22 Replies Latest reply on Feb 10, 2011 5:41 AM by digitlman

    PPBM5 Questions and GPU Rendering Concerns

    digitlman Level 1

      Bill and Harm, i figured i woud post this here instead of a private email so as to get some opinions and input from others.

       

      I am a bit confused with the logic behind the gpu testing and want to make sure i understand how this is working and make sure i dont have any problems with my machine.

       

      I just wiped out my 8 core, 20GB ram, SSD, gtx 460/2GB, machine and put a fresh win7/64 on it with cs5 and nothing else.

       

      i am planning on doing various tests with the SSD, then SATA Raid, then my 4gbit fibrechannel in combinations.

       

      My concern is over the gpu acceleration and timeline rendering. My basis for wanting these gfx cards with cs5 is on the PPBM results showing all kinds of double digit speed gains with MPE enabled.

       

      I have 8 of these machines and need to upgrade them all so i am concerned to make sure what i am doing and what i am getting for it.

       

      my first obvious question for everyone else is when you load the benchmark in what do you see on the timeline? i have all yellow except for two small red lines near the end of each section where the curtain and cube effects are at.

       

      Now when i hit the spacebar to play the timeline at starts out ok but then hits the 8 second mark into it and chokes out the rest of the way skipping frames and drawing maybe 3-5 frames per second until it gets through the rest.  Is this what happens for everyone else or does your system play the yellow area smoothly?

       

      Is this a by-product of my manually putting the 460 into my supported cards file and premiere not properly calculating what the gpu can and cannot accelerate?

       

      Now as for the benchmark testing itself when i hit enter it simply renders out those 2 effects that have the red line then starts playing the same choked out skippy video as before.

       

      So if this speed is calculated and then the speed of my cpu rendering the entire timeline then how is this an effective comparison? if i simply remove those two effects at the end then i still have that big complicated timeline and yet if i hit enter it just plays choppy video instantly and nothing is rendered, so compared to my cpu thats a bazillion times faster but i dont get anything out of it as the video doesnt actually play with the gpu.

       

      My real concern and need is flat out rendering speed for encoding. i had high hopes that getting a video card with hundreds of cores on it would magically render out mpeg 2 files 10-20x faster then my cpus can.

       

      i saved out a pic of my cores when i am playing the timeline, you can see they are not being maxxed out.

       

      Before i wiped my system out i tried the trial version of the mainconcept gpu encoder and it went slower with the gpu than it did with my cpus. Their email response to me was that gpu rendering is more of an advantage against slow cpu systems. and my cpu's were too fast in comparison. I find this hard to believe in that my system is several years old now. thier own performance comparison charts show a tremendous speed increase over modern I7 processors.

       

      On a side note - DONT INSTALL MAINCONCEPT TRIAL ENCODER!- it hijaacked my mpeg encoding/decoding even when i wasnt using it and it was watermarking all my stuff in premiere with its logo!

       

      I also have been using a program called canopus procoder which does great encoding with multi cores, a standard dv to mpeg2 encode on my system maxxes all 8 cores and renders about 14x faster than realtime.

       

      I hoped a cuda card could increase this speed for me and render faster then that directly out of premiere. I would think during rendering your cpu's and gpu's would work together at 100% usage to go as fast as possible. i would like to see 20-30x realtime speed for mpeg2 encoding in my situation.

       

      Thanks and i look forward to any input or help or even a simple "click on this and yoru problem is solved" 

       

      ppbmcpu1.jpg

        • 1. Re: PPBM5 Questions and GPU Rendering Concerns
          Jeff Bellune Level 6

          Please don't start the same topic in a different forum.  This is where your question belongs.

           

          -Jeff

          • 2. Re: PPBM5 Questions and GPU Rendering Concerns
            Harm Millaard Level 7
            Now when i hit the spacebar to play the timeline at starts out ok but then hits the 8 second mark into it and chokes out the rest of the way skipping frames and drawing maybe 3-5 frames per second until it gets through the rest.  Is this what happens for everyone else or does your system play the yellow area smoothly?

