5 Replies Latest reply on Dec 19, 2013 3:24 PM by Alex - DV411

    Pr 7.2 update's OpenCL performance improvement: a lot? a little?

    Alex - DV411 Level 2

      (forked from "AMD and Nvida cards." thread)

       

      The latest Pr CC (7.2) update touts improved OpenCL performance. Anyone has any clues, how much of an improvement it was?

       

      (I.e. are AMD cards any closer now to NVidia's performance-to-buck ratio with respect to GPU acceleration in Pr?)

        • 1. Re: Pr 7.2 update's OpenCL performance improvement: a lot? a little?
          DMH79 Level 2

          I wish there was a way to really tell or measure any performance improvements. I'm guessing since I read that there was "improvements" and I was ready to feel like it was improved...almost like taking a pill and feeling "something", yet not really knowing if it's the real thing or a placebo effect.

           

          For me (currently using AMD 6970M on a 2011 iMac) I only have access to OpenCL so I have no way of knowing if CUDA would be better/smoother. I have noticed an little improvement in things since the update, (minus a few lingering bugs that drive me nuts), but I also switched to a SSD for my media/projects so I'm wondering if that's where the small, but noticeable improvements came from. They are subtle though and there's always room for improvement.  I guess, I'd love to see a fully how a crazy maxed out computer handles my workflow and then I'd know how much "better" it can really get.

          1 person found this helpful
          • 2. Re: Pr 7.2 update's OpenCL performance improvement: a lot? a little?
            Alex - DV411 Level 2

            Thanks DMH79, great points... which put a spotlight on another question: how to measure GPU acceleration performance in Premiere Pro. PPBM is a great benchmark that is very helpful in determining the GPU acceleration effect as it measures performance with and without it... Are there other, perhaps simpler benchmarks? (I was never able to run PPBM as it gave me various error messages.)

            • 3. Re: Pr 7.2 update's OpenCL performance improvement: a lot? a little?
              DMH79 Level 2

              Actually, after working on another project, I'm not sure I see an improvement...or at least "notice" one. I've actually noticed some weird issues that weren't there before like clips skipping. I'll be watching a clip and every once in a while the actual clip will jump forward a bit as if that middle second or so was missing. Not often, but a couple times in my current project. Maybe that's not OpenCL related, but it's odd and new to 7.2 and still there in 7.2.1. When I turn off GPU Acceleration, the clip is there in it's entirety. Interesting. Also, I shut down the project, opened it back up and enabled GPU acceleration (OpenCL) again and that clip no longer had the issue. I don't render anything and I'm editing native 60fps AVCHD on a 24p timeline usually with Warp Stabilizer added. 

               

              Would be nice to "measure" these improvements. Granted, one can tweak a little something and call it a "performance improvement" so who knows to what extent we should expect it to be better.

              • 4. Re: Pr 7.2 update's OpenCL performance improvement: a lot? a little?
                JEShort01 Level 4

                Alex,

                 

                I've used the dropped frames counter to study GPU performance, assuming that you are doing all of your testing with the rest of the system remaining the same.

                 

                I like the various PPBM7 timelines + some multi-layer projects that I have prepared with AVCHD, 4k, etc.

                 

                I will give the credit to Dennis Radecke for this technique. I think it was on his Adobe blog that he was reporting "dropped frames" for various video cards awhile back.

                 

                Regards,

                 

                Jim

                1 person found this helpful
                • 5. Re: Pr 7.2 update's OpenCL performance improvement: a lot? a little?
                  Alex - DV411 Level 2

                  Indeed, he was measuring the effects of Maximus configurations in CS5.5. In combination with timing rendering or exports using PPBM or other GPU accelerated projects, that's probably all one needs to measure the effect of GPU acceleration...

                   

                  Now all we need is for someone to publish their numbers before and after Pr 7.2 update, on an OpenCL system, and hopefully similar numbers with a comparable Cuda card.... Anyone?