Skip navigation
Don_1959
Currently Being Moderated

Premiere Pro CS5.5 mercury playback engine will be supported on GTX 680

Mar 30, 2012 5:28 PM

Could someone please confirm if Premiere Pro CS5.5 mercury playback engine will be supported on GTX 680.

 

thanks

 
Replies
  • Currently Being Moderated
    Mar 30, 2012 7:10 PM   in reply to Don_1959

    My EVGA GTX 680 is working perfectly, my PPBM5 scores on MPEG2 -DVD encoding has dropped about 12% from my GTX 580!

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 30, 2012 9:55 PM   in reply to Bill Gehrke

    Eeesh.  That's not much of an improvement.

     

    I don't suppose you'd have a $200 card around to compare it to, eh Bill?  Like a 560 Ti?

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 30, 2012 10:26 PM   in reply to Jim Simon

    Don't bother asking this, Jim. I did some testing with a non-Ti GTX 560 (which costs about $180 to $190), and found it to perform only about on a par with a GTX 285 - nearly 120 seconds in the MPEG-2 DVD test. And the regular GTX 560 Ti is also nowhere near as fast as the limited-edition 448-core version (which is actually closer to a GTX 570 than to a standard 560 Ti).

     

    By the way, based on my estimates, the regular GTX 560 Ti would have taken twice as long as the GTX 680 in the MPEG-2 DVD test in Bill's massively overclocked i7-2600K system.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 30, 2012 10:25 PM   in reply to Don_1959

    Supported, no, but after applying the 'hack' it works as Bill said.

     

    GPU-test-PPBM5.5-new.jpg

    Jim, Bill's improvement will probably show a MPEG2-DVD score of 53 seconds. The 560 Ti 448 scored 68, so the improvement with the 680 is 22%. That is a big improvement, because it was run in a PCIe-2.0 slot, not a PCIe-3.0 slot, and secondly, this is the improvement by the number of CUDA cores on the hardware accelerated parts of the test and that is mainly scaling and frame rate conversions. The encoding is still CPU bound.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 9:57 AM   in reply to Harm Millaard

    the improvement with the 680 is 22%. That is a big improvement

     

    I'm inclined to disagree.  250% of the cost for 22% improvement is not a very good deal in my book.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 10:35 AM   in reply to Jim Simon

    as a hobbyiest i would agree.

    as a weekend warrior (8-12 weddings a yr) probably worth it, as would most speed increases that are affordable.

    as a daily pro not even a consideration to drop that little. time it money

     

    Scott

    ADK

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 10:59 AM   in reply to Jim Simon

    Jim,

     

    You are making a fundamental error in your thinking. Suppose Red Bull could increase their top speed on Sepang from 298 km/h to 334 km/h, that is a 12% increase. They would leave everybody in the dust, including Alonso, Hamilton, Button and the rest. But in lap times their performance would only improve by 2% at most. 2% is not impressive. But they would win every race. Would that be impressive (and boring)? In this case we are not talking about 2% but about 22%.

     

    Furthermore your cost comparisons are off. The 560 Ti 448 is around € 240, the 680 € 440, so that is 45% price difference, not 250%.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 12:59 PM   in reply to Don_1959

    Well if you are in the market now, the Gtx 680 is $500, which is exactly the price as the average Gtx 580 price, I would be more than happy to see a 12% improvement for the same money. I guess my question would be is there anything we will see in CS6 that will further optimize the performance.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 4:02 PM   in reply to FloridaG

    I’m new to the MPE and have just started looking at benchmarks and it appears from the graph above (PPBM5) that CUDA acceleration hardly scales beyond 192 cores with H.264. Is this generally the case with MPE and H.264 as I am only using AVCHD so aren’t concerned about the MPEG2 acceleration which scales much better?

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 4:29 PM   in reply to Harm Millaard

    The 560 Ti 448 is around € 240,

     

    I was looking at the normal 560 Ti, which runs about $200 on Newegg.