             

            Is that before or after rendering?

             

            On my system, which is a simple i7-920 with only 12 GB RAM and a GTX 480, with the H.264 sequence, plus an additional extra fifth track of 10 seconds of AVCHD with the corner pin effect and keyframed opacity in both parts of the timeline, I get:

             

            Dropped frames on the H.264 timeline with an extra 5-th track, out of 1,560 frames
            Unrendered
            Rendered
            Playback at full resolution3246
            Playback at half resolution756

             

            This was tested while there were a number of applications running, including anti-virus, disk optimizer, FF with around 10 sites active, etc. so not ideal to test. The area where dropped frames occur is from around 20 seconds in the time-line. Before that no dropped frames at all, even with the additional 5-th track in it.

             

            You can design 100's of time-line tests, all different and measuring completely different things. This is one time-line and when we disgned it we had a number of practical considerations.

             

            At that moment only Quadros could handle more than three tracks but the 285 was limited to three tracks maximum. This limit has been lifted since, but back then, this was a practical limitation. Second, we had to take in mind that not everybody would have CUDA enabled cards, but we still wanted them to participate in the benchmark, not shy them away by extremely long render times. As you see, there are a number of people that have to wait 5 - 7 minutes for the rendering to finish. I commend these people on the patience they have to run this test. It should not be more demanding or time consuming. During the testing before release and with the help of numerous beta testers, we found CUDA/MPE results in the range of 6 - 30 seconds and with software MPE around 70 - 260 seconds. That appeared doable. Since then the MPE technology has evolved and improved, which is why we now have low scores of 4.

             

            Whatever you think about the effects included in the test, since everybody has done the same test, the results have a common basis and are comparable.

             

            There will be a new version, probably for CS6, that will entail even more complex time-lines. The reason is that for statistical accuracy we need to reduce measurement errors as much as possible and with MPE ON scores around 4 seconds, the margin of error is too big IMO. I definitely want to have a test with a low score in the range of 20 - 30 seconds, but do not want to have people with lesser machines die of old age while waiting for the rendering to finish. I also do not want to increase the download size.

             

            There are films on YouTube where Adobe promotes MPE and it shows MPE gains of around 30 x. In our current benchmark the gains are around 10 x. Possibly in the new version it may only show gains around 5 x, because of other effects used, etc.

             

            You always have to take care in interpreting benchmark results. They can tell quite a lot, but there are cases known of 'How to lie with statistics'. That is why we offer people to get further explanation on their results if they don't understand them.

             

            In your case, the numerous frames dropped on a fresh install, running an 8-core system with 20 GB and a GTX 460 does not sound normal. Maybe you have freshly installed the OS and CS5, but have not yet 'tuned' your system, like turning off the Sidebar, indexing, compression and the like. Have a look here: Adobe Forums: Guide for installing and tuning a Vista... which says Vista, but also applies to Win7.

             

            I hoped a cuda card could increase this speed for me and render faster then that directly out of premiere. I would think during rendering your cpu's and gpu's would work together at 100% usage to go as fast as possible. i would like to see 20-30x realtime speed for mpeg2 encoding in my situation.

             

            Encoding is still done by the CPU, not the GPU. Scaling, blending and some effects are handled by the GPU, but the majority is still done by the CPU.  Even if you have a lot of scaling and blending going on, the traffic to and from the CPU - GPU means that one or the other component is waiting for the other component. Therefore you will not likely see both at 100% utilization. For instance (at least on my system) the GPU load on rendering the H.264 timeline is always below 50%, because the GPU is waiting for my slow CPU to come up with new data to process.

             

            Hope this answers your question somewhat.

            • 3. Re: PPBM5 Questions and GPU Rendering Concerns
              JEShort01 Level 4

              Harm,

               

              How to you "test" for dropped frames for playback?

