8 Replies Latest reply: Jul 29, 2012 10:10 PM by shooternz RSS

    Tesla card in Maximus not being used

    philip_38 Community Member

      I spent a few thousand dollars buying a Quadro 6000 and a Tesla M2090, but the later inever gets used at any time by Premiere CS6. I tested the usage with the command below and the GPU and memory is always zero, when playing back, when exporting, etc. The card is simply ignored.

      Is there something I am missing? I am exporting a 45 minute video for Youtube only with acceleted effects and it takes about 40 minute to generate the file. Sould it not take much less time?

      Philip

       

       

      C:\Users\Administrator>nvidia-smi -i 0 -q -d MEMORY,UTILIZATION,POWER,CLOCK,COMPUTE

       

       

      ==============NVSMI LOG==============

       

       

      Timestamp                       : Sun Jul 29 17:12:42 2012

       

       

       

      Driver Version                  : 297.03

       

       

      Attached GPUs                   : 2

       

       

      GPU 0000:05:00.0

          Memory Usage

              Total                   : 5375 MB

              Used                    : 9 MB

              Free                    : 5366 MB

          Compute Mode                : Default

          Utilization

              Gpu                     : 0 %

              Memory                  : 0 %

          Power Readings

              Power Management        : N/A

              Power Draw              : N/A

              Power Limit             : N/A

          Clocks

              Graphics                : 50 MHz

              SM                      : 101 MHz

              Memory                  : 135 MHz

          Max Clocks

              Graphics                : 650 MHz

              SM                      : 1301 MHz

              Memory                  : 1848 MHz

       

      my configuration:

      C:\Users\Administrator>nvidia-smi -q

       

       

      ==============NVSMI LOG==============

       

       

      Timestamp                       : Sun Jul 29 17:18:05 2012

       

       

      Driver Version                  : 297.03

       

       

      Attached GPUs                   : 2

       

       

      GPU 0000:05:00.0

          Product Name                : Tesla M2090

          Display Mode                : Disabled

          Persistence Mode            : N/A

          Driver Model

              Current                 : TCC

              Pending                 : TCC

          Serial Number               : 1.0203E+12

          GPU UUID                    : GPU-fde0f3fc-6d39-30fe-d1c4-274c8d289af0

          VBIOS Version               : 70.10.4E.00.00

          Inforom Version

              OEM Object              : 1.0

              ECC Object              : 2.0

              Power Management Object : 3.1

          PCI

              Bus                     : 0x05

              Device                  : 0x00

              Domain                  : 0x0000

              Device Id               : 0x109110DE

              Bus Id                  : 0000:05:00.0

              Sub System Id           : 0x088710DE

              GPU Link Info

                  PCIe Generation

                      Max             : 2

                      Current         : 1

                  Link Width

                      Max             : 16x

                      Current         : 16x

          Fan Speed                   : N/A

          Performance State           : P12

          Memory Usage

              Total                   : 5375 MB

              Used                    : 9 MB

              Free                    : 5366 MB

          Compute Mode                : Default

          Utilization

              Gpu                     : 0 %

              Memory                  : 0 %

          Ecc Mode

              Current                 : Enabled

              Pending                 : Enabled

          ECC Errors

              Volatile

                  Single Bit

                      Device Memory   : 0

                      Register File   : 0

                      L1 Cache        : 0

                      L2 Cache        : 0

                      Total           : 0

                  Double Bit

                      Device Memory   : 0

                      Register File   : 0

                      L1 Cache        : 0

                      L2 Cache        : 0

                      Total           : 0

              Aggregate

                  Single Bit

                      Device Memory   : 0

                      Register File   : 0

                      L1 Cache        : 0

                      L2 Cache        : 0

                      Total           : 0

                  Double Bit

                      Device Memory   : 0

                      Register File   : 0

                      L1 Cache        : 0

                      L2 Cache        : 0

                      Total           : 0

          Temperature

              Gpu                     : N/A

          Power Readings

              Power Management        : N/A

              Power Draw              : N/A

              Power Limit             : N/A

          Clocks

              Graphics                : 50 MHz

              SM                      : 101 MHz

              Memory                  : 135 MHz

          Max Clocks

              Graphics                : 650 MHz

              SM                      : 1301 MHz

              Memory                  : 1848 MHz

          Compute Processes           : None

       

       

