Can Lucidlogix Right Sandy Bridge’s Wrongs? Virtu, Previewed

[Supertrek32]

From what I've read on the doom9 forums - where the developers of many codecs and containers hang out - Quick Sync isn't anything revolutionary in terms of encoding.

With settings for a comparable output in terms of quality, x264 is just as fast as intels Quick Sync. There's also the added bonus of being able to fine-tune that quality up or down and all the other little tweaks which you simply can't do on intel's mostly-closed platform. Sandy Bridge is only a revolution in the encoding department in the fact that it's the fastest CPU.

The only area Quick Sync shines in encoding is on a laptop, since it means less power used and a longer battery. But who does serious encoding on a laptop anyway?

 

[sblantipodi]

Because the discrete GPU can't do what Quick Sync does

thanor the reply cangelini but what a discrete GPU + AVX from CPU can't do and quick sync does?

 

[boletus]

[citation][nom]lradunovic77[/nom]Again Intel and AMD move to integrate graphic chip into CPU is good for mobile useless for anything else. It is far from being smart solution for desktops unless they can pack GTX580 capable card into....mmmm i don't think so.[/citation]

I gave up on using that long ago. The best solution is to keep the front page open on a separate tab or or window and use it like a table of contents.

 

[boletus]

Oops, meant to quote hp79's comment about the drop down menu.

 

[cangelini]

[citation][nom]sblantipodi[/nom]thanor the reply cangelini but what a discrete GPU + AVX from CPU can't do and quick sync does?[/citation]

AVX? Not even in use yet 😛

 

[palladin9479]

Umm the only thing demonstrated here is that if your app is optimized for something then it runs better on that thing. "QuickSync" is just hardwired encoding into the CPU rather then using software to feed the algorithms to the CPU. Hardwireing math heavy algorithms should gain a performance increase of at least an order magnitudes, typically an increase of 10x if the entire operation is being done inside memory, half that if your factor in outside I/O and other bottlenecks. SB actually fails pretty badly here, of course its faster then any software implemented methods but we should of seen a much better boost then what was presented.

If you want to see what the performance difference should be, get a Via Nano platform going on with a Linux distro and properly compiled kernel (with Padlock included) and OpenSSL. Run the OpenSSL benchmarking test using dynamic engine (software encryption) with AES-256 and 2048 bit encryption key, then run the same benchmark again using padlock engine (built in CPU encryption) and watch the output skyrocket. It allows real time encrypted file system access without any performance penalty on a low-power 1.6Ghz CPU. Most likely the current implementation has a few issues that they'll fix on the next release "QuickSync 2" or something.

 

[tpi2007]

Chris!

Your commenting system is now completely broken!

When you look at the comments from the article, the first one that shows up is

"^^-*edit while your desktop is showing through the intel card*",

but when you look at it, the first and second page of comments are the same. If you look at the comments through the forums, there are many comments before that one.

One more thing: where is my comment ? It was the first on the article when I wrote it yesterday. Where is it ? Nowhere!


Why is it also taking forever for the comment rating system to be fixed ? It has been months now. What could be so hard to fix, when it actually worked before ? This is slowly destroying the community.

 

[cangelini]

tpi,

Believe me, everyone here in the US is anxious to see our team in France fix the community-related bugs. They were reported a loooong time ago, and finally today I see that they've been assigned to techs. Please expect a fix in the very near future!

Chris

 

[sblantipodi]

AVX? Not even in use yet 😛

this isn't true, there is many software that uses AVX since the release of SP1 for windows 7, only three names. 😉

Arcsoft media converter 7
Cyberlink Media espresso 6.5
Badaboom 2.0

now you can answer my questions, what quick sync can do that AVX+CUDA can't do?

 

[cangelini]

1) Please feel free to share links to release notes with information about current apps optimized for AVX.

2) Once I'm able to test Quick Sync against an app optimized for both CUDA and AVX instructions, I'll be able to more definitively tell you whether performance is any better and if the CUDA-based output still looks like crap

 

[sblantipodi]

1) Please feel free to share links to release notes with information about current apps optimized for AVX.