               

              I too am seeing a cleaner, smoother, playback for CPU rendered PPBM5 timeline vs. the same timeline played back after a GPU render.

               

              Thanks,

               

              Jim

              • 4. Re: PPBM5 Questions and GPU Rendering Concerns
                Harm Millaard Level 7

                Sorry,

                 

                Not at liberty to tell. NDA...

                • 5. Re: PPBM5 Questions and GPU Rendering Concerns
                  JEShort01 Level 4

                  Digitl,

                   

                  Using a i7-860 PC with a GTX 460 GPU, when I do a "Sequence / Render Entire Work Area" render the GPU render is just under twice as fast as the CPU render. Both play back extremely cleanly with the "render bar" showing full green for both.

                   

                  I'm not sure exactly what is going on here, but it does seem that the GPU is either speeding up something, or requiring that less sophosticated render files are required for playback.

                   

                  While 2x the render speed is not as impressive as the 15x rendering effects only with MPE on (MPE on vs. MPE off and rendering "effects in work area" - Enter key render only), I really do require smooth playback for my editing purposes.

                   

                  This also begs the question, how would a GTX 580 perform in this same situation. Would the "effects only" render play back more cleanly? Would the GTX 580 speed up the "render entire work area" case?

                   

                  BTW, all my testing and comments are using the base case PPBM5 timeline.

                   

                  Regards,

                   

                  Jim

                  • 6. Re: PPBM5 Questions and GPU Rendering Concerns
                    digitlman Level 1

                    harm, thats kind of funny actually, you show us a report of your dropped frames on playback then say you cant tell us how because of NDA, yet we obviously know you have an app that reports the dropped frames during playback

                     

                    But thats ok, anyways as far as your ppbm goes i love it and dont mean to desparage it in any way, i have been a big supporter and love the work you and bill have put into it. I am really just trying to have a better understanding of how it works.

                     

                    My concern is more about premiere as opposed to ppbm.  I want to know if my system matches what everyone else sees, when i load the project is get this:

                     

                    ppbmtimeline.jpg

                     

                     

                    i just want to make sure everyone else has the same tiny red sections that premiere feels it can't play.

                     

                    previously premiere has always either played in realtime at 30fps or shown a red line which needs to be rendered. then if you render it you get full 30fps playback on the entire timeline.

                     

                    now with this new gpu workflow i can only render those two tiny little red sections and the rest of my timeline plays at about 2 or 3 fps from 8 seconds into the timeline until the end.

                     

                    So I actually have no physical way to see my video at full 30fps without changing to software mode and rendering the entire timeline?  or am i missing something here?

                     

                    Isn't premiere supposed to recognize when it cant keep up with 30fps playback and change that section of the timeline to red?

                     

                    And on a side note i still have to say i am fairly annoyed that an adobe employee took it upon themselves to move my thread into the unpopular generic premiere hardware forum, when this is a very specific premiere cs5 question thread. I have never even looked in this forum for any cs5 posts before now and had hoped to get more feedback from users in the main premiere cs5 forum.  A lot of companies like to treat their customers well and the old addage "the customer is always right" doesn't seem to apply anymore. My company probably has purchased over $1,000,000 in adobe software licenses over the years which i feel would make me a valuable customer, and i have some important questions i need to ask about the new version of your software and how it is supposed to work.

                     

                    thanks

                    • 7. Re: PPBM5 Questions and GPU Rendering Concerns
                      digitlman Level 1
                      function(){return A.apply(null,[this].concat($A(arguments)))}

                      JEShort01 wrote:

                       

                      Digitl,

                       

                      Using a i7-860 PC with a GTX 460 GPU, when I do a "Sequence / Render Entire Work Area" render the GPU render is just under twice as fast as the CPU render. Both play back extremely cleanly with the "render bar" showing full green for both.