      GPU 0000:03:00.0

          Product Name                : Quadro 6000

          Display Mode                : Enabled

          Persistence Mode            : N/A

          Driver Model

              Current                 : WDDM

              Pending                 : WDDM

          Serial Number               : N/A

          GPU UUID                    : GPU-8ab57ebc-584d-5d9d-330a-c689df499b66

          VBIOS Version               : 70.00.57.00.03

          Inforom Version

              OEM Object              : N/A

              ECC Object              : 1.0

              Power Management Object : N/A

          PCI

              Bus                     : 0x03

              Device                  : 0x00

              Domain                  : 0x0000

              Device Id               : 0x06D810DE

              Bus Id                  : 0000:03:00.0

              Sub System Id           : 0x076F103C

              GPU Link Info

                  PCIe Generation

                      Max             : 2

                      Current         : 2

                  Link Width

                      Max             : 16x

                      Current         : 8x

          Fan Speed                   : 35 %

          Performance State           : P0

          Memory Usage

              Total                   : 3071 MB

              Used                    : 3041 MB

              Free                    : 29 MB

          Compute Mode                : Default

          Utilization

              Gpu                     : 0 %

              Memory                  : 0 %

          Ecc Mode

              Current                 : Disabled

              Pending                 : Disabled

          ECC Errors

              Volatile

                  Single Bit

                      Device Memory   : N/A

                      Register File   : N/A

                      L1 Cache        : N/A

                      L2 Cache        : N/A

                      Total           : N/A

                  Double Bit

                      Device Memory   : N/A

                      Register File   : N/A

                      L1 Cache        : N/A

                      L2 Cache        : N/A

                      Total           : N/A

              Aggregate

                  Single Bit

                      Device Memory   : N/A

                      Register File   : N/A

                      L1 Cache        : N/A

                      L2 Cache        : N/A

                      Total           : N/A

                  Double Bit

                      Device Memory   : N/A

                      Register File   : N/A

                      L1 Cache        : N/A

                      L2 Cache        : N/A

                      Total           : N/A

          Temperature

              Gpu                     : 84 C

          Power Readings

              Power Management        : N/A

              Power Draw              : N/A

              Power Limit             : N/A

          Clocks

              Graphics                : 573 MHz

              SM                      : 1147 MHz

              Memory                  : 1494 MHz

          Max Clocks

              Graphics                : 573 MHz

              SM                      : 1147 MHz

              Memory                  : 1494 MHz

          Compute Processes

              Process ID              : 4348

                  Name                : C:\Program Files\Adobe\Adobe Premiere Pro CS6\Adobe Premiere Pro.exe

                  Used GPU Memory     : Not available in WDDM driver model

        • 1. Re: Tesla card in Maximus not being used
          J. Simon Community Member

          Do this.

           

          Turn off hardware acceleration in PP and export again.  If it takes longer, then all is well, hardware acceleration is working properly.  If it still takes the same 40 minutes, then something is wrong.

          • 2. Re: Tesla card in Maximus not being used
            philip_38 Community Member

            The question is diferent: if it takes the same 40 minutes with and without the Tesla, then whay would you spend the money?

            I think that there is a confuguration file somewhere that has only a C2075 in it, and I need to add the M2090.

            Any idea?

            • 3. Re: Tesla card in Maximus not being used
              J. Simon Community Member

              The test will determine if hardware acceleration is actually in use.  I don't trust metering tools to tell the whole story, as I've seen some that suggest no acceleration is occurring when I know for a fact it is.

              • 4. Re: Tesla card in Maximus not being used
                Harm Millaard Community Member

                Tesla M2090, but the later inever gets used at any time by Premiere CS6

                 

                Return it and get a refund. The Tesla M is for servers only and requires Windows Server 2008 as an OS. It is not supported with Windows 7 and will not work even with the 'hack' applied.

                 

                The only Tesla that will work is the C2075 and for the meager amount of an additional € 2300 (that is in addition to the Quadro card) it will give you ZERO, NIL performance improvement over a € 400 GTX card. And that is only in single CPU systems, because Tesla will not work in dual CPU systems, if HP is to be believed.

                 

                Results of Maximus solutions with Quadro 6000 + Tesla C2075 (for only € 6000) are no better than a single GTX 560 Ti for around € 240. The only situation where these €€€ may be well spent is if you need 10 bit output for your € 2000+ each monitors.

