The emerging High Efficiency Video Coding (HEVC) codec has revealed its versatility at the IBC 2013 trade show in Amsterdam, working equally well for SD content at low bit rates on mobile devices and high-resolution ultraHD delivered to wall-size displays. The full range of potential HEVC use cases was well covered by the demonstrations, with many vendors jumping on the bandwagon even when they had nothing of their own to show. Among the 12 or so vendors with HEVC encoders or decoders of their own, Ericsson highlighted the ability of HEVC to deliver high-quality content to mobile devices, while Elemental Technologies showed how HEVC, increasingly referred to as H.265, enabled far superior video quality at low bit rates compared with its predecessor H.264.
Envivio has taken this further by featuring a three-way shootout comparing HEVC with both H.264 and MPEG-2 for real-time video encoding, in this case, though, highlighting particularly the bit savings that can be achieved for a given video quality. The quality per bit gain is achieved via various techniques that enable the encoding algorithms to zoom in on the parts of the picture that are changing most between successive frames. The cost is in higher processing requirement, but this will be comfortably absorbed by improvements in available power at a given price on the encoding side. For decoding the emphasis has been on keeping the processing load as light as possible so that it can be done by small, relatively inexpensive handheld devices.
Several vendors demonstrated HEVC for online encoding using MPEG-DASH adaptive-bit-rate streaming (ABRS) at IBC. Thomson Video Networks showed its VisualOn's OnStream MediaPlayer+ on Nexus tablets decoding HEVC-encoded MPEG-DASH streams, while Allegro DVT showed how its AL1200 and AL2200 support TS output and MPEG-DASH with HEVC encoding.
One of the clear messages was how HEVC is accelerating the trend toward software rather than hardware encoding, according to the British analyst firm Rethink Technology Research. As the firm pointed out this is because HEVC, unlike H.264, has comprehensive in-built support for parallel processing, which dovetails well with the latest off-the-shelf multicore chips.
Until now software encoding has suffered from inferior performance compared with hardware methods and HEVC, with its greater computational demand, has been designed to take advantage of generic processors in order to contain costs. Then transcoding vendors can exploit the greater flexibility and upgradeability of the software approach. Some vendors have been betting on software winning the transcoding argument for some time, such as Elemental Technologies and Envivio. At IBC French encoding vendor ATEME was also demonstrating how a software approach meant that its Titan Live/Titan File carrier-grade video transcoding system for multiscreen delivery and its Kyrion DR8400 universal Integrated Received Decoder have both been upgraded to support HEVC, despite having been introduced over four years ago. Hardware encoders of that vintage would now only be of use for legacy infrastructures.
The distinction between the hardware and software approach for encoding or any repetitive process is becoming blurred, though, as Canadian video infrastructure company Haivision highlighted at IBC. Haivision’s approach would be classified as software, as it runs on the latest off-the-shelf Intel CPUs rather than dedicated silicon. But the company pointed out, these CPUs themselves incorporate multiple-GPU cores, which they utilize directly for both the encoding and decoding processes. In effect, the GPU cores are multiple processors exploited by HEVC’s parallel processing. In this way, the need for dedicated silicon even for high-resolution HEVC encoding is being eroded. HEVC encourages this trend through its implicit support for parallelism, especially via Wavefront Parallel Processing (WPP). WPP allows a picture to be split up into multiple components, while preserving links between them that take account of movement of objects such as heads or balls within the picture. It allows the components to be encoded in parallel without destroying the implicit links between them that enable movement to be properly taken into account of in frame prediction.
It is not quite correct, then, to say that software has won the encoding battle. It is more that the codec has evolved to take full advantage of dedicated features within generic processors.
