DivX 6.6 with SSE4-Instructions ?

[crnagora]

Hi folks,

on the IDF: Intel 3,33 GHz „Penryn“ (Core2Duo Successor) presentation they showed some benchmarks made with the newest DivX 6.6 Alpha:

http://img444.imageshack.us/img444/338/34215377rp6.th.jpg (http://img444.imageshack.us/img444/338/34215377rp6.jpg)http://img295.imageshack.us/img295/992/penrynidfpressbrief0026kc0.th.jpg (http://img295.imageshack.us/img295/992/penrynidfpressbrief0026kc0.jpg)

Looks impressive (Even if 38fps -> 18fps isn't 211% faster ;) :) ). Hope there is some Quality gain too with the new DivX 6.6 Codec.

Sadly it seems Digital56k isn't posting here anymore. Would love to hear some "inside" stories ;).

If some of the DivX guys is around: I'm willing to beta-test ;).

[rakf1]

I could only find Divx 6.6 beta for Mac on Divx.com, does anyone know where I can find Divx 6.6 codec for Windows?

[DigitAl56K]

Hi folks,
Sadly it seems Digital56k isn't posting here anymore. Would love to hear some "inside" stories ;).


I hate that DigitAl56K guy! ;)

Actually inside stories are a little scarce right now because we have had only a short time to investigate the new instructions so far. You can imagine that getting time to test code on pre-production hardware is difficult!

To explore performance we compared a windowed full search in SSE4 (obviously only available on the Penryn) to the SSE2 equivalent (i.e. no cheating!). The SSE4 code was clearly faster, and the Penryn chip itself is just a screamer regardless.

I've seen different websites quote different numbers depending on the two platforms being compared (i.e. code executing on Penryn vs. Penryn with and without SSE4 or Penryn vs QX6800). Keep in mind that DivX 6.5.1 doesn't perform full search, so comparisons are between 6.6 alpha versions performing the windowed full search and not 6.5.1 to 6.6 alpha generally.

DivX 6.6 for Windows isn't yet available from our website.

[Awatef]

I hate those people that give false information! 211%! They think people are dumb! "Oh! We forgot to divide by 2! We're very sorry!"
Another reason to ignore Intel...

[Romario]

DigitAl56K, can you tell us when DivX 6.6 will be released, in October or little earlier?

[DigitAl56K]

I hate those people that give false information! 211%! They think people are dumb! "Oh! We forgot to divide by 2! We're very sorry!"

They didn't need to divide by 2, they just confused "x% faster" with "x% performance". They showed the data too though, so it was reasonably clear what they meant.

@Romario: We will have a release out much sooner than that :)

[Schultz]

Looks impressive (Even if 38fps -> 18fps isn't 211% faster ;) :) ). Hope there is some Quality gain too with the new DivX 6.6 Codec.


Look at the picture again the DivX was showed seconds NOT fps.
So it took the new one 18secs to complete instead of 38secs. Which is 2x the performance. Since over twice as fast. 38/18 = 2.11 so they math is correct and performance numbers are correct.

[Romario]

DigitAl56K, how close you know SSE4 instructions, can your team implement these fantastic instructions well?

Are you sure that DivX 6.6 will be SSE4 optimised?

[crnagora]

Look at the picture again the DivX was showed seconds NOT fps.
So it took the new one 18secs to complete instead of 38secs. Which is 2x the performance. Since over twice as fast. 38/18 = 2.11 so they math is correct and performance numbers are correct.

You are right about the fps thing. But not about the math :). At least not where I live. Doubled Performance = 100% faster. 200% faster would be 4 times faster. So the math isn't correct. :stupid: :p

I hate that DigitAl56K guy! ;)


Cheers the Red-Head is alive. Glad to see you posting again.

[DigitAl56K]

DigitAl56K, how close you know SSE4 instructions, can your team implement these fantastic instructions well?

We are currently researching the best ways to use the new instructions - we saw them for the first time only very recently.

Are you sure that DivX 6.6 will be SSE4 optimised?

I think what we might do is roll out 6.6 very soon, including the windowed full search that Intel demonstrated (i.e. in SSE2 and SSE4 variations). Those lucky enough to get their hands on these chips could then verify the numbers. More involved optimizations of the codec would follow later - as I say we're already looking into it.

Glad to see you posting again.

Thanks! I'm always around somewhere, even if it's just to lurk. Who could miss the AACS debacle? ;)

[Romario]

My advice to you, DigitAl56K, is not to rush into DivX 6.6 release. There is NO NEED for rush.

P.S. Ideal scenario: DivX team somewhere in mid-summer release DivX 6.6 in both version, DivX 6.6 SSE4 only and DivX 6.6 SSE2 only. I hope that you understood me.

[ARDA]

If anyone wants to read about sse4

ftp://download.intel.com/technology/architecture/new-instructions-paper.pdf

[Schultz]

You are right about the fps thing. But not about the math :). At least not where I live. Doubled Performance = 100% faster. 200% faster would be 4 times faster. So the math isn't correct. :stupid: :p


The math is correct the wording they used it in isn't. Its the same point if you are running at 100% your speed. And you increased your speed by 50% you are now running at 150% your normal speed.

