huge advantage over any of the Richland or Trinity series APUs. Intel’s transcoding makes of specialized hardware subunits whereas AMD’s solution is to make all computations on the stream processors. Considering the specifics of the transcoding algorithm, the latter solution doesn’t look optimal.
By the way, we can note that the Intel HD Graphics 4600 is slower than Intel’s earlier integrated graphics cores in terms of transcoding. It doesn’t mean that Quick Sync has become worse. Intel has just implemented new speed/quality profiles which focus on higher-quality results. The transcoded video outputted by the Haswell’s integrated graphics core is better by default than what we get with the earlier solutions, but at the expense of some speed.
Every time we talk about video transcoding technologies implemented in today’s graphics hardware, and especially about Intel Quick Sync, we have to admit that they are largely supported by not-very-popular commercial utilities. This situation is about to change, though. For example, QuickSync and OpenCL have been introduced into a beta version of the free software tool HandBrake. Let’s see what performance we can expected from the tested graphics cores if we use HandBrake for our transcoding.
Intel’s Quick Sync is quite astonishing again and ensures much higher speed in comparison with the Socket FM2 configurations. AMD’s APUs may improve, though. The OpenCL support in HandBrake is not yet polished off. The utility only uses OpenCL to resize the video frames but not to recompress them. That’s why AMD products are going to perform better in the future versions of HandBrake.