2) Once I'm able to test Quick Sync against an app optimized for both CUDA and AVX instructions, I'll be able to more definitively tell you whether performance is any better and if the CUDA-based output still looks like crap

intel used both Cyberlink mediaespresso and Arcosft mediaconverter to show the AVX improvements in their events on sandy bridge so simply use google, for badaboom http://badaboomit.com/node/541

Probably what you don't understood is that Quick Sync isn't anything more than AVX + Intel GPU, so you can do better with AVX + CUDA running a discrete graphics.

Feel free to write a review with a software that supports AVX like the one I saied and you will understand that I'm right 😉

 

[cangelini]

I'm sorry, but I was at IDF and saw the demos. Right before Sandy Bridge launched, I asked Intel for examples of AVX-optimized software, and there were none. The link you provided has no mention of AVX. And I very regularly test with both MediaEspresso and MediaConverter--neither employs AVX as far as either Intel, CyberLink, or Arcsoft has told me. If you can show me proof to the contrary, I'd be happy to investigate!

There is a big difference between AVX, an extension to x86, and Quick Sync, a fixed-function block consisting of a multi-function codec for high-performance decoding, video scalar, a denoise filter, and color processing. Yes, the execution units are used in the encode stage. However, I have received no indication that this fixed-function logic is related to AVX.

 

[sblantipodi]

I'm sorry, but I was at IDF and saw the demos. Right before Sandy Bridge launched, I asked Intel for examples of AVX-optimized software, and there were none. The link you provided has no mention of AVX. And I very regularly test with both MediaEspresso and MediaConverter--neither employs AVX as far as either Intel, CyberLink, or Arcsoft has told me. If you can show me proof to the contrary, I'd be happy to investigate!

There is a big difference between AVX, an extension to x86, and Quick Sync, a fixed-function block consisting of a multi-function codec for high-performance decoding, video scalar, a denoise filter, and color processing. Yes, the execution units are used in the encode stage. However, I have received no indication that this fixed-function logic is related to AVX.

you have received no indication but it can't be different
there is no difference, quick sync = AVX+integrated graphics

time will give you your answer no need to argue anymore since there is no specs that confirm what I'm saying,
for now you are right, tomorrow we will talk again 😉

 

[cangelini]

Because I'm never satisfied with "trust me, I'm right and you're wrong," I went to Intel for more clarification This comes straight from Hong Jiang, the lead architect on Quick Sync.

"Quick Synch Video is not using AVX. It is using the high quality, high performance fixed function codec pipelines built into all 2nd Gen Core processors. AVX is a different technology that ISVs can also use and there could be a workflow where a video might be processed using SW algorithms that benefit from AVX, and then encoded for final archival or sharing using the Quick Sync Video. But the two underlying features of the processor are independent."

Hope that helps!!
Best,
Chris

 

[sblantipodi]

Because I'm never satisfied with "trust me, I'm right and you're wrong," I went to Intel for more clarification This comes straight from Hong Jiang, the lead architect on Quick Sync.

"Quick Synch Video is not using AVX. It is using the high quality, high performance fixed function codec pipelines built into all 2nd Gen Core processors. AVX is a different technology that ISVs can also use and there could be a workflow where a video might be processed using SW algorithms that benefit from AVX, and then encoded for final archival or sharing using the Quick Sync Video. But the two underlying features of the processor are independent."

Hope that helps!!
Best,
Chris

it helped, I know that making people like you angry is the only way to know more 😀
when will you release a comparison using Quick Sync and sandy bridge + Cuda?

as far as I understood Quick Sync is deactivated when a discrete graphics is installed in the system, if this is true, what
is the sense of quick sync? Who don't have a discrete today?

