15 Replies Latest reply on May 20, 2010 5:33 AM by Ernst Lunding

    No effect of Mercury Engine, CS5

    Ernst Lunding

      When NVIDIA launched Elemental I bought it – but it turned out to be waist of money. Encoding was considerably slower with Elemental.

       

      I have now upgraded my Production Premium CS4 to CS5 – and encoding with the 64 bit CS5 is app. 40% faster than the 32 bit CS4 – this is what you would expect.

       

      But using the heavily promoted Mercury engine has absolutely no effect. I get the same export or encoding times whether I use the GPU Acceleration or the software only.

       

      Can anybody give an explanation?

       

      My PC is a Dell T7500, 64 bit Vista Ultimate, 48 GB RAM, 2 Intel Xeon W5580 with HT enabled (16 ‘processors’), 2 NVIDIA  Quadro FX 5800 (latest driver).

       

      Ernst,

      Denmark

        • 1. Re: No effect of Mercury Engine, CS5
          Jeff Bellune Adobe Community Professional

          Be aware that you get the Mercury engine goodness in CS5 whether you have software *or* hardware rendering.  The 40% increase you've seen is likely the result of the MPE and not just a switch to a 64-bit OS.

           

          The most likely explanation for why you don't see any improvement when GPU acceleration is used, is that your render is CPU-intensive.  That  is, any GPU-accelerated effects are taking less time to render than the  decoding/encoding time of the footage itself.  What format/codec is used for your source footage, and what format/codec is used for your export?

           

          -Jeff

          1 person found this helpful
          • 2. Re: No effect of Mercury Engine, CS5
            Ernst Lunding Level 1

            Thanks.

            Source footage is AVCHD (Sony) - export is H264 (Blue Ray)..

            Ernst

            • 3. Re: No effect of Mercury Engine, CS5
              Harm Millaard Level 7

              Ernst,

               

              With software MPE you get around the same encoding time as you get with hardware MPE. From the looks of it not much benefit, but under the hood you get significant gains, because MPE enabled encoding uses MRQ, which with software MPE would take around 4 times as long. So in the end result you get better quality in the same time.

               

              With rendering you get around a 4-fold performance increase in render times.

               

              AVCHD is very demanding on your CPU.

              1 person found this helpful
              • 4. Re: No effect of Mercury Engine, CS5
                SuperOperater Level 1

                Well, with those CPUs and ESPECIALLY WITH 48GB of RAM - if I understood corectly - I am really wondering if anything can go faster that it already works with such a "beast" of WorkStation!!!

                • 5. Re: No effect of Mercury Engine, CS5
                  Jeff Bellune Adobe Community Professional

                  Yeah.  Both of those codecs require more CPU time than almost any effect or combination of effects.

                   

                  -Jeff

                  • 6. Re: No effect of Mercury Engine, CS5
                    Ernst Lunding Level 1

                    Thanks to Harm and Jeff for answers.

                     

                    AVCHD is very demanding - this is why I bought this very expensive PC, which Dell called 'a monster'. Encoding or export time is basically not a problem (H264 encoding( CB35 Mbps, max quality) is carried out in real time, and blue ray MPEG2 (CB40 Mbps) much faster) , but I had expected to see a time difference between GPU acc. and software. This was the reason for my question.

                     

                    Elemental should with CS4 also facilitate GPU acceleration but with same PC I experienced a considerably slower encoding - less for Blue ray MPEG2 than for Blue ray H264. This was a surprise for me.

                     

                    Ernst

                    • 7. Re: No effect of Mercury Engine, CS5
                      Harm Millaard Level 7

                      Ernst,

                       

                      If you would run the PPBM4 Benchmark twice, once with MPE and once without MPE and submit the results to Bill and possibly to me by PM, we will all see what a monster you have, how your system compares to others and maybe where you have to look for further improvements. It works the same for both CS4 and CS5.

                      • 8. Re: No effect of Mercury Engine, CS5
                        Ernst Lunding Level 1

                        Harm,

                         

                        I have tried it some time ago, but the statistical script ( or whatever you call it) did not work and I could not figure out why. I will try again tomorrow and see if I more lucky this time.

                         

                        Best regards

                         

                        Ernst

                        • 9. Re: No effect of Mercury Engine, CS5
                          Harm Millaard Level 7

                          Ernst,

                           

                          If you have any problems, let me know. I admit it is not intuitive, but if you go to Explorer, move to the PPBM directory and use Copy address from the address bar:

                           

                          17-5-2010 15-19-28.jpg

                           

                          and then double click on the Statistics.vbs file to start it, then on the first question use CTRL-V and repeat that for the second question, you are almost done.

                          • 10. Re: No effect of Mercury Engine, CS5
                            Ernst Lunding Level 1

                            Harm,

                             

                            I have carried out the tests, but I cannot make the statistics.vbs work. I get the message (translated from Danish): Path not found, Code 800A004C. The path I have entered IS correct - I have also copied the address as you suggested but with same result. I have done 4 tests.

                            I tried the test on my old editing PC (CS4) and her I had no problems. This is very peculiar. Any explanations?