                       

                      I'm not sure exactly what is going on here, but it does seem that the GPU is either speeding up something, or requiring that less sophosticated render files are required for playback.

                       

                      While 2x the render speed is not as impressive as the 15x rendering effects only with MPE on (MPE on vs. MPE off and rendering "effects in work area" - Enter key render only), I really do require smooth playback for my editing purposes.

                       

                      This also begs the question, how would a GTX 580 perform in this same situation. Would the "effects only" render play back more cleanly? Would the GTX 580 speed up the "render entire work area" case?

                       

                      BTW, all my testing and comments are using the base case PPBM5 timeline.

                       

                      Regards,

                       

                      Jim


                      Hi jim, thanks for the info, you snuck your post in while i was typing mine.

                       

                      I didnt realize there was a separate force render all function up there, that gives me full speed playback.

                       

                      So now i guess my question for harm and bill is why not use that function for the ppbm comparison speed test for rendering?

                      • 8. Re: PPBM5 Questions and GPU Rendering Concerns
                        digitlman Level 1

                        so interesting, with software rendering it takes my machine 2:15 to render out the timeline and with mpe mode it takes 1:35 so now i know my actual MPE performance gain is less than 2x faster than my cpu.

                         

                        So this begs the question for bill and harm, as to why dont you render the full timeline to compare the performance gain?

                         

                        It would make sense if the yellow parts of the timeline played in realtime and actually didnt need rendering then the red area rendering would be an accurate benchmark.

                         

                        I just really wanted to make sure there was not anything wrong with my setup and maybe other users are getting 30fps playback in the yellow timeline areas and i was not because adobe isnt supporting my gfx card properly yet or something.

                         

                        I guess i have been spoiled in that i have a whole bunch of matrox axio machines which have been doing realtime 1080p playback for years and would accurately indicate if it was too complicated and would red line those sections to indicate rendering needs.

                         

                        thanks!

                        • 9. Re: PPBM5 Questions and GPU Rendering Concerns
                          Harm Millaard Level 7
                          And on a side note i still have to say i am fairly annoyed that an adobe employee took it upon themselves to move my thread into the unpopular generic premiere hardware forum, when this is a very specific premiere cs5 question thread.

                           

                          This is a user-to-user forum, so any moving of threads from one forum to another is an action by a Mod, not an Adobe employee.

                           

                          Mods deserve our praise for the thankless, volunteer work they do to make our lives nice on the forum.

                           

                          As to unpopular generic hardware forum, I do not quite agree. It generates less traffic than for instance the generic DreamWeaver forum, but in general the content is quite valuable and the questions posed do not tend to be of the 'noob' qualitry that hundreds of repeated same questions on the general PR CS5 or CS4 forum often are.

                           

                          Questions tend to be better documented, better phrased, more to the point than on the general CS5 and CS4 forums.

                           

                          So this begs the question for bill and harm, as to why dont you render the full timeline to compare the performance gain?

                           

                          For the simple reason it would defeat the presence of hardware MPE.

                           

                          It would make sense if the yellow parts of the timeline played in realtime and actually didnt need rendering then the red area rendering would be an accurate benchmark.

                           

                          The yellow bar indicates that PR sees it timeline as being of moderate complexity and can often be played back in realtime without rendering. Whether that is the case in an individual situation, depends on a lot of different factors, CPU, memory, disk I/O, processes running, etc, that are not included in the analysis to determine whether a timeline should be red or yellow. It is not that intelligent.

                           

                          Keep in mind that previews are exactly that. It is not the final output. And keep in mind that even red parts can play in real time, even if not rendered, at least on a fast system.

                           

                          In your case the dropped frames can mean there is still some tuning ahead of you to get better performance and less dropped frames.