 

[cangelini]

I don't get mad; I just look for the truth ;-)

We actually have something like this already--it's very in-depth! Check it out and let me know what you think. The problem with CUDA is that its quality is noticeably very bad:

http://www.tomshardware.com/reviews/video-transcoding-amd-app-nvidia-cuda-intel-quicksync,2839.html

 

[sblantipodi]

I don't get mad; I just look for the truth ;-)

We actually have something like this already--it's very in-depth! Check it out and let me know what you think. The problem with CUDA is that its quality is noticeably very bad:

http://www.tomshardware.com/reviews/video-transcoding-amd-app-nvidia-cuda-intel-quicksync,2839.html

wow, great article it is what I'm searching for, I will read it tomorrow.
Anyway I can't understand who will use quick sync if quick sync is disabled when discrete graphics is installed in the system.
Do you know if it is true that quick sync is disabled in this case?

Thanks Cangelini.

 

[cangelini]

It is true. This is a limitation that I don't like. And that's exactly why Lucidlogix's Virtu software is so interesting--it lets you use Quick Sync while discrete graphics are installed.

 

[uther100]

I don't understand how this is helping things? The software can successfully copy the frame buffer from a discrete graphics card to an integrated one, great but what does that accomplish?

If they get it working so that it lets you shut off discrete graphics in a laptop when not needed that is great. I don't see where any of this helps with encoding though?

Is it saying that the gpu can encode faster when it does not have to display a screen so you can use the integrated graphics for your display and fully commit a discrete gpu to the encode process?

Someone break it down for me.

 

[sblantipodi]

It is true. This is a limitation that I don't like. And that's exactly why Lucidlogix's Virtu software is so interesting--it lets you use Quick Sync while discrete graphics are installed.

I really can't understand the sense of Quick Sync if it can't be used if a discrete is installed in the system.
Is this a new features? Is this an interesting features? For who? Do you know if Z68 have this limit also?

 

[Guest]

@sblantipodi -- why don't you try doing some READING and RESEARCH on the 'net instead of burying this forum with questions whose answers are out there?

(serve him right if Chris answers "Do you know if Z68 have this limit also? " with "yes, I do know." rather than answering the question sblanti evidently meant to ask.)

 

[Guest]

@sblantipodi "Why care about quick sync when you have a discrete GPU?"

Because the new HD3000 graphics are MUCH faster when transcoding video. If you don't work with video, then no, there is nothing here that will be of interest to you.

 

[dogman-x]

Would this work with Adobe CS5?

 

[cangelini]

[citation][nom]dotScience[/nom]Hi, for those of us who are waiting to build our next system around an SB 2x00K CPU and who:1. do not anticipate doing any video transcoding2. are not gamers, not doing anything with 3d modeling/rendering, but who do work with very large files in PhotoShop, and do apply filters to layers that can take a long time, sometimes.May we respectfully ask your opinions on, very rough estimates on:1. if you are using an SB 2x00 K chip, and NOT making use of discrete built-in transcoding, is power consumption of the CPU, and heat output, less: under either use at rated speed, or use where the SB CPU is over-clocked ? I would assume this transcoding block unit cannot be "disabled" by some future motherboard's bios.2. any guesses as to when commercial availability of Z68 chipset motherboards are going to available to mere mortals ?3. assuming your primary need for heavy-duty graphics processing while using this to-be-built SB 2x00 K CPU system is PhotoShop (but, not involving PS's 3d facilities): would you assume you would need to use a video card, and you would get more benefit from one that supported DirectX 11 vs. DirectX < 11 ?Excellent article, and always appreciate your comments, Chris A.: you wield Occam's Razor sharply thanks ! Bill[/citation]

Bill,
I don't often get back to the comments of older stories. Please feel free to start up a conversation any time on Twitter--I'll definitely see it there!

The block cannot be disabled, no.

Sounds like Z68 is planned for May 8th-14th--somewhere in there.

If you're looking at Photoshop for a primary use case, go for a more powerful CPU and pay less attention to graphics, IMO.

Hope that helps a bit!
Take care,
Chris

 

[vania007]

I was seriously considering an upgrade of my gaming PC, now I will wait until Intel can somehow make Quicksync available for everyone with a discreet graphics card (perhaps make it a part of the south bridge?), am sorry but Lucidlogix just doesn't seem like a sacrifice I might be willing to make, I rather lose Quicksync than FPS.