27 Replies Latest reply on May 4, 2010 9:39 PM by shooternz

    MPE hardware GPU vs software test

    DVDmike Level 1

      I've got the 285 card, so there is no need to hack into anything.  But on my system, MPE GPU slows things down.

       

      Here is what I did, I took one 720p D90 (Nikon Dslr) clip and put it into a new sequence.  The clip plays back fine with GPU enabled

       

      I add the same clip on video track 2 and move it over a few seconds and scale it down to 40% of its orginal size (like PIP).

       

      At this point, I still have no red, yellow or green lines in my sequence.  I play back the sequence and once it gets into the PIP portion, the sequence pays back badly.  I am getting maybe every 8th frame or something like that.  It is very jittery and barely looks like video at all.  The task manager reports virtually no CPU being used.

       

      Next, I change the settings to software only and delete previews.  I play back the sequence and it plays back virtually fine, even during the PIP section (which is redlined).  Again, minimal CPU being used.

       

      Is there some setting that I need to adjust to get MPE GPU working correctly?  Is this an issue with retrieving data from disk?

        • 1. Re: MPE hardware GPU vs software test
          Jeff Bellune Adobe Community Professional

          Driver version?

           

          -Jeff

          • 2. Re: MPE hardware GPU vs software test
            DVDmike Level 1

            I downloaded the latest driver from Nvidia last week.

             

            driver.JPG

            • 3. Re: MPE hardware GPU vs software test
              Harm Millaard Level 7

              Limiting factors in system performance have always been CPU, memory and disk I/O. The weakest link determined overall performance. Now that GPU performance is also a major factor, it is logical to assume that the impact of the weakest link will become more distinct. Often this will be the disk setup.

               

              It is telling that the top performing machines in the http://ppbm4.com benchmark all have (massive) raid arrays and slower performers with the same CPU and the same memory but with more limited disk setups are further down the list.

               

              If the GPU speeds up rendering and encoding, disk setup will become a bottleneck even earlier than in the past.

              1 person found this helpful
              • 4. Re: MPE hardware GPU vs software test
                DVDmike Level 1

                In my case, if I/O was why my performance is not as expected, why does it work better in software-only mode?  If my I/O is fast enough for software only mode, why wouldn't it be fast enough for GPU processing with a supported card?  I do not recall Adobe stating that we needed raid systems instead of single 7200 SATA drives for smooth preview playback.  If fact, I even saw someone do a web demo on a laptop with 5400RPM drives with 3 or 4 1080p and even red footage that had smoother playback than I am experiencing.

                • 5. Re: MPE hardware GPU vs software test
                  Stan Jones Adobe Community Professional & MVP

                  I can't find the perfect thread to add this, and it is relevant here:

                   

                  Bill G and Harm have set up a great benchmark option to avoid guessing about problems - now with more telling use of video clips.  I believe that anytime you are experiencing problems that are potentially hardware related, you should run the benchmark and use that as data to sort it out.  Kudos again to Bill and Harm.

                  • 6. Re: MPE hardware GPU vs software test
                    Harm Millaard Level 7

                    One reason that can have a major impact is that software MPE is using the well known encoding or rendering standards, but if you enable hardware MPE, maximum render quality is always used, and I believe also maximum bit depth.

                    • 7. Re: MPE hardware GPU vs software test
                      Harm Millaard Level 7

                      Stan,

                       

                      Bill and I are working on a PPBM5 benchmark with actual footage from various sources. Our dilemma is to keep the download file small, the testing time not too long, and to show, next to the current PPBM4 results, the advantages of MPE. Currently I'm testing with AVCHD, HDV, XDCAM-EX, Canon 7D and some Red material to get a nice mix of what one can encounter in real life.

                       

                      I don't know if you alluded to that.

                      • 8. Re: MPE hardware GPU vs software test
                        cts51911 Level 1

                        Harm,

                         

                        I'm trying to build a cost effect raid 0 setup and wondered if you can give some suggestions.

                         

                        1)  Brand/Model of SATA 7,200 rpm drives?

