A new solution from Cinemartin leverages pro and consumer Nvidia GPUs to achieve speedy 4K and HD encoding.
When the successor technology to the ubiquitous h.264 codec was standardised as h.265, we published an in-depth piece describing how its compression techniques built on its predecessor. In that piece, we discussed the key limiting factors in video compression. We discovered that the thing that keeps us from going further and further is not a lack of mathematical cleverness, at least not at the moment.
Instead, the main limiting factor is how difficult a codec is to decode, because the consumers who will receive the video content want their home electronics to be cheap. The aim with h.265 is that it will provide the same image quality at half the bitrate, with a maximum of three times the work required to decode it compared to h.264. This seems like a bad deal – three times the work for twice the effectiveness – but it's a good use for general advances in microelectronics, even if that 2:1 advantage isn't quite always seen in practice.
What's often overlooked is the difficulty of encoding material. On one hand, that's sort of OK, because content creators only need to do it once, whereas it'll be decoded many times. On the other, well, every video chat application needs to encode video and that's hard work. Every codec since Cinepak, of the early 1990s, has been highly asymmetrical, requiring a lot more work to encode than decode, and that remains the case with h.265. Recognising this, the designers made specific decisions to allow it to be more easily encoded on multi-core systems, which were uncommon at the time h.264 was developed, but are par for the course now.
Faster-than-realtime 4K encoding?
Taking advantage of this, particularly in respect of the enormous core counts offered by modern graphics cards, is Barcelona-based Cinemartin. It's tricky company to describe, with offerings ranging from a 15mm rod baseplate to various After Effects and Premiere plugins. Its most recent announcement concerns an HEVC encoder for its Cinec transcoding platform that's designed to use Nvidia GPUs, including both the Quadro series and the lower-cost GTX cards. Performance, as demonstrated in its video presentation (below), indicates faster-than-realtime encoding for 4K material and performance of several times realtime for HD. Encoding 4K h.265 is very hard work and standalone devices dedicated to the task are expensive.
There's always a need for caution in codec comparisons, of course. Most codecs can be set up to operate quickly and coarsely and there is no absolutely reliable way to compare the output of competing encoders. Sheer signal-to-noise numbers can be misleading, in terms of how the pictures look subjectively to humans. General-purpose applications for GPUs are no longer particularly new or exciting as an approach, although that's certainly the key to performance, in this case, and that faster-than-realtime result for 4K material is likely to turn heads, especially since it's shown on a GTX 980 graphics card worth only a few hundred units of currency.
If there's a more general thread to be drawn from this, it's the irresistible encroachment of software and general-purpose computers into realms previously occupied by expensive boxes. There are downsides to this, mainly because workstations and operating systems and software are, in combination, often less reliable than a fixed piece of custom-built electronics. The cost-benefit analysis, however, tends to obliterate those concerns fairly handily.
Check out the full press release for the Cinemartin HEVC encoder on Page Two!