     

    But your math is off.  $500 is 250% of $200.  ($200 x 250% = $500.)  The difference is your baseline, which in this case is the cheaper card, not the more expensive one.

     

    So for 250% more money, you only get 20 some percent improvement.  That's just very disappointing.  I think we were all hoping for upwards of 300% improvement (since it has 3x the cores).

     

    Now if you want to go the other way, with the 680 as the baseline, you'd say that for 60% less money, you still get 80% of the performance.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 4:32 PM   in reply to eclipse_crow

    Hardware acceleration is useful with rendering, previewing, and on certain parts of the export process, i.e. scaling, frame rate adjustments, blurring and blending, but not with encoding. Encoding is purely CPU dependent and that is why you do not see any benefit with the H.264-BR test, because there is no scaling involved. The MPEG2-DVD test has scaling all over and that is why the cores on the video card impact on the total time.

     

    If you export to anything else than BR, then scaling comes into the picture and thus the benefit of the 680 card. If you start with AVCHD and only end up with BRD, then it makes no sense to get this 680 video card for the encoding process.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 4:37 PM   in reply to Harm Millaard

    would something like the Matrox H.264 MAX card be used by Premiere if you were exporting a project in H264 ?

     

    Chris

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 4:38 PM   in reply to Jim Simon

    I think Im going to wait and see how the" Quadro" version of the Kepler card performs when it is released later this summer.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 4:57 PM   in reply to Jim Simon

    Jim,

     

    Your comment was based upon results from the 560 Ti 448 card and that is € 240 or more. The 680 is € 440. That is 45% difference: € 440 - (0.45x440) = € 440 - € 198 = € 242

     

    Your comparison is based on a € 200 GTX 560 model, which only achieves a score of around 120 seconds, which is 55% less expensive (€ 440 - (0.55x440) = € 440 - € 242 = € 198) and more than 225% slower (120 / 53 = 2.26)

     

    So for 55% less expenditure, you lose more than 225% performance. That is very disappointing and would rank in the 'penny-wise-pound-foolish' savings. (But that is how your system can be ranked, at least in the past). There is another major flaw in your reasoning, you are only considering number of CUDA cores, nothing else. You are disregarding all the other aspects of the new architecture and that makes your argument very weak and flawed. You have to consider the number of cores AND the memory bus in combination. Something you obviously overlooked, but although the number of cores tripled in comparison to the 580, the memory bus was reduced to 256 bits width, a 50% reduction. That means that at best the expected performance gain could be 0.5 (memory bus) x 3 (number of cores) = 1.5 and not your 300%.

     

    To summarize:

     

    Now if you want to go the other way, with the 680 as the baseline, you'd say that for 60% less money, you still get 80% of the performance.

     

    For 55% less money, you get 44% of the performance on this test.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 5:05 PM   in reply to lasvideo

    The Quadro that will be introduced next is likely based on the GTX 5xx, probably with VRAM in the range of 4 - 6 GB, and I don't know about naming conventions, but imagine something like this:

     

    1. Quadro 7000, 4GB, 384 cores, $ 1200,

    2. Quadro 8000, 5GB, 448 cores, $ 2000,

    3. Quadro 9000, 6GB, 512 cores, $ 4000.

     

    They will still be underspecced, underpowered and overpriced.

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 6:04 PM   in reply to Harm Millaard

    Yes, Harm. It was me who finally declared that the performance of the plain non-Ti GTX 560 (336 cores) is commensurate with its cost. It, like you said, costs 55% less - but also performs at only 44% (relative to the GTX 680). There are few cheaper cards than the GTX 680 whose performance is faster than its cost implies (the GTX 560 Ti 448 is one such card).

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 8:19 PM   in reply to RjL190365

    well the whole "value" question is determined by more than one variable. Since I'm coming back to PC from Mac, I need a new card so whats the best bang for the buck?

     

    If I had a gtx 580 already , that would be different.

     

    If I was satisfied with my current card, there would need to be a very good reason to upgrade, not only because a new cards has come.out.