                         

                        2)  Onboard X58 Raid good enough?  (2x drives)

                         

                        Wild Fly   4.2ghz ranks pretty high with a 4x 500 Raid 0 on your database at PPBM4....No mention of a 3rd party raid controller in the notes.  What was used?

                         

                        Thanks, Chris

                        • 9. Re: MPE hardware GPU vs software test
                          Harm Millaard Level 7

                          Chris,

                           

                          1. Samsung Spinpoint F3, WD Caviar Black RE3, Seagate 7200.12

                           

                          2. That is what I use for my 2 disk array. The Marvell one.

                           

                          Wild Fly did not mention his raid controller, so it is safe to assume he used the on-board controller.

                          • 10. Re: MPE hardware GPU vs software test
                            cts51911 Level 1

                            Thanks Harm!

                             

                            I'm looking forward to your PPMB5.

                            • 11. Re: MPE hardware GPU vs software test
                              Harm Millaard Level 7

                              Chris,

                               

                              One correction:

                               

                              Bill is the Founding Father, Bill hosts it and I just try to help him. So it definitely is not My PPBM5 test. It is Bill's test with some help from me and others who have willingly made their source material available for testing. And I am grateful for their contributions.

                              • 12. Re: MPE hardware GPU vs software test
                                Stan Jones Adobe Community Professional & MVP

                                Bill and I are working on a PPBM5 benchmark with actual footage from various sources.  ....  I don't know if you alluded to that.

                                Yes, I'm a bit ahead of your actual status.  So hurry up!

                                 

                                I just strongly believe in the methodology, both for benchmarking and for troubleshooting.

                                • 13. Re: MPE hardware GPU vs software test
                                  Wil Renczes Adobe Employee

                                  If no yellow is showing up over the timebar, something's not right.  Can you confirm that the MPE acceleration option is on & sticking (ie, when you reopen the project settings, is it still on, or is it resetting itself back to software after you select it), and maybe a screen shot showing the timeline & the effects control panel with the top video layer selected (so we can see what filters you have applied)?

                                  • 14. Re: MPE hardware GPU vs software test
                                    digitlman Level 1

                                    Maybe you could try a different video clip for your test? I'm sure not every codec/format out there is accelerated with cuda. i personally have never heard of or used a nikon dslr video file.  Maybe try editing some other format and see what happens. try an avi, or mpg, or h264 or mxf file.

                                    • 15. Re: MPE hardware GPU vs software test
                                      DVDmike Level 1

                                      I can definitely confirm that that the settings are sticky.

                                       

                                      I will try to find some other non-DSLR HD clips and see what happens.  I should have some HDV clips on an old drive somewhere.  Hopefully, they are not cineform converted.  Back when I edited HDV, I was using cineform with CS3.  The machine in question here is new and has no cineform software/codecs installed.  So if the clips that I have saved were converted, then I won't be able to use them.  Other than that, I have tons of DV footage that I could try.  But I would expect it to work regardless.  I also could  render out HDV from Digital juice compositor's tool kits and other royalty free DJ products for testing.    But if anyone had any sample HD clips that were small enough that I could download, we could compare results with the same exact clips and settings.

                                       

                                      I will report back once I find some good clips to use...

                                       

                                      Thank you for your help.

                                      • 16. Re: MPE hardware GPU vs software test
                                        DVDmike Level 1

                                        I created some 1080p footage from Digital Juice

                                        in .avi format.  The uncompressed 1080p files are like 2GB each.  So even a single

                                        clip does not play back at 1/4 resolution.

                                         

                                        I used PPro CS5 to encode them into DVCPro 100 HD and brought them back into CS5.  I have had a difficult time getting the playback to be anything but smooth.  I am up to 5 layers and effects of some sort on 4 of them.  Here, I can really see the difference between software and GPU modes.  The GPU mode tops out at about 25% cpu in the most intensive areas.  The software mode gets up to 90% CPU in some spots.

                                         

                                        So what we have learned here is that the Nikon DSLR footage does not play back well with MPE.  And that DVCPro is everything that is advertised.

