Personal View site logo
Graphics card for video editing
  • 129 Replies sorted by
  • OK.

    Yep, I agree if you need very good performance GTX 670 seems sweet spot.

    but most probably in real things GTX 640 won't be much behind.

    Looking at http://www.studio1productions.com/Articles/PremiereCS5.htm it seems that Adobe products scaling is not the best.

    It requires someone to make proper test for DaVinci.

  • "It is always good to see some real benchmarks or measurements."

    For CS5.5 http://ppbm5.com/ For DaVinci i havent seen any... but i also would like to see some.

    "Are you sure that you need 4Gb?"

    I don't know exactly how many memory Premier needs for "GPU" 4K editing, but according to people that is testing it in the Adobe Hardware forums it is needed more than 1GB (sorry dont remember exactly the amount but according to them 1,5GB is enough ) and having a lot of memory helps to improve performance.

    http://forums.adobe.com/message/4290928

    On the other hand (although I think it is not very important for this conversation) having a lot of memory is important for GPU rendering as, the whole scene must be stored in the card memory. (I guess this can vary depending of the render engine)

    http://jeffpatton.net/2010/11/gtxquadrotesla-my-opinion-on-todays-gpu-selections-for-rendering/

    I just wanted to express that having that big amount of memory can be useful for a lot GPU computing applications, even if now you don't need it.

  • I have several of these cards in different computers, and looking at the "load" using GPUZ, even the GT 240 is never running at peak. That is, the program just doesn't ever use all of the GPU power even on the 240. I have a passively cooled GT 240 for my Video/Audio daw that is silent, and everything runs fine. Ray tracer in After Effects works if you hack the raytracer text file, exactly the way it is done in Premiere. If anything, playback is even smoother in CS6. I don't see a big difference using DDR5 as opposed to DDR3, but the testing shows a slight increase in throughput. An i7 with hyperthreading of course makes a big difference, and you need 12 gig ram. I plunked for the 16 gig because it is ridiculously cheap. I do believe that scaling, for all kinds of pan and scan, etc, is better using the CUDA. My GT 240s will also power the Catleap 27" monitor no problem. Lastly, native GPU acceleration in Speedgrade requires at this time a real Quadro card, if there is a hack for this I have not found it. Waiting for the budget Tesla clone :)

  • Speedgrade is another beast. I can load 4K footage, grade it and play back real time butter smooth on an older AMD Phenom II x6, 16GB, GTX460 system.

    From another forum/GPU thread, I read:

    SpeedGrade is in a league of its own in terms of performance. Instead of using GPU compute, i.e. use the GPU to mimic CPU functions, it directly leverages the GPU's shaders through OpenGL. The result is performance an order of magnitude better than the likes of Premiere Pro or Resolve. Interestingly, the SpeedGrade team recommends Quadro GPUs

  • In is very interesting conflicting data about SpeedGrade.

  • Radeon 7970 looks like an excellent contender.

    Issue is that both CS6 and DaVinci do not support AMD cards :-)

    So, if you need card today, NVidia is the only way.

  • Swapped out a GTX460 2gb with a GT640 2gb and see an increase in rendering and real time playback in PP CS6 with CUDA hack. I can't believe what you can get for $100 in performance. And no need for power connectors! Being able to run 3 monitors (2 dvis 1 Display Port) was also a plus.

  • Being able to run 3 monitors (2 dvis 1 Display Port) was also a plus.

    Do you have a link to exact card you use, as most GT 640 seem to have 2xDCI, HDMI and VGA.

  • @vas907

    OK. Almost the same as in first post.

  • For some reason using the CUDA hack with GT 640 I had to use Compatiblity performance mode to get PP CS6 to recognize the card.

  • Speedgrade runs like a turtle on my GT 240. I have read that it works with a GT 240, so maybe it is a driver issue. @vas907 did you do anything special to get speedgrade running with the GTX 460? And what version driver did you use?

  • It seems than Speedgrade just needs OpenGL 2.0 graphic card, therefore it should work fine with AMD cards... any experience with it?

    http://www.adobe.com/products/speedgrade/tech-specs.html

    Another interesting comparisons between GTX 680 and AMD 7970:

    http://www.geeks3d.com/20120427/clbenchmark-new-opencl-benchmark-for-windows-tested-hd-7970-vs-gtx-680/

    http://techreport.com/articles.x/22653/7

  • @DrDave Didn't have to do anything special when I was running the GTX 460, and I was running 301.42

  • Is anyone reading this own a new retina MacBook pro ? I thought I made a big mistake to order one hoping to use open gl on cs6.