                             

                            The only results that I can give at present is therefore the elapsed queue time which with MPE GPU acc. was 4:36 and 4:36 (2 tests), without GPU acc. 4:09 and 4:15 (2 tests). Does that mean anything to you?

                             

                            Ernst

                            • 11. Re: No effect of Mercury Engine, CS5
                              Curt Wrigley Level 4

                              The GPU HW portion of MPE does not help with previewing native clips.   As mentioned; thats all CPU and disk.  The HW MPE accelerates effects.

                               

                              So; if your "monster" can preview AVCHD in real time smoothly; with HW MPE you should be able to toss in some color correction, transitions, etc.. and it will still playback smoothly in realtime.

                              • 12. Re: No effect of Mercury Engine, CS5
                                Ernst Lunding Level 1

                                I have found a solution. When opening the Benchmark project in CS5,CS5 will leave the project untouched and use a new name. Even if this is in the same folder the Statistics.vbs file will not recognize the path - therefore it is necessary to overwrite the original file name. Then it works.

                                I have done several tests, but I cannot beat your results Harm. Benchmark time: 27,7 secs, AVI encoding time: 12,2 secs, rendering time 6 secs. (MPE GPU acc.), Without MPE GPU acc. I get the following results: Total Benchmark time: 45,2 secs; AVI encoding time: 10 secs; Rendering time: 26 secs. Conclusion: MPE GPU acc. has an effect.

                                 

                                I have made a small test with AVCHD which is of interest for me as I am recording AVCHD 1920x1080 and burning the final project to Blue Ray disks using H264 (max quality 35 Mbps):

                                 

                                Project length: 1:45

                                No rendering in PPRO

                                Maximum rendering quality in Media encoder.

                                Blue Ray encoding (PAL) H264, VB1 35 Mbps

                                Video effects: I have used 3 GPU accelerated effects in all 15 clips (Gaussian Blur, Procamp, Fast Colour Corrector and cross dissolve between the 15 clips)

                                 

                                                      Elapsed queue time

                                GPU acc. No video effects     01:45

                                GPU acc. Video effects     01:50

                                No GPU acc. no video effects     01:45

                                No GPU acc. Video effects     03:27

                                 

                                As you can see there is no difference in Blue Ray encoding if you do not use video effects - but a marked effect if you do.

                                As far as play back is concerned I have no problems playing back AVCHD clips in the highest resolution whether I use MPE or not.

                                 

                                Ernst

                                • 13. Re: No effect of Mercury Engine, CS5
                                  Harm Millaard Level 7

                                  Ernst,

                                   

                                  You are absolutely right. I had completely forgotten because I started testing with CS5 in february or march and had forgotten that this was indeed the pitfall. When creating the queue you have to adjust the file names.

                                   

                                  I'm sorry I did not think about that. I will make a separate posting on this issue. Thanks for helping remind me of this.

                                  • 14. Re: No effect of Mercury Engine, CS5
                                    Harm Millaard Level 7

                                    Ernst,

                                     

                                    If I understand you correctly, your results are:

                                     

                                    TOTAL: 27.7, AVI: 12.2, MPEG: 9.5, RENDER: 6, with MPE and

                                     

                                    TOTAL: 45.2, AVI: 10.0, MPEG: 9.2, RENDER: 26, without MPE

                                     

                                    Your render times with/without MPE are in line with other results. The increase in AVI times with MPE turned on is what we have encountered till now. The weak link is your setup appears to be your disk setup. Your increase in performance from MPE is about equal to what I experienced.

                                     

                                    Can you PM me the Output.txt file. I will include it and forward it to Bill for publication, unless you have already sent the results to Bill.

                                     

                                    PS. Thanks for your earlier remark. I made a separate post about it. Greatly appreciated.

                                    • 15. Re: No effect of Mercury Engine, CS5
                                      Ernst Lunding Level 1

                                      Harm,

                                       

                                      You have understood me correctly.

                                      I have attached two output files to you - I have not sent them to Bill.

                                      I think you are right about my disk setup, which is: one SAS (134GB) for OS, one SAS (419GB) which I use for media cache and one SAS (419GB) for project.

                                       

                                       

                                      I repeated my test from yesterday using video effects which are not GPU accelerated. I added to all clips: auto colour, auto contrast, auto levels, convolution kernel, shadow highlights + cross dissolve. Cross dissolve is GPU acc.

                                       

                                      Without any effects the total elapsed queue time (rendering + encoding) was 1:45 for my 1:45 AVCHD-project whether I used MPE GPU acc. or not.

                                      With the not GPU acc. video effects the total rendering and encoding time was 4:03 without GPU acc. and 4:25 with GPU acceleration. This means that (with my PC) it is slightly faster not to use MPE GPU acceleration if I use the not GPU accelerated video effects.

                                       

                                      Ernst

                                       

                                      PS.: If you have any comments I will not be able to respond until next Wednesday as my wife and I are on a opera and symphony concert trip to the German city Dresden. Hopefully the weather is better in Dresden than here in Copenhagen.