                                         

                                        Are there any native editing modes that are 1920x1080 with pixel aspect of 1.0 that I can try as well?

                                        • 17. Re: MPE hardware GPU vs software test
                                          shooternz Level 6

                                          I will run a test with some canon DSLR clips I have.

                                           

                                          I never had any issue with it in CS4 so lets find out about it in  CS5

                                           

                                           

                                          OK.  Done

                                           

                                          I dumped some Canon DSLR 1920 x 1080 (30fps) into my 720P  25fps project (Scaled to frame size on ) and it runs perfectly with out a Red Render bar.

                                           

                                          There ya go.

                                           

                                          Just added another layer of it and applied Opacity > Screen Blend.

                                           

                                          Still perfect.

                                          • 18. Re: MPE hardware GPU vs software test
                                            Wil Renczes Adobe Employee

                                            Ah, I think I have a theory.  What's the actual dimensions of the source footage? (I haven't played yet with the Nikons)   And, what sequence are you putting them in?  Lastly, are you applying 'Scale to Frame Size'?  (I think that's where it flips into software mode...)

                                            • 19. Re: MPE hardware GPU vs software test
                                              shooternz Level 6

                                              Who are you asking Wil ?

                                              • 20. Re: MPE hardware GPU vs software test
                                                DVDmike Level 1

                                                the Nikons are 720p.  But I created the comp by dragging the Nikon clip into the new comp (sequence) icon.  So the sequence matches the comp.

                                                • 21. Re: MPE hardware GPU vs software test
                                                  Wil Renczes Adobe Employee

                                                  Okay, thanks.  So much for that idea.  I'll ask around, I think some folks on the QE team have been testing out the Nikons - I'll see what I can find out.

                                                  • 22. Re: MPE hardware GPU vs software test
                                                    DVDmike Level 1

                                                    Tomorrow, I will create some new clips from the D90 just to make sure that it is not something with my other original clips.

                                                     

                                                    thanks for you rhelp

                                                    • 23. Re: MPE hardware GPU vs software test
                                                      DVDmike Level 1

                                                      Note that some portions of my 5 layer sequence that works well (DVCPRO HD) is redlined.  But the redline plays back smoothly.

                                                       

                                                      The Nikon clips that do not play back are not even yellow lined and do not play back smoothly.

                                                       

                                                      Since my dvcproHD clips played back fine, I decided to go back to the Nikon DSLR project and try it again.  Low and behold it looks fine now.  In fact, I can add more video tracks and my dvcpro footage and it is still OK.

                                                       

                                                      Since I had problems, I have not made any system changes, any disk changes, nothing other than installing Quicktime and the DJ Juicer program.  The project was the same too.  I did not make any preferences changes in PPro either.  So I have no explanation for the difference in results from one day to the other.

                                                       

                                                      I will try it again tomorrow and see what happens......

                                                      • 25. Re: MPE hardware GPU vs software test
                                                        DVDmike Level 1

                                                        You think that installing QT might have helped my CS5 performance?  I am not using a QT files in any of my test projects.  But I will need to in

                                                        the future.

                                                        • 26. Re: MPE hardware GPU vs software test
                                                          the_wine_snob Level 9

                                                          If you feel the need for QT, explore the versions. For me, the last known good was 7.5.5. I had been led to believe that 7.6.6 was good too, after some anti-Adobe versions in the 7.6.x series, but recent reports seem to indicate that it does not work with Adobe. You will need to check this out for yourself and choose the version accordingly. I would steer away from any 7.4.x versions, as many ruined all things Adobe.

                                                           

                                                          Good luck,

                                                           

                                                          Hunt

                                                          • 27. Re: MPE hardware GPU vs software test
                                                            shooternz Level 6

                                                            You think that installing QT might have helped my CS5 performance?  I am not using a QT files in any of my test projects.  But I will need to in

                                                            the future.

                                                             

                                                            Remove it and see what happens.

                                                             

                                                            FWIW : I am working with QTPro  7.6.6

                                                             

                                                            ProRes is a great format as an intermediate from Mac platforms should you have the need like I do.  I also use a lot of .movs in my reverse workflow.