as a programmer I have some idea that translating CPU rendered algorithms to CUDA/Shader code is going to alter the look of the effects to some extent.
I haven't been a programmer in a looooong time, but as a current active video producer, I can say that offloading the rendering to the GPU should have NO effect on any effect other than the speed at which it renders. If any other effect is created, something is wrong. (And when I say "should" here I don't mean according to hardware and software theory, I mean that's the way we producers NEED it to work. NO difference whatsoever other than speed of rendering.)
I'm wondering if this is a GTX 470 quirk? (fingers crossed it's not
Since that's an unsupported card, I would actually hope it IS a quirk, as the alternative is that ALL cards work improperly. That would not be good.
I'm a video maker AND a programmer...so on reflection I guess I was being a little bit timid, and too nice - first time post and all. So I agree:
offloading the rendering to the GPU should have NO effect on any effect other than the speed at which it renders.
Let's hope some people with other cards try comparing effects...
Trying your test with a 480, gives the following results:
1. MPE on:
2. MPE off:
There is a clear difference and that shows the advantage of using MPE. This is not a bug or deficiency, but the difference between MPE on or off.
What happens is that with MPE on, everything is handled with maximum quality, something that is not done when MPE is off, in order to get at least reasonable performance for those without a CUDA card.
The noise effect is not displaying correct with MPE on
I think it has to do with the Alpha Channel.
I filed a bugreport for someting simular (GTX285).
with MPE on, everything is handled with maximum quality
Is it possible to use Maximum Quality with MPE set to software? If so, the results should be identical, with the speed at which it renders being the only difference.
I have now tried a series of tests to ascertain where and how these effects are rendering differently on my system. To eliminate differences produced by Preview quality differences, I am only looking at the output of Exported movies with "Render at Maximum Quality" checked in the export window.
The expectation I (and I think others) have is that files rendered in this way should be identical if rendered with or without the Mercury Playback System enabled on the GPU. Obviously I am not using a certified card - and that is the main point of my post: to ascertain whether users of certified (or at least diferently uncertified;-) hardware are getting the same or different results.
To do the test - and to allow others to easily test the results on their systems - I have created a PP5 project that uses a series of color bars with individual effects applied to each of the bars, and then a set of color bars with 3 effects layered up in one order, and then in the reverse order - more about the need for this in a sec! - for now props to PPBM4 for the idea of using color bars.
In total I have 7 color bars with:
1. No effects - just to check that there is no color variation without effects applied
2. Noise Filter only - with "Use Color Noise" unchecked (ie., black and white noise)
3. Noise Filter only - with "Use Color Noise" checked (ie.,color noise)
4. Tint only
5. Fast Color Corrector only
6. FCC, Tint, Noise
7. Noise, Tint, FCC
The PP5 project file is available here: http://www.mediafire.com/file/0jxw2zgecmt
I should probably have done this with more effects and combinations, but for now this test is proving to me that, at least with a GTX 470, something - indeed several things - are going wrong.
* Firstly, the Noise filter is rendering (and indeed Previewing) using only black and white noise with MPE/GPU - it's a different consistency too - but it's the absence of color that's is really changing the effect;
* Both Tint and FCC are producing slightly different color values across all the bars, usually between 1-4 across all RGB values (checked in Photoshop);
* Finally, MPE/GPU is reversing the order the filters are being rendered!!!! This is the main culprit producing the extremely different colors and appearance in the test image/preview renders in my orginal post.
I've uploaded a couple of stills from the tests below (these are taken from the Rendered output files) - obviously there is little point in uploading the FCC and Tint tests as the forums' resizing/compression makes seeing the difference nigh impossible, but if you'd like to take a look at my A-B comparisons I've put a zip with them all here (1 thru 7 as per above as PNGs): http://www.mediafire.com/?2ndywoynoyj
Software render is on the Left. MPE/GPU is on the right.
This one is Noise with "Use Color Noise" checked (it's hard to see, but the MPE/GPU Noise is coming out monochrome):
This one is FCC, Tint & Noise:
And this one is with those effects in reverse order (ie., Noise, Tint, FCC):
Anyone want to have a go???
Realised the MediaFire link for the PP5 project adds a .xml suffix to the file name no good reason. In case that's why no one has tried it out, here's a zipped version: http://www.mediafire.com/?wemng2znz0g
Sort of thought people would be worried about effects not rendering accurately, and effect layers getting reversed when switching between hardware and software MPE...
Obviously worried someone from Adobe enough for them to PM me for a copy of the file...but not enough for them to get back to me with their results...
In the interests of science lets pull together and narrow it down to whether it's a GTX 470 problem, or goes wider...
I tried your test and uploaded it to youtube at this link: http://www.youtube.com/watch?v=bghMD12zfgQ
System: Intel i7 X980 @3.33 Gh ASUS Geforce GTX480 PCI-E 2.0 1536 MB DDR5 Graphics Card with hack.
Exported to a Microsoft AVI file using "Maximum Render Quality". Uploaded the AVI file to YouTube and let YouTube do the streaming compression.
1 person found this helpful
I can't confirm any other cards but it happens the same on my machine with a 470 card when I used your project. FWIW.
I couldn't watch the youtube vid. Error.
@Rowby: Thanks for doing the test on a GTX 480...could you confirm a couple of things?
1) Is the video you posted done with MPE enabled on the GPU? (I'm assuming it was)...and
2) Did you try it with MPE switched to software mode? If so would you mind uploading the video, or alternatively just letting us know if you see the reversal of effect layers problem, noise issue, etc ?
Just wanting to confirm that you are seeing the same problems we are on GTX 470 hardware...
@Thrill Media: Thanks for confirming you see the same problem on your GTX 470 - although it's not exactly making me happy it's good to know it's consistent across this particular hardware setup...
Anyone with a GTX 285 or Quadro card able to confirm if this is happening on certified cards?
Sorry to hassle, but was just wondering if you could let me know if the test you did on the GTX 480 confirmed the results we are seeing on the GTX 470...?
Bit hard to tell from a single render - needs to be done with MPE set to GPU and to Software...perhaps I am missing something?
PS: I've installed the latest drivers from Nvidia (v: 258.96 r: July 12, 2010) - still have the same problem.