25 Replies Latest reply on Dec 18, 2012 7:19 PM by RjL190365

    Long Rendering/Encoding Sequence Time

    RareTacticGaming

      Okay I am rendering a 25 minute video and it takes up to 6 hours to render! I rally can't believe it too. I think I have a decent PC, well anyways here are the export settings and my PC specs. This is in Adobe Premiere CS6. Please Help!

       

      Specs:

      AMD Athlon II X2 250 Processor @3.00GHz

      4GB DDR3 Corsair Vengeance RAM

      AMD Radeon HD 6770 1GB DDR5

       

      Export Settings:

      Format H.264

       

      Summary:

      Output

       

      1280x720

      30fps

      Prgressive

      CBR @ 8Mbps

       

      Source:

      1280x720

      59.94fps

      Progressive

        • 1. Re: Long Rendering/Encoding Sequence Time
          Harm Millaard Level 7

          That is to be expected on such a system, that is way underpowered. With that CPU, memory and video card and presumably a very lacking disk setup, such a system is at least 20 times slower than a fast Intel based desktop, so where you need 6 hours that fast system would need around 18 minutes and that makes sense.

          • 2. Re: Long Rendering/Encoding Sequence Time
            RjL190365 Level 4

            Guess what? Your system just barely surpasses Adobe's official minimum requirements to run Premiere Pro CS6 at all. You see, AMD CPUs, especially those with only two physical cores, never performed well in the Adobe programs to begin with because those CPUs support only four of the many instructions in the SSE 4.x set (thus, SSE 4.x support in the AMD CPUs is very incomplete). Second, the Windows version of Premiere Pro does not support OpenCL GPU acceleration at all at this time; GPU acceleration in MPE is NVIDIA CUDA-specific. Thus, not only is your PC handicapped by the AMD CPU, but also by the AMD GPU that cannot use MPE GPU acceleration at all. Thus, six hours to render a 25-minute project is about right for that system. Intel CPUs perform much better in Adobe.

             

            And not only that, but no PC with only a dual-core CPU performs as fast as a PC with even a mediocre-performing quad-core CPU in Premiere. The best that I can do with a dual-core Sandy Bridge Intel i3 CPU is around 500 seconds in the PPBM5 benchmark while even the cheapest of the full-power quad-core Sandy Bridge i5 desktop CPUs can run that same benchmark in about 360 seconds.

            • 3. Re: Long Rendering/Encoding Sequence Time
              RareTacticGaming Level 1

              Okay I'm thinking in getting an 17 3820, so how long do you think that will take me to render with the above settings? Also, what Nvidia Graphics card do you recommend for renderign and gaming purposes. I have a budget of around £130 or something around that for a new graphics card. I want it to be somethignas good or better than my HD 6770. Any suggestions? Thanks btw!

              • 4. Re: Long Rendering/Encoding Sequence Time
                Bill Gehrke Most Valuable Participant

                RTC, this does not exactly compare to your specs, but it is about as close as I can come for a trial encoding run.  I have a 20 minute long  1920 x 1080 AVCHD (H.264) project that is more complex (I believe) than your project because it has several necessary video effects.  I have used Fast Color Correction and also Brightness and Contrast adjustments.  My test computer is a three generation old i7-980X hex core with hyperthreading overclocked to 4.2 GHz has 24 GB of RAM and at this moment has an nVidia GTX 560 Ti 448 core GPU.  Here are the export settings:

                Sermon-Test-H.264.jpg

                To encode this it took me 6 minutes and 47 seconds!

                 

                Your i7-3820 should do well compared to your current 6 hours.  Be careful of the LGA 2011 socket CPU's they have fussy requirements on the RAM   A middle-of-the-road GTX 560 Ti like I used is about $200 to $250 dollars

                • 5. Re: Long Rendering/Encoding Sequence Time
                  RjL190365 Level 4

                  Unfortunately, you might be limited by that £130 budget. You will likely be stuck with a GTX 550 Ti, which is actually slower than your current HD 6770 in most games (but can use MPE GPU acceleration in Premiere unlike the HD 6770), at that price. A non-Ti GTX 560 would be an inprovement - but the cheapest one (as I found on scan.co.uk) is just over your stated £130 limit plus VAT.

