3 Replies Latest reply on Oct 31, 2011 8:29 PM by RjL190365

    $80 GT 440 Scores Almost Same As GTX 570 In PPro Benchmark Test

    flyingfish4 Level 1

      Just ran across this benchmark article that says any Nvidea CUDA card with at least 96 CUDA cores and GDDR5 memory will perform as almost as well as any other CUDA card in Premiere Pro. 


      The $80 GT 440 did just about as well as the GTX 570.  


      I'm thinking about getting the GT 440 instead of the 570 for my new build because of this. 


      Any feedback would be greatly appreciated.


      For more info, please see the 3 page article at:  http://www.studio1productions.com/Articles/PremiereCS5.htm

      ------------------------------------------------------------------------------------------ -------------------------------------------------------------------------

      Here is an excerpt of the article:


      First, our test system:

      We used an HP computer running an AMD Phenom II X4 (quad core) computer running at 3.2 Ghz with 8 gigs of RAM and a single Samsung 7200 rpm SATA 3.0Gb/s hard drive.  The computer was running Window 7 Home edition. All of the video cards had DDR5 memory.


      Testing Methods - I used Adobe Premiere CS5.5 and two different benchmark tests.  Each test was run 5 times on each video card, to come up with an average time for rendering the time line.  When I changed out each video card, I would uninstall the NVidia video card driver and then download and install the latest driver for the video card being tested. 


      All tests were run with the MRQ (Maximum Render Quality) set to ON.  To measure the time, I used a stopwatch.


      Note: When I ran the PPBM5 benchmark with the Mercury Playback Engine in Software mode, I only ran it only 1 time with each video card installed, since the video card's GPU would not be used.  The 373 Seconds listed in the chart is the average time from all 5 video cards I tested.  There was only a 2 to 3 second difference in the software mode due to other processes running in Windows.


      MPE GPU = I was running the Mercury Playback Engine using GPU Acceleration.

      MPE Software = I ran the test with the Mercury Playback Engine using Software only.


      First Benchmark Test - I used Adobe Premiere CS5.5 and the PPBM5 benchmark project that uses 4 video tracks.

      Each video card had at least 1 Gig of DDR5 memory.

      The chart below shows the time it took to render the timeline with multiple effects and dissolves.




      Video Card# of Cuda CoresMPE GPUMPE Software
      GT-2409634.2 Seconds373 Seconds
      GT-4409633.3 Seconds373 Seconds
      GTX-47044831.6 Seconds373 Seconds
      GTX-545 14432.8 Seconds373 Seconds
      GTX-550 Ti19231.9 Seconds373 Seconds
      GTX-570 48031.5 Seconds373 Seconds

      Second Benchmark Test (Updated 07/10/11) - I used Adobe Premiere CS5.5 and a 10 minute time line with 3 tracks and multiple effects and dissolves.  Each video card had at least 1 Gig of DDR5 memory.

      Video Card# of Cuda CoresMPE Hardware
      Minutes & Seconds
      MPE Software
      Minutes & Seconds
      GTX-545 1445:3740:12
      GTX-550 Ti1925:3540:12
      GTX-570 4805:2940:12

        A word about the Benchmarks and Results: The benchmark tests were run only rendering the timeline.  This was done so I could show you the difference in using the Premiere in GPU mode vs. software mode using various video cards.


      Your results will vary from mine simply because our systems are different.  From the chipset on the motherboard, to the memory chips, brand of video card, etc. all of these things will make a difference in the benchmarks.  If you run your own tests, make sure you run your tests with the Maximum Render Quality set to ON for both Software mode and GPU mode. All of the video cards had DDR5 memory.


      Explaining the Test Results

      Let me make this very clear, Adobe Premiere uses ALL of the CUDA cores on the video card.


      So you are probably wondering what's going on with the results?  When you look at the GT240, which only has 96 CUDA cores and the GTX570 has 480 CUDA cores.  Clearly the GTX570 is a much more powerful video card and should blow away the GT240.  Technically, you would be correct in that assumption.


      However, the video card's GPU is out running the rest of the computer, causing the GPU to have to wait for the data from the rest of the computer.  This is why there is not much of a speed difference in the benchmark tests we ran.


      If you look at the second Benchmark, you will see the GT240 video card with 96 CUDA cores took 5 minutes and 41 seconds to render the timeline, while the GTX570 with 480 CUDA cores took 5 minutes and 29 seconds to render the same timeline.  The GTX570 was 12 seconds faster because of the faster GPU, wider memory bandwidth and the more CUDA cores.


      NOTE: You might have notice the GT240 and GT440 both have 96 CUDA cores, but the GT440 card was a few seconds faster in the second benchmark.  The reason is the GT440 has a faster GPU clock speed.


      I have received numerous emails from people who said, when the upgrade from one video card to another, they didn't get any performance boost.  And we didn't see that much either with our test system and the benchmarks above.  However, I have had a couple of emails stating they did see a small increase in timeline rendering speed.  The reason why some people don't see any difference and other do see a slight difference can be attributed to how their computer system is configured.