Graphics Reference
In-Depth Information
1.4
Complexity, Parallelism, Hardware, and Economies
of Scale
When a technical design such as HEVC is new, its practicality for implementation is
especially important. And when they emerged as new standards, H.261, MPEG-2,
and AVC were each rather difficult to implement in decoders—they stretched the
bounds of what was practical to produce at the time, although they each proved to be
entirely feasible in short order. In each of those cases, major increases in computing
power and memory capacity were needed to deploy the new technology. Of course,
as time has moved forward, Moore's law has worked its magic, and what was once
a major challenge has become a mundane expectation.
Thankfully, HEVC is less of a problem than its predecessors in that regard [
1
].
Although its decoding requirements do exceed those of the prior AVC standard,
the increase is relatively moderate. The memory capacity requirement has not
substantially increased beyond that for AVC, and the computational resource
requirements for decoding are typically estimated in the range of 1.5-2 times those
for AVC. With a decade of technology progress since AVC was developed, this
makes HEVC decoding not really so much of a problem. The modesty of this
complexity increase was the result of careful attention to practicality throughout
the design process.
Moreover, the need to take advantage of parallel processing architectures was
recognized throughout the development of HEVC, so it contains key new features—
both large and small—that are friendly to parallel implementation. Each design
element was inspected for potential serialized bottlenecks, which were avoided as
much as possible. As parallelism is an increasingly important element of modern
processing architectures, we are proud that its use has been deeply integrated into
the HEVC design.
Another key issue is power consumption. Today's devices increasingly demand
mobility and long battery life. It has already been well-demonstrated that HEVC
is entirely practical to implement using only software—even for high-resolution
video and even using only the computing resources found in typical laptops, tablets,
and even mobile phones. However, the best battery life will be obtained by the use
of custom silicon, and having the design stability, well-documented specification,
and cross-product interoperability of a well-developed international standard will
help convince silicon designers that investing in HEVC is appropriate. Once broad
support in custom silicon is available from multiple vendor sources, economies of
scale will further take hold and drive down the cost and power consumption to very
low levels (aside, perhaps, for patent licensing costs, as further discussed below).
Indeed, this is already evident, as some custom-silicon support is already emerging
in products.
Encoding is more of a challenge than decoding—quite a substantial challenge,
at this point. HEVC offers a myriad of choices to encoders, which must search
among the various possibilities and decide which to use to represent their video most
effectively. Although this is likely to present a challenge for some time to come,
