8 Replies Latest reply on Apr 16, 2013 1:57 AM by Harm Millaard

    Playback decompression question.

    Ron Lee's Education

      Hello. I'm working with Premiere CS6, an Nvidia GTX 470 card, a 2.83 GHz quad CPU, and 8 GB ram, and am curious about what I can reasonably expect in terms of playback decompression. I mention my graphics card, and I am using GPU acceleration, but as I understand it, that has nothing to do with codec decompression.


      Currently, I have a timeline with 3 tracks of the same DSLR clip (AVC, 45 Mbps) offset by a few seconds, with varying levels of opacity so that each clip shows. If I start playback somewhere in the middle, where all clips are active, the CPU gets pegged, and I drop frames. But if I start playback where there is only one track, and let it play a few seconds and then run into the second and third tracks, it all keeps playing just fine. Any ideas what's going on here?


      I guess I'm curious about both why there's a difference starting on one track as opposed to many, and also if anyone can tell me what's reasonable to expect out of this CPU codec-wise. Thanks very much.

        • 1. Re: Playback decompression question.
          John T Smith Adobe Community Professional & MVP

          You might want to run the CS5 Benchmark http://ppbm5.com/ to see how your computer rates... and where you are a lot slower than "fast" computers

          • 2. Re: Playback decompression question.
            Harm Millaard Level 7

            Would that be a Xeon E5440 CPU?


            One thing you are noticing is a shortcoming in the internal buffering within PR with version CS6. It was OK in CS5 and 5.5, but got broken in CS6. If my first assumption about the CPU is correct, then the problem is more noticeable than on modern CPU's, because the 54xx range is rather dated in architecture, more than 5 years old, and the system has only 8 GB memory.


            If you start from a position with only one track, the prefetching has sufficient time to get that single clip from disk into memory, but getting all the clips in the different tracks from disk into memory at the same time can slow things down considerably, especially if you use older and reasonably filled disks and the limited amount of memory does not help here.


            PS. Just saw your submission. It is a Core 2 Quad Q9550. I'll add the result shortly, but that system is pretty slow, around rank # 129 out of 143. It is just too weak for fluidly playing AVCHD material from 3 or 4 tracks. You could try lowering the playback resolution to 1/4.


            Ron Lee.png

            Ron Lee-RPI.png

            Ron Lee gauge.png

            As you can see, your H.264-BR score is pretty bad and that is the relevant test for AVCHD material.

            For 3 - 4 tracks of AVCHD material you should aim for around 75% or better. In your case with high bit-rate and low compression, you may lower that a bit to maybe 60 - 65%, but not lower.

            • 3. Re: Playback decompression question.
              Ron Lee's Education Level 1

              Thanks John and Harm, very enlightening. Couple more questions, if I might. First, I tried playback at 1/4 resolution, and it's the same. This is what I would expect, since, as I understand it, the bottleneck here is the codec decompression, and the amount the CPU has to decompress is the same regardless of the playback resolution. Or is it? There is much I don't know. Thoughts?


              Secondly, just wondering roughly what CPU/RAM combo would fall in the Economical range on the graph above. Decent i5 and 16gb? Thanks again.

              • 4. Re: Playback decompression question.
                Ron Lee's Education Level 1

                Oh, one more question. I have Premiere set to use 6.5 gb ram, and leave 1.5 for everything else, but when I play this timeline, it's only using 1 gb. Anything I can do there?

                • 5. Re: Playback decompression question.
                  Jim_Simon Level 8

                  PP will use what it needs.

                  • 6. Re: Playback decompression question.
                    Ron Lee's Education Level 1

                    Thanks Jim. That's what I would expect, just trying to find where the bottlenecks are. Interestingly, I'm now trying the same things with half as much ram installed (4 instead of 8gb), so Premiere is set to have 2.5 gb instead of 6.5 gb to play with. Doing what it was doing before, where it was using 1 gb, it never went over 800 mb, so there is a bit more going on here.


                    Also, when doing CPU intensive tasks, often I see the CPU not more than 50% used. I would expect it to be used as much as possible. Is this indicating a bottleneck in terms of getting data to the CPU for crunching?

                    • 7. Re: Playback decompression question.
                      Ron Lee's Education Level 1

                      Hello Harm, I just posted a second set to the test site, but mistakenly put version 5.5 instead of 6.0. Could you fix that, or should I repost? Thanks.

                      • 8. Re: Playback decompression question.
                        Harm Millaard Level 7

                        I'll adjust it. No need to submit again. I'll have your results ready in the next hour.


                        Using only 4 GB memory, instead of 8 GB resulted in an increase in Total Time from 636 to 708 seconds and happens for a very large extent in the MPEG test (+ 75 seconds).   The other differences can partly be attributed to +/- 1 second measurement erors. This is what it looks like now:


                        Ron Lee-4.png

                        The bottleneck in your system is mainly the CPU, lacking HT, hyper-threading and the lack of L3 cache. Further the amount of memory (8 GB) is another bottleneck, as you can see from the MPE Performance. Your video card, the 470, can easily do much better, but is held back by the bottleneck of CPU and memory and lacking L3 cache. Especially the H.264 test shows the impact of the missing L3 cache. Your disk setup is quite decent, and may even be negatively influenced by the CPU and memory, just like the video card.