                   

                  In other words, your planned upgrade would become CPU-heavy and GPU-light. A better-balanced choice would be an i7-3770K or even an i5-3570K plus a GTX 560 Ti.

                  • 6. Re: Long Rendering/Encoding Sequence Time
                    Bill Gehrke Most Valuable Participant

                    Here is an alternate sort of temporary path for you to consider.  The most important upgrade on your current system would be the GPU.  Since you are probably going to do it anyway why don't you get the GPU as your first option and install it in the current computer and just see how much that helps.  You do have to add a line to the Adobe file "cuda_supported_cards.txt".  Before you do make sure you have a large enough power suppy for the new card and the correct connectors to supply the card.

                    • 7. Re: Long Rendering/Encoding Sequence Time
                      RareTacticGaming Level 1

                      What do you mean they have fussy requirements on RAM? Like that some RAM won't be compatible?

                      • 8. Re: Long Rendering/Encoding Sequence Time
                        RareTacticGaming Level 1

                        I found the processors available on Amazon UK, and the i7 3820 comes to around £227 and the 3770k comes to £276?! Is that correct, I thought the 3820 would be more expensive?

                        • 9. Re: Long Rendering/Encoding Sequence Time
                          RjL190365 Level 4

                          The difference in price is offset by the motherboard that each CPU requires.The i7-3820 requires an X79 motherboard, which on average costs significantly more money than the average Z77 motherboard used for the i7-3770K. Plus, X79 motherboards require RAM sets of four or eight to perform their best whereas Z77 boards only need RAM in pairs. You cannot use an Intel CPU on an AMD motherboard: The socket and CPU are completely incompatible with one another.

                          • 10. Re: Long Rendering/Encoding Sequence Time
                            RareTacticGaming Level 1

                            The RAM sets are fine anyways, as I always buy in 4s. But are there any difference between the 3820 and teh 3770 other than the cache difference? I think I will get the 560 as long as it is better than my current graphics card. Stretching the budget to £140 would be alright.

                            • 11. Re: Long Rendering/Encoding Sequence Time
                              RareTacticGaming Level 1

                              And also I forgot to mention, does my motherboard support the cpu's? And how do i check that?

                              • 12. Re: Long Rendering/Encoding Sequence Time
                                RjL190365 Level 4

                                Your current motherboard is only for AMD CPUs. None of the Intel CPUs will fit your motherboard at all. Not even physically.

                                • 13. Re: Long Rendering/Encoding Sequence Time
                                  RareTacticGaming Level 1

                                  Sorry to ask but what motherboards do you recommend to me that will be compatible to all of the above etc? Can you recommend from the extreme series via the link below? Lol this is basically getting a new PC all over again. Thanks!

                                   

                                  http://www.intel.com/content/www/us/en/motherboards/desktop-motherboards/motherboards.html

                                  • 15. Re: Long Rendering/Encoding Sequence Time
                                    Bill Gehrke Most Valuable Participant

                                    Here is a brand new and very inexpensive GTX 600 series, a GTX 640 that was just announced today that should be quite good for the $110 price, it has 384 cores so in theory should perform as good as the GTX560 Ti.

                                    • 16. Re: Long Rendering/Encoding Sequence Time
                                      RjL190365 Level 4

                                      Bill,

                                       

                                      I just looked at the GT 640. The retail version only has the memory bandwidth that's barely higher than a DDR3 version of the GT 440: It uses only 128-bit DDR3 video RAM with a total memory throughput of only 28.5 GB/s - just less than half that of the 57.7 GB/s memory throughput of a reference GTS 450. Also, Kepler shaders only run at the GPU core speed while Fermi shaders run at double speed. Thus, if anything, the GT 640 (in this 128-bit DDR3 form) will actually be slower than a GTX 550 Ti or even a GTS 450. However, it will be noticeably faster than a GT 440 DDR3 because 384 CUDA cores in Kepler roughly equals 192 CUDA cores in Fermi.