                          • 10. Re: PPBM5 Questions and GPU Rendering Concerns
                            digitlman Level 1

                            I dont understand how you say that defeats the mpe? with mpe on i got about 75% performance increase in rendering which to me is an accurate result of the mpe acceleration. With the current setup the mpe yellow line is a false representation of what mpe is "supposed" to do in realtime. Like i said before if you had left out those two transitions near the end then you wouldnt even have anything to render, they only comprise a small 5% section of the timeline.

                             

                            anyways, i have to differ with your opinion of realtime playback for lots of people that is a normal workflow. although we do tapeless workflow for our rendering to our broadcast server and rendering to dvd or bluray. We still do realtime playback output to tape as hardcopy backup and as a delivery medium to other broadcasters that require tape delivery. we output realtime firewire to DV and HDV, as well as SDI realtime output to Beta tapes and realtime HD-SDI output to HD tape formats.

                             

                            In CS4 this always worked fine in realtime, usually nothing needs rendering but if some parts have too many layers or effects then they get a red line and premiere would render those sections out and the rest would all play in realtime perfectly.

                             

                            The purpose of my testing is to discover exactly how cs5 and mpe work by themselves, I am now planning on installing my axio drivers and see how the combination of the two compares. It is interesting in other threads i have read talking about the difference with what the gpu rendering works on as opposed to how the cpu needs to decode various file formats slows things down. I am curious to see with the axio card decoding my video how much better performance i may see from the gpu. just as a test i might re-encode your ppbm video into matrox lossless uncompressed codecs and see how much faster it might perform.

                             

                            I realize this aspect doesnt apply to most cs5 users, but it is a big deal for us as we have a whole bunch of systems with axio cards with need to get nvidia gpu cards in and upgraded to cs5 and figure out a proper workflow for everything.

                             

                            thanks for all your help

                            • 11. Re: PPBM5 Questions and GPU Rendering Concerns
                              JEShort01 Level 4
                              function(){return A.apply(null,[this].concat($A(arguments)))}

                              ...when I do a "Sequence / Render Entire Work Area" render the GPU render is just under twice as fast as the CPU render. Both play back extremely cleanly with the "render bar" showing full green for both.

                               

                              I'm not sure exactly what is going on here, but it does seem that the GPU is either speeding up something, or requiring that less sophosticated render files are required for playback.                                                                                  

                               

                              What is the explaination for this - is the GPU assisting with the file rendering, or is the reason for the faster render with "MPE on" due to less elaborate files being necessary for "green bar" playback files when a MPE GPU is available?

                               

                              function(){return A.apply(null,[this].concat($A(arguments)))}

                                                                                                                                                This also begs the question, how would a GTX 580 perform in this same situation. Would the "effects only" render play back more cleanly? Would the GTX 580 speed up the "render entire work area" case?          


                              Can anyone comment if a GTX 480 / 580 level of GPU (currently using 1MB GTX 460) would allow for improved timeline playback, with significantly fewer dropped frames when I simply do a "effects only" render with MPE on?

                               

                              Also, would a faster GPU speed up the "render entire work area" case?

                               

                              Harm, I'm hoping that you may have some input here, especially since you are armed with a special playback "drop frame" detector!

                               

                              Thanks,

                               

                              Jim

                              • 12. Re: PPBM5 Questions and GPU Rendering Concerns
                                digitlman Level 1

                                well the gpu certainly does help speed rendering for certain effects or functions it seems.

                                 

                                i did some testing on 2 machines with task manager cpu opened and a gpu meter both running so i could see what was going on.

                                 

                                cpu bounces around from 40-80% and gpu is the same bouncing from 0-80% in various sections, in software mode gpu is 1% the whole time.

                                 

                                anyways my results from ppbm5 h264 full timeline render:

                                 

                                on an old slow 8 core machine with a gtx 460 2GB card

                                 

                                cpu: 4:42

                                gpu: 3:19

                                 

                                on a little faster 8 core machine with gtx 460 2GB card

                                 

                                cpu: 2:16

                                gpu: 1:35

                                 

                                 

                                I think this reflects a fairly accurate representation of the rendering speed increase you get with this gfx card. It does help out but it is not a massive double digit multiplication speed increase.