     

    Only a 20% increasr I hear? Thats a lot. I work in news and that could be enough time to make or break a clip.making it to air. I would pay $500 for that performance increase without question.

     

    The quadro cards simply dont offer the value, as this is where I started looking. The consumer gaming cards like the GTX were not in the running until I saw how they outperformed the quadro boards.

     

    So if we are talking $200-$500 to make yourself more productive, its certainly.money well spent if you livelyhood depends on it

     

    Chris

     
    |
    Mark as:
  • Currently Being Moderated
    Mar 31, 2012 8:33 PM   in reply to Jim Simon

    Does anyone know if Adobe and Nvidia got together to allow Premiere Pro 6.0's Mercury Playback Engine to support say, four 680's in SLI?  In other words will the new version of Premiere Pro support Mercury compatible

    cards in SLI?

     
    |
    Mark as:
  • Currently Being Moderated
    Apr 1, 2012 4:56 AM   in reply to Harm Millaard

    Thanks Harm,

     

    “Encoding is purely CPU dependent and that is why you do not see any benefit with the H.264-BR test, because there is no scaling involved.” Harm

     

    But there is a benefit in the H.264 test as the score scales from 194 to 57 as the CUDA cores scale from 32 to 448. The scaling drops off steeply at 192 cores though or possibly even earlier but it’s not possible to say as the graph has no data points between 192 and 32 cores.

     

    “Hardware acceleration is useful with rendering, previewing, and on certain parts of the export process, i.e. scaling, frame rate adjustments, blurring and blending, but not with encoding. If you start with AVCHD and only end up with BRD, then it makes no sense to get this 680 video card for the encoding process.” Harm

     

    For the type of workflow that I outlined (AVCHD/BRD) I presume there are still benefits to using the MPE in certain areas of the process so what class of nVidia card is recommended in this case?

    My system is constrained by my desire for silent computing so I will use a 3rd party GPU heatsink/fan to achieve the best performance whilst keeping the noise in check.

    I imagine I am limited to roughly 100 to 125W TDP for the GPU so am trying to determine the best card to use within this constraint.

    The current favourite is a GTX 560 Ti 448 which I would need to down clock and even under volt if possible to keep the power consumption in check.

    But if my workflow doesn’t benefit from more than 192 cores I might as well stick with a GTX 550 Ti.

     
    |
    Mark as:
  • Currently Being Moderated
    Apr 1, 2012 8:14 AM   in reply to eclipse_crow

    Here's a possible reason for the slow or bottlenecked H.264 encoding performance with NVIDIA GPUs with fewer than 192 CUDA cores:

     

    The cheaper cards employ TurboCache (using the slower system RAM as part of the video RAM) much more frequently than the higher-end cards do. This, in turn, causes the PC to utilize the page file much more frequently than it should have for that same amount of installed system RAM since there is much less physical system RAM available with the lower-end cards than with the higher-end cards. Of all the 500 series GPUs, the GT GPUs all utilize TurboCache while the GTX GPUs don't. There is much back-and-forth switching between the system RAM and the on-card VRAM in PCs equipped with such cheapo GPUs. This is also the reason how the GT 240 performs about 20 seconds slower than the GTX GPUs in the H.264 test.

     

    Also, what GPU you can use in your system will also depend on the airflow inside your case. If your "quiet" case has such bad airflow (especially on the intake) that you can't use even a regular GTX 560 Ti without overheating, it's time to get a new case.

     
    |
    Mark as:
  • Currently Being Moderated
    Apr 1, 2012 8:44 AM   in reply to RjL190365

    The 9500 GT which is listed in the graph doesn’t use Turbo Cache as far as I can see although I do appreciate the technical point you are making.

    I wasn’t aware that nVidia were still using that in any 5 series cards as I thought it went out with the 6200 and low end 7100/7300 series! Are you sure?

     

    “Also, what GPU you can use in your system will also depend on the airflow inside your case. If your "quiet" case has such bad airflow that you can't use even a regular GTX 560 Ti without overheating, it's time to get a new case.” RjL190365

     

    The issue is that cooling a card with a 170W TDP in SILENCE is not trivial so it’s not something that I am aiming for. I can easily cool such a card in my setup but noise is the constraining factor as I outlined above.