                                       

                                      What we really need is a retail GT 640 with GDDR5 RAM instead of DDR3 RAM.

                                       

                                      On the other hand, the GeForce GT 545 currently comes in two versions: one with 192-bit DDR3 RAM and another with 128-bit GDDR5 RAM. In this particular case, the two versions perform almost equal to one another because the two have almost equal memory throughput, and that the wider bus width of the DDR3 version of the GT 545 makes up for the lack of throughput efficiency of this older RAM technology. (One version of the GT 640 that's supplied only to OEMs is a renamed GT 545 DDR3.)

                                       

                                      Message was edited by: RjL190365

                                      • 17. Re: Long Rendering/Encoding Sequence Time
                                        avidpremiereuser

                                        I read what you wrote about amd, but since my box is so new and my problem seems to be overkill im going to post.


                                        gigabyte mobo

                                        amd phenom II X6 1055T
                                        16 gb ram
                                        560 ti
                                        64 windows 7
                                        running 2 WD blues in raid, premiere off 2nd production specfic drive volume (no storage, launching avid and premiere from this drive alon

                                         

                                        export settings in premiere 6 as follows

                                         

                                        mpeg 2
                                        ntsc 720 x 480 29.97 quality 5, progressive

                                         

                                        vbr 2 pass min 4 target 6.20 max 8

                                        dolby digital 192 48k

                                         

                                        there are four exports (using media encoder)

                                         

                                        each export is 70 min in length. 

                                         

                                        im getting 16 hours for each one.

                                         

                                        is this normal even with a hexacore and that much ram?

                                        • 18. Re: Long Rendering/Encoding Sequence Time
                                          Harm Millaard Level 7

                                          16 hours for 2-pass roughly translates to 8 hours single pass. Given that your system is at least 5 times slower than a fast Intel system, but probably around 10 times slower or worse, that gives a normal encoding time on a fast system of less than 50 minutes. That sounds about right.

                                          • 19. Re: Long Rendering/Encoding Sequence Time
                                            RjL190365 Level 4

                                            The 16 hours for a two-pass MPEG-2 encode of a 70-minute project sounds about right for an AMD hexa-core CPU-powered system, especially one that's running at just its stock CPU speed. Remember, AMD CPUs lack complete support of the SSE 4.x instructions that are fully supported by even the slowest of the current Intel CPUs. That alone condemns the fastest of the AMD CPUs to roughly the level of an old quad-core Intel Core 2 Quad or a dual-core i3. So in other words, the fastest AMD CPU is slower in Premiere than even the slowest of the current Sandy Bridge or Ivy Bridge quad-core Intel i5 CPUs.

                                            • 20. Re: Long Rendering/Encoding Sequence Time
                                              RjL190365 Level 4

                                              RjL190365 wrote:

                                               

                                              Bill,

                                               

                                              I just looked at the GT 640. The retail version only has the memory bandwidth that's barely higher than a DDR3 version of the GT 440: It uses only 128-bit DDR3 video RAM with a total memory throughput of only 28.5 GB/s - just less than half that of the 57.7 GB/s memory throughput of a reference GTS 450. Also, Kepler shaders only run at the GPU core speed while Fermi shaders run at double speed. Thus, if anything, the GT 640 (in this 128-bit DDR3 form) will actually be slower than a GTX 550 Ti or even a GTS 450. However, it will be noticeably faster than a GT 440 DDR3 because 384 CUDA cores in Kepler roughly equals 192 CUDA cores in Fermi.

                                               

                                              What we really need is a retail GT 640 with GDDR5 RAM instead of DDR3 RAM.

                                              And that retail GT 640 will be the least expensive Kepler-derived GPU available on a retail graphics card - at least for the foreseeable future. There is only one Kepler-derived GPU that's below the GT 640 - and that's an OEM-only version of the GT 630, which is a GT 640 with half of its shader/texture units disabled. Other than that, all of the GT 6xx cards below the GT 640 are rebrands of the earlier lower-end Fermi-derived cards (for example, the retail GT 610, GT 620 and GT 630 are the GT 520, GT 430 and GT 440, respectively, with new names).