[crnagora]

The math is correct the wording they used it in isn't. Its the same point if you are running at 100% your speed. And you increased your speed by 50% you are now running at 150% your normal speed.

The whole point was that they made a mistake there: "211% Faster" isn't right. The right term would be 110% faster than the sse4-less cpu (As 38s would be 100%).

But anyways...lets stop this math-mumbo-jumbo (I'm getting woozy :stupid: :p ).

@DigitAl56K
Is DivX 6.6 only about speed or can we expect some quality gain too ? Oh one thing about DivX 6.5.1: Is "Enhance I-Blocks" and "Enhance Textures (6.4)" the same thing ?

[DewAsmara]

Hi Digital56K,
I hope new DivX 6.6 have abilities to convert DVD not only convert their AC3 into MP3 but also DTS into MP3. If possible I loved to attached the original AC3/DTS Sound with the DivX Video, because MP3 seems not too suficient compare with AC3. Right now temporary I convert DVD into AVI using MPEG4 standard codec and AC3 Sound, before I used DivX Converter but then quite dissapointed with MP3 Sound Quality, even for Video Codec DivX still much better than MPEG4 standard codec.

[video_magic]

Hello, I wondered if the download will have 'runtime CPU detection' , so it will work on any X86 chip and use the detected supported instructions - that would seem the best approach to me as a user.
I don't know whether this would work for 64-bit also or whether You would have to make a separate 64-bit download.

Anyway, it's always good news to have a new faster version :thanks: , the saved time is a very important factor to Me as a user , almost as much as quality results (which still is the priority reason for choosing anything)

[Isochroma]

Fudzilla: 2.33GHz Intel Wolfdale tested (http://www.fudzilla.com/index.php?option=com_content&task=view&id=2300&Itemid=1)

HKEPC has got their hand on a 45nm Wolfdale ES sample clocked at 2.33GHz with 6MB of L2 cache and a 1,333MHz FSB and they've put it through the paces and put it up against a Core 2 Duo E6550.

HKEPC has run a wide range of tests on the two CPUs and the Wolfdale is faster in every single test, although in some cases only by half a percent or less. But in some test, CineBench 9.5 for example, it's between 7.5 to 19.2 percent faster.

But the most impressive number is the alpha version of DivX 6.6 with SSE4 support, over 115 percent performance increase, down from 69 seconds to 32 seconds when encoding a 1080p MPEG2 file to MPEG4.

Looking at the game tests it seems to be about 10 percent faster on average, although for some reason it's more than 31 percent faster in Half-Life 2, up some 22 fps compared to the E6550.

It's also using a lot less power and running about 10-12 Centigrades cooler than the E6550. Even at full load the Wolfdale ES only draws 59W compared to 83W for the E6550.

You can check out more benchmark numbers here (http://www.hkepc.com/bbs/hwdb.php?tid=837360&tp=Intel-Penryn-Preview&rid=837360) and if you know Chinese, there's a whole article to go with the results.

[Wolfman]

Arrrghh.. I was just about to jump on the 6550 bandwagon and then something like this comes along... 100%+ increase on one of the main cpu workloads is just too good to passover. Have to stick with my xp 2600 for a while longer. Wolfman + Wolfdale 3.2 = happiness?

[_xxl]

Where to find that cpu?

[Manao]

100%+ increase on one of the main cpu workloads is just too good to passoverIt's just a trick however. As it was said elsewhere on this forum, the DivX 6.6 version was modded to include an unefficient & slow ME algorithm that could however greatly be sped up with SSE4. So SSE4 doesn't speed up DivX 6 by 115%. ( in other words, it's a commercial stunt )

A penryn in itself, however, will speed up a program in comparison to a conroe. And SSE4 is usefull for video encoding, but don't expect anything more than 5 to 10% speed up, in the best cases.

[DeathTheSheep]

the DivX 6.6 version was modded to include an unefficient & slow ME algorithm
You know the DivX code, or is this speculation, or what?

[Manao]

It's not speculation, I don't know DivX code, but, IIRC, Digital56K admitted as much here. I don't remember where I read that, but I'm pretty sure of it.

[Manao]

Here's the link : http://forum.doom9.org/showthread.php?t=125417

Look at the first post, the screen capture specifies : the experimental motion search can not be directly compared to the normal encoding modes with DivX codec versions 6.6 or priorWhich amounts to what I said : they implemented the algorithm best suited for SSE4 ( dumb exhaustive search ), only for this version, and only for benchmark purposes. They don't use that algorithm for real encodings for the same reason you don't use it for x264. It's overslow and doesn't bring any quality boost at all.

[DeathTheSheep]

Little OT, but the newer exhaustive search in x264 now means baseline compression faster than realtime now (and with subme7, single-threaded no less), at least for QVGA encodes. And at least in terms of SSIM, it always amounts to a higher score there, too, from my tests.

Not that I use it anyway; hence the OT designator. :)

[burfadel]

Too bad that the pixel shader support from Media Player Classic can't easily be implemented in to x264 to offload the motion estimation!