     

    I’m trying to determine how important the CUDA core count is when just working with AVCHD as so far I haven’t seen any data that focuses on that in multiple areas of the workflow. There’s no point in me buying a more powerful card than I need and after market fans/heatsink if there aren’t tangible benefits over a card with a much lower CUDA core count.

     
    |
    Mark as:
  • Currently Being Moderated
    Apr 1, 2012 9:08 AM   in reply to eclipse_crow

    eclipse_crow wrote:

     

    The 9500 GT which is listed in the graph doesn’t use Turbo Cache as far as I can see although I do appreciate the technical point you are making.

    I wasn’t aware that nVidia were still using that in any 5 series cards as I thought it went out with the 6200 and low end 7100/7300 series! Are you sure?

     

    The GT 520 will also suffer in H.264 performance because it has only 48 CUDA cores (and also due to its low-bandwidth, low-throughput 64-bit DDR3 VRAM). That's the only 500 series GPU available in the consumer market below the GT 545 (which is basically a cut version of the GTX 550 Ti).

     

    And if NVIDIA doesn't use TurboCache in its current GPUs, then the only explanation I can offer is that certain operations in Premiere also rely on the bandwidth of the graphics card's onboard VRAM. The very slow H.264 result from the GeForce 9500 GT is partially due to its use of very slow, low-bandwidth DDR3 VRAM on the card (Bill's cards used DDR3, but many 9500 GT cards use even slower DDR2 VRAM).

     
    |
    Mark as:
  • Currently Being Moderated
    Apr 1, 2012 11:18 AM   in reply to RjL190365

    GPU memory bandwidth might well be the issue with CUDA performance along with the core count. I suppose nVidia aim to match the two parameters where possible.

     

    A Kepler replacement for the GTX 550 Ti sounds promising if AVCHD processing doesn’t scale beyond 192 cores. It should consume around 70W for the same performance and would be borderline for not even requiring a PCIe power connector.

     

    I’m assuming TurboCache is dead as if a similar technology is present in a card it is highly likely that a company’s legal department would insist that any marketing info made it clear if 1GB of RAM includes shared main memory.

     
    |
    Mark as:
  • Currently Being Moderated
    Apr 2, 2012 8:07 AM   in reply to eclipse_crow

    Hi guys!

     

    I read all the conversations about GTX680 here and other topics, but Im not smarter.Is it a good time to replace my old gtx 470?The GTX 680 is worth it? Or better to wait until Q4 2012? The price of GTX680 is my top limit. Maybe the future cards will be more expensive? There will be GTX700 series in 2012? What do you think?

    When will be any PPBM with gtx680?

    I have x79 platform with SB 3930K.I would investing a vga card for years, so I can wait a few months.But If you say there is no anything sure about nvidia roadmap maybe I purchase now.

    Thanks.

     
    |
    Mark as:
  • Currently Being Moderated
    Apr 7, 2012 7:51 AM   in reply to RjL190365

    I would like to add to my previous statement that the faster cards really need fast expensive CPUs or heaviliy overclocked CPUs in order to see the full advantage of them (as Harm stated many times). I tested my GTX 560 Ti 448 on my main system with the CPU at its stock (or default) clock speeds, and discovered that the MPEG-2 DVD score went from 69 seconds to a whopping 116 seconds. (By comparison, my old GTX 470 went from about 127 seconds with the CPU at default clocks to only 107 seconds with the CPU at the exact same overclocked 4.6GHz speed as the 69-second result from the GTX 560 Ti was attained from.)

     
    |
    Mark as:

More Like This

  • Retrieving data ...

Bookmarked By (0)

Answers + Points = Status

  • 10 points awarded for Correct Answers
  • 5 points awarded for Helpful Answers
  • 10,000+ points
  • 1,001-10,000 points
  • 501-1,000 points
  • 5-500 points