                                              • 21. Re: Long Rendering/Encoding Sequence Time
                                                Bill Gehrke Most Valuable Participant

                                                Randall, yes that is a pretty lousy memory solution.  But when looking at the lineup there sure is a big hole in the 600 series with the GTX 670 at $400 and this wimp of a GT 640 at $110.  I would guess that we will see GTX 650's and 660's before to long.

                                                • 22. Re: Long Rendering/Encoding Sequence Time
                                                  RjL190365 Level 4

                                                  Bill, I just read a review of the GT 640 on the Tom's Hardware site, and discovered that its performance falls in between a GTS 450 and a GDDR5 version of the GT 440. That's in line with what I had predicted a few posts up - and disappointing since the GTS 450 has only 192 CUDA cores. Predictably, the GT 440, despite the higher VRAM throughput of its GDDR5 variant, is held back by its 96 CUDA cores.

                                                   

                                                  As for the gap in the 600 series, there is only one 600-named GPU that's in between this wimp of a GT 640 and the GTX 670 - and that is a 600 only in name, the OEM-only GT 645 (which as I knew is a renamed GTX 560 SE with only 288 cores and a 192-bit memory bus using GDDR5 VRAM). Three other versions of the GT 640 are available to OEMs: A lower-clocked version of the retail GT 640, a version of the GT 640 with retail clocks but GDDR5 VRAM and a GT 640 that's really a GT 545 DDR3 (192-bit) with a new name.

                                                   

                                                  Finally, as of now no 650s or 660s have been announced yet. This is because the yields on the 28nm GPUs still have not yet been as high as either NVidia or AMD would have liked.

                                                  • 23. Re: Long Rendering/Encoding Sequence Time
                                                    Bill Gehrke Most Valuable Participant

                                                    Randall,

                                                     

                                                    Take a look at this site with GPU rumors.

                                                     

                                                    It has rumored specs on GTX 640, 650, 660Ti cards.  All have GDDR5.  Only the GTX 660 Ti has the GK 104 GPU chip (like the GTX 670 and 680) and other decent memory bandwidth specs and it has a forecast release date of August 7.

                                                    • 24. Re: Long Rendering/Encoding Sequence Time
                                                      RjL190365 Level 4

                                                      Bill, I just did take a look at that site. As I stated in my previous post, there has been no official announcement from NVidia or any of its card brand partners about any of the Kepler-derived GPUs between the GT 640 and the GTX 670. And the final specs on the released versions of those intermediate 600 series GPUs may very well be different from what had been leaked, especially since the leaked specs on the "GTX 660 Ti" specified greater memory throughput than even the GTX 680 (but only half the cores of the GTX 680).

                                                      • 25. Re: Long Rendering/Encoding Sequence Time
                                                        RjL190365 Level 4

                                                        Looking back almost six months, and since that time there are four new retail Kepler-derived GPUs that shipped.

                                                         

                                                        First off is a GTX 660 Ti: The real GTX 660 Ti is a GTX 670 with lesser memory bandwidth (192-bit GDDR5 memory at 144.4 GB/s instead of 256-bit GDDR5 memory at 192.8 GB/s).

                                                         

                                                        The non-Ti GTX 660 has the same memory bandwidth as the GTX 660 Ti, but it uses a different GPU - GK106 with 960 CUDA cores (the GTX 660 Ti uses the GK104 GPU with 1344 CUDA cores).

                                                         

                                                        The GTX 650 Ti uses a GK106 GPU like the GTX 660 - but with only 768 CUDA cores and only 128-bit GDDR5 memory at 86.4 GB/s.

                                                         

                                                        The GTX 650 is another weakling (though not quite as weak in real-world performance as the GT 640): It is basically a higher-clocked GT 640 (GK107-based) with higher-bandwidth GDDR5 memory (80 GB/s versus 28.5 GB/s for the DDR3 memory on the GT 640).