At this rate, it would take a full two weeks to encode 10 seconds of HD video. On a Core i7. This is not merely slow; this is over 1000 times slower than x264 on “placebo” mode.
Purely for the sake of example, say a fast computer now is 100x faster than one in May 2010. It still takes 3.4 hours to encode 10 seconds of video. That's 100 CPU-days for a 2 hour movie.
As long as decoding is fast I don't see a problem. I would love to spend substantially more time encoding to get a better size/quality output, and I'm sure most mass-media distributors feel the same way.
It would be interesting if we ever ended up in a situation where compression is so expensive that only studios/distributors are able to throw the kind of CPU power at the problem that's needed to do it in a reasonable amount of time. I'm sure they wouldn't mind consumers being faced with having to spend 2x as much storage/bandwidth for unauthorized versions (or get less quality in the same file size). Though I'd bet such a state would only serve to spur distributed encoding schemes to ramp up…
>Computers have gotten much faster since May 2010?
Really?
I've wondered this for a while.... from my (idiotic consumer) perspective it seems like we are moving at a snail's pace. If I buy an iMac today, it comes with the exactly same processor as it did 13 months ago. Almost exactly the same as a year before that (all had quad cores at various configurations). And it's not really any different browsing custom PC vendors if I want to keep prices reasonable.
Shouldn't I be able to get a quad-core 7ghz for $2000 by now according to the corollaries of Moore's Law?
I am not an expert in this field, but I agree with you from the consumer perspective. E.g., the new macbook airs don't seem radically faster than my macbook from a few years before, it is just thinner. It seems like Moore's law is dead/dying and instead being replaced by Koomey's law, as others have noted
GPUs are much faster, and we have things like Quick Sync. And of course FPGA densities are skyrocketing still. But the integrated memory controller and hyperthreading of Nehalem was the last big jump in software encoding. Until 8 core goes mainstream, that's unlikely to change much.
(Note that you can buy 6 core Sandy Bridge E workstations, but you could do that with Westmere on 1366, too)
I recall the days where I'd start mp3-encoding a freshly-ripped CD and find something to do for an hour or two. I recall the days where I'd render the demo files for PoV-Ray at the highest quality at my PC's max VGA resolution (800x600) and then go to sleep, wake up in the morning to check the progress (half-way done -- w00t!), then come home from work to see one of Dan Farmer's eerie creations completely rendered.
I have no problem with processes that are so computationally-intensive that they are rendered impractical. Hardware will either catch up and it will be adopted, or it won't.
The proof is in the pudding, though. How does the H.265 encoded stuff look?
Perhaps the new spec is aiming to benefit huge media companies that can afford vast runs on a GPU cloud to encode, thus preventing us little folks (or the warez scene) from taking advantage of halving the bandwidth/space requirements (or doubling the video quality).
13 comments
[ 4.6 ms ] story [ 34.7 ms ] threadComputers have gotten much faster since May 2010?
Really?
I've wondered this for a while.... from my (idiotic consumer) perspective it seems like we are moving at a snail's pace. If I buy an iMac today, it comes with the exactly same processor as it did 13 months ago. Almost exactly the same as a year before that (all had quad cores at various configurations). And it's not really any different browsing custom PC vendors if I want to keep prices reasonable.
Shouldn't I be able to get a quad-core 7ghz for $2000 by now according to the corollaries of Moore's Law?
http://www.economicsofinformation.com/2011/09/is-koomeys-law...
Sandy Bridge was barely any improvement at video encoding over Nehalem, clock for clock (the first i7):
http://www.anandtech.com/show/4083/the-sandy-bridge-review-i...
And Ivy Bridge is at most 10%:
http://www.anandtech.com/show/5626/ivy-bridge-preview-core-i...
GPUs are much faster, and we have things like Quick Sync. And of course FPGA densities are skyrocketing still. But the integrated memory controller and hyperthreading of Nehalem was the last big jump in software encoding. Until 8 core goes mainstream, that's unlikely to change much.
(Note that you can buy 6 core Sandy Bridge E workstations, but you could do that with Westmere on 1366, too)
I think the article's next line has it right: the tested algorithm might have been for a future computer from space.
1. At the reported speeds, it would take 3 years on a 100 machine cluster to encode a 1.5h movie like this.
2. It's 10x better today (fact?). So you'll only need 4 months on that cluster.
I have no problem with processes that are so computationally-intensive that they are rendered impractical. Hardware will either catch up and it will be adopted, or it won't.
The proof is in the pudding, though. How does the H.265 encoded stuff look?
Perhaps the new spec is aiming to benefit huge media companies that can afford vast runs on a GPU cloud to encode, thus preventing us little folks (or the warez scene) from taking advantage of halving the bandwidth/space requirements (or doubling the video quality).