                                • 13. Re: PPBM5 Questions and GPU Rendering Concerns
                                  Harm Millaard-38t4Q4 Level 1

                                  There is definitely something wrong with your setup. Dropped frames and these results...

                                   

                                  95 seconds is the second worst score overall, only a single laptop with a GTS 360M scores worse.

                                   

                                  Nearly 2.5 times slower than the worst 8 core system with 38 seconds and the next 8 core scores 23. All the rest do far better. There are even non-accelerated systems doing better than that. The fastest 460's take around 5 or 6 seconds.

                                   

                                  The truth is often more in large numbers than single observations and yours is a clear exception.

                                   

                                  I think this reflects a fairly accurate representation of the rendering speed increase you get with this gfx card. It does help out but it is not a massive double digit multiplication speed increase.

                                   

                                  ONLY for your system. If 100 people report that the top speed of a certain brand/model/type of car is say 150 MPH and you report on that same brand/model/type your top speed of 25 MPH, who would you believe? One would question if you ever got it out of first gear.

                                  • 14. Re: PPBM5 Questions and GPU Rendering Concerns
                                    digitlman Level 1

                                    Harm, i think you are completely missing what i am doing with this experiment.  My apologies if i am not wording this very well, sometimes i have issues explaining myself properly, i will try to clarify my test.

                                     

                                    Your current timeline render benchmark is only for pressing enter and rendering nothing but those 2 little effects at the end of each section only. your fastest results are about 4 seconds. then you compare that time of rendering about 3 seconds of the timeline with the result of rendering the entire 52 second timeline and get a result of 10-18x faster render speeds.

                                     

                                    I am comparing rendering the ENTIRE timeline by going up the the sequence pulldown menu and click "render entire work area"

                                     

                                    i did this with the gpu on and with it off to compare the rendering speed for the entire timeline to see how much faster it renders with the gpu enabled.

                                     

                                    those were the times i got for doing that. I feel it gives a fairly accurate result of the GPU rendering gain.

                                     

                                    on my system i get about 1.43x speed gain with the gpu enabled during the complete timeline render.

                                     

                                    thanks

                                    • 15. Re: PPBM5 Questions and GPU Rendering Concerns
                                      Harm Millaard Level 7

                                      Got it. But it appears to me like you are forcing a render where no render is necessary. So, to make an anology, you and a friend are travelling from Mexico City to Toronto. You have a plane ticket (hardware MPE) and your friend has a bus ticket (software MPE). If you were to travel separately, you would arrive much earlier than your friend, but you deliberately choose to travel by bus too, Now you arrive at the same moment in Toronto. IMO hat does not make sense.

                                      • 16. Re: PPBM5 Questions and GPU Rendering Concerns
                                        digitlman Level 1

                                        Thanks, now we are almost on the same page. how is it you say i dont need to render the timeline if my computer can't actually play the timeline? at certain points in the middle of the timeline it is dropping about 26 of 30 frames. even your much faster overclocked i7 cpu is dropping 330 frames during playback according to your report. This was my original question at the start is can anyones computer play the yellow line area in realtime without dropping a single frame? The MPE should play yellow areas 100% perfect or they should be red and need rendering. When i record my video out to tape it must be flawless or it's no good. my system can only play 16 total seconds out of the 52 second timeline without dropping any frames.  (first 8 seconds of each section)

                                         

                                        There are two different aspects to gpu usage in premiere now, one is realtime MPE playback of the timeline and the other is actual rendering speed.

                                         

                                        MPE playback performance can only be defined by 100% flawless playback, anything else doesn't count and needs to be rendered.

                                         

                                        Actual rendering speed is pretty straightforward and myself and the other gentleman both got similar results in our first tests of rendering the entire timeline.