                 

                Is there something I am missing? I am exporting a 45 minute video for Youtube only with acceleted effects and it takes about 40 minute to generate the file. Sould it not take much less time?

                 

                Transcoding is purely a CPU matter. There is no acceleration with CUDA enabled cards, apart from scaling, blending and blurring. All the rest is only CPU dependent.

                • 5. Re: Tesla card in Maximus not being used
                  philip_38 Community Member

                  THE TRUTH, THE TRUTH AND NOTHING BUT THE TRUTH ABOUT PREMIERE AND NVIDIA

                  I finally have the picture clear. In order to reach this "illumination" I assembleda laboratory with

                  a) a $25.000 server Dlel R901 with 24 cores at 2.7 Ghz and 128 Gb of RAM and Windows Server 2008 R2 Enterprise

                  b) a Quadro 6000 + Tesla M2090

                  c) a Geforce GTX 285

                  d) a EVGA GeForce GT 640

                  e) a Desktop computer with an I7 980 with 25 GB of RAM and Windows Server 2008 R2 Enterprise

                  f) a Mac 10.7.4 with an 8-core I7 990 at 3.2 Ghz


                   

                  Background: I shoot a 45 minute video every saturday for a non-for-profit institution, as a hobby.  That is why I push this so hard. My goal is to shorten the time it takes from shooting the video and uploading it to Youtube. This involves one video track, one title, and some CUDA accelerated effects like RGB curves , zooming plus some audio enhancement.

                  Some facts: Even the cheapest Nvidia card available, the GT 640, can do realtime CUDA effects, whoever tells you different is trying to defraud you. It does work, at least in my one video-track scenario. The bottleneck is not realtime display, is exporting it to H264. When Premiere is exporting H264 at 1280x720 highest possible quality, it uses the CUDA engine in full. I proved it. It also uses the CPU, but without CUDA, even if pre-rendered, it takes a lot longer to export, several times longer. Some people say that when exporting Premiere does not use the CUDA engine, but they are flat wrong.


                  Quadro 6000 + Tesla M2090, COMPLETELY UNNECESSARY. This is what happens: you would think that having the two most expensive cards in the Nvidia portfolio, the time it takes to export the video would shorten, but it is wrong. The geniouses at Nvidia/ Premiere only use in card at the time, I proved it. If the Tesla is present in the system, they use it above the Quadro, and the Quadro is idle at 0% while the Tesla is running at 90% GPU and it does require cooling or it smells like it is catching fire. All in all, it will not save any time over using the Quadro 6000 alone. However, why would you ever spend the money on either card, when an old GTX 285 can do the same exact job for only 10% more time, I mean going from 40 to 45 minutes? You need to be high on drugs to spend almost $8.000 in hardware for something that a $200 card can do in only 10% more time. It is all a marketing fad. Unless you have 10 video tracks and dozens of effects, no need to spend a dime over $200, even a $100 GT 640 will do it.

                   

                  THE COMPUTER: does it matter? No it does not. My $25.000 server will still export the same video in the same 40 minutes, using the cheapest card, the GT 640, which being half-length, is the only one that I could fit in my server. In my desktop box with an I7 980 and 25 GB of RAM, it takes the exact same time to export the same video, so save the server for a higher purpose.


                  It gets even better. I don't use Windows 7, but Windows server 208 R2 with full desktop composition, which is the same as Windows 7 for any video editing purpose. But I also have a Mac 10.7.4, with 25 GB of RAM, built by me, with a I7 990 (8 cores at 3.2 Ghz). Premiere takes around 10 times more time to export the same exact video with the same exact CUDA effects and card on the Mac, as compared to Windows. A complete catastrophe. It uses the Mercury Engine for realtime, but apparently not for exporting the video to H264, while in Windows it does use it just fine. Something is wrong here, very wrong. This time I think Adobe is to blame. Please be advised that hard drives are not to blame on this difference. I am not an idiot. In both Operating Systems I created a 2 GB RAM disk and exported to this device. So please stop arguing that raid cards and other gimmicks can shorten export times. It is unrelated, since there is nothing faster than a RAM disk. On the Mac I even tried to store the source on the same RAM disk as the exported file, and it did not substantially improve the time it takes to export.

                   

                  Conclusion: if you can find it in Craigslit, buy yourself