  • @tinbeo

    It is topic about desktop graphic cards :-)

  • @stip

    This thing is mostly useless marketing shit. They want to sell Quadro :-) Just most people do not want to buy it.

  • haha, yeah that's true!

  • @vas907 Thanks for the info--are you running the standard DPX files or mov files? The DPX files are the ones that play slow on my system. @Vitaliy agreed. Even a GT 240 is only using 30-40 percent of the GPU for premiere pro.

  • @vas907 is the 640 faster than the 460 in Speedgrade? TIA.

  • @DrDave Yeah, it seems that in PPro the problem is either getting the software to utilise all the available cores (or some other bottleneck, perhaps memory bandwidth).

    According to Studio1's benchmarks (http://www.studio1productions.com/Articles/PremiereCS5.htm) there's hardly any difference in timeline rendering speeds across the various cards, there's lot more variance in DVD encoding.

    I took their listed results for CUDA rendering on CS5.5 and compared them to the GT-240. The figures below are seconds and a ratio compared to the GT-240.

    Benchmark One: AMD X4 @ 3.2 GHz / 8GB
    
    Video Card    Time Line Render    Export to MPEG-2 DVD
    None          373.0 / 10.91       387 / 1.38
    GT-240         34.2 /  1.00       281 / 1.00
    GT-440         33.3 /  0.97       275 / 0.98
    GTX-470        31.6 /  0.92       230 / 0.82
    GTX-545        32.8 /  0.96       258 / 0.92
    GTX-550 Ti     31.9 /  0.93       246 / 0.88
    GTX-570        31.5 /  0.92       193 / 0.69
    GTX-680        30.0 /  0.88       184 / 0.65
    Quadro 2000    32.5 /  0.95       257 / 0.91
    Quadro 4000    31.6 /  0.92       242 / 0.86
    
    Benchmark One: i7-920 @ 3.05 GHz / 16GB
    
    Video Card    Time Line Render    Export to MPEG-2 DVD
    None          114.0 / 9.91        176 / 0.97
    GT-240         11.5 / 1.00        181 / 1.00
    GT-440         11.5 / 1.00        180 / 0.99
    GTX-470        10.0 / 0.87         98 / 0.54
    GTX-545        11.0 / 0.96        168 / 0.93
    GTX-550 Ti     11.0 / 0.96        159 / 0.88
    GTX-570        10.0 / 0.87         97 / 0.54
    GTX-680         9.0 / 0.78         86 / 0.48
    Quadro 2000    11.5 / 1.00        166 / 0.92
    Quadro 4000    11.0 / 0.96        155 / 0.86
    
    Benchmark Two: AMD X4 @ 3.2 GHz / 8GB
    
    Video Card      Time Line Render
    GT-240          341 / 1.00
    GT-440          337 / 0.99
    GTX-470         334 / 0.98
    GTX-545         337 / 0.99
    GTX-550 Ti      335 / 0.98
    GTX-570         329 / 0.96
    GTX-680         316 / 0.93
    

    Studio1 also had this to say about H.264: "The difference between a 96 cuda core video card and a 480 cuda core video card was 7 seconds when exporting in the MPE GPU mode."

    So today, unless you're doing MPEG-2 encoding there doesn't seem to be a great reason to upgrade from the GT-240.

    EDIT Fixed benchmark two's ratios.

  • Personally, I went for GTX 670. We'll see how it'll behave, especially with upcoming DaVinci.

    Also think that GTX 560 and it's variants are also not bad, quite cheap, especially if you get it on sale.

    Of course, they are larger and consume more than GT 640, so may be it is better to just plug second GT 640 later if you need it.

  • With 1344 cuda cores you could cook a Pizza on it!

  • Its very rare that an app can supply enough data to hog 1344 CUDA on a single PCIe 10x lane. I dont have the references, but its better to have 2x cards on 2x PCIe slots and so on.. Also better for power management..As the unused cards can just stay idle.