                                        • 17. Re: PPBM5 Questions and GPU Rendering Concerns
                                          Bill Gehrke Most Valuable Participant

                                          I am late getting into this thread but clear up a few technical issues for me.  You keep mentioning 8 cores and a slightly older system, could you tell us what you really have.

                                           

                                           

                                          1. Do you really have 8 cores or are you using a single four core processor with 8 threads from hyperthreading. 
                                          2. Are you using two processors with four cores each
                                          3. Are you using AMD or Intel.
                                          4. What actually is the processor model number
                                          5. What is your project disk system

                                           

                                          These are important factors that we get from full submission of PPBM.

                                          • 18. Re: PPBM5 Questions and GPU Rendering Concerns
                                            digitlman Level 1
                                            function(){return A.apply(null,[this].concat($A(arguments)))}

                                            Bill Gehrke wrote:

                                             

                                            I am late getting into this thread but clear up a few technical issues for me.  You keep mentioning 8 cores and a slightly older system, could you tell us what you really have.

                                             

                                             

                                            1. Do you really have 8 cores or are you using a single four core processor with 8 threads from hyperthreading. 
                                            2. Are you using two processors with four cores each
                                            3. Are you using AMD or Intel.
                                            4. What actually is the processor model number
                                            5. What is your project disk system

                                             

                                            These are important factors that we get from full submission of PPBM.

                                             

                                            hey bill, thanks for your input. my system is a little dated by i7 standards but its

                                            dual xeons with 8 cores total, 3.2ghz  X5482 with 20GB ram, SSD os disk and 4gbit Fibrechannel 8x1TBR6 raids (4 of them)

                                            geforce 460 2GB video card

                                            • 19. Re: PPBM5 Questions and GPU Rendering Concerns
                                              Harm Millaard Level 7

                                              MPE playback performance can only be defined by 100% flawless playback, anything else doesn't count and needs to be rendered.

                                               

                                              That is utopia. There will almost always be a few dropped frames during playback. In my case I had only 6 dropped frames not your claimed 330. OK, I had to render those small red segments. Forcing a render of the complete timeline still gives me 6 dropped frames during playback. That happens on the cuts between 25i and 29.97p or 29.97i and when switching from XDCAM to AVCHD or vice versa. That is a CPU matter, not an MPE matter.

                                               

                                              In short:

                                               

                                              A. Without hardware MPE I have 6 dropped frames after rendering.

                                              B. With hardware MPE I have 6 dropped frames after rendering.

                                              C. With hardware MPE and rendering the whole timeline, I have 6 dropped frames.

                                               

                                              Neither are flawless in your definition. The only difference is that A takes 73 seconds, B takes 4 seconds, C takes 49 seconds.

                                               

                                              There is no difference in the ultimate result on playback, same dropped frames. The only difference is how you get from Mexico City to Toronto, by bus (A), by plane (B) or by skipping the plane and deliberately taking the bus (C). The ultimate result is that you end up in Toronto.

                                               

                                              The only conclusion you can draw is that lowly i7-920 with only 12 GB memory may still be too low specced to avoid all dropped frames in this very complex timeline. More complex than the standard H.264 BR timeline.

                                               

                                              Another analogy:

                                               

                                              You have a printer with a 500 sheet paper tray (hardware MPE). You can also use manual sheet feeding (software MPE). You deliberately bypass automatic sheet feeding on a 500 page print job (not using hardware MPE) and only using manual feeds to mimic softeware MPE. What are you now measuring? Your own dexterity of feeding sheets into the printer...

                                              • 20. Re: PPBM5 Questions and GPU Rendering Concerns
                                                Bill Gehrke Most Valuable Participant

                                                I hate to tell you this but I just junked my Supermicro X7DWA-N system, now I only had E5410's which was fine for CS4 but not up to really sucessfully running CS5.  The shameful part was that it while I was able under special circumstances to push it to 3.0 GHz it was not really practical to overclock the beast..

                                                • 21. Re: PPBM5 Questions and GPU Rendering Concerns
                                                  JEShort01 Level 4
                                                  function(){return A.apply(null,[this].concat($A(arguments)))}

                                                   

                                                  MPE playback performance can only be defined by 100% flawless playback, anything else doesn't count and needs to be rendered.


                                                  I agree with you Digitlman that full renders are required using my current system.

                                                   

                                                  My test results for a 4-core overclocked i7-860 (4GHz) and using the PPBM5 timeline are as follows:

                                                       5 seconds - MPE "render effects only" (PPBM MPE test); unacceptable playback quality

                                                       86 seconds - MPE "render timeline" - "render entire work area" (PPBM CPU test); perfect playback quality

                                                       53 seconds - MPE "render entire work area" (39% faster than CPU only); perfect playback quality

                                                   

                                                  Harm, you really seem to get into analogies, so for me the 5 second MPE render option is like taking a fast plane, train, or Ferrari ride with muddy windows. You could say that it is you arrived quickly, but the journey is not very enjoyable and you miss a lot of details along the way. I like fast cars, but will not compromise on seeing out of the windows.

                                                   

                                                  Cheers,

                                                   

                                                  Jim

                                                  • 22. Re: PPBM5 Questions and GPU Rendering Concerns
                                                    digitlman Level 1
                                                    function(){return A.apply(null,[this].concat($A(arguments)))}

                                                    JEShort01 wrote:

                                                     

                                                    function(){return A.apply(null,[this].concat($A(arguments)))}

                                                     

                                                    MPE playback performance can only be defined by 100% flawless playback, anything else doesn't count and needs to be rendered.


                                                    I agree with you Digitlman that full renders are required using my current system.

                                                     

                                                    My test results for a 4-core overclocked i7-860 (4GHz) and using the PPBM5 timeline are as follows:

                                                         5 seconds - MPE "render effects only" (PPBM MPE test); unacceptable playback quality

                                                         86 seconds - MPE "render timeline" - "render entire work area" (PPBM CPU test); perfect playback quality

                                                         53 seconds - MPE "render entire work area" (39% faster than CPU only); perfect playback quality

                                                     

                                                    Harm, you really seem to get into analogies, so for me the 5 second MPE render option is like taking a fast plane, train, or Ferrari ride with muddy windows. You could say that it is you arrived quickly, but the journey is not very enjoyable and you miss a lot of details along the way. I like fast cars, but will not compromise on seeing out of the windows.

                                                     

                                                    Cheers,

                                                     

                                                    Jim

                                                     

                                                    Yes Jim your results are similar to mine, thanks for your input.  I agree that perfect playback is required for any NLE. i have been using premiere for about the last 14 or so versions since V4 in the early 90's and have always been able to record video out to tape perfectly.

                                                     

                                                    It's not like i could fedex a Million dollar commercial tape to FOX for the superbowl and tell them "dont worry if it appears to have some dropped frames, i rendered it in 4 seconds"

                                                     

                                                    Harm, i am not trying to be offensive or start a war or anything i am just trying to understand how cs5 works now. You are a great guy and a tremendous help to the adobe community and i appreciate any help you can give me with getting all of our editing systems running well on CS5. I am confused by your chart you posted and i interpreted the dropped frames chart with 324 in the unrendered category as 324 dropped frames that were not rendered. and the 75 in half resolution made sense as only 75 unrendered frames dropped at half resolution. It looked like 6 of the rendered frames were dropped in both playback modes. I had 5 other video editors here look at your chart and they all had the same conclusion.

                                                     

                                                    So if you had 6 dropped frames in both full and half resolution then what is the other column with the 324 value for?

                                                     

                                                    Not that it matters anyway, as you said it was not an ideal test situation and those results are not really important.

                                                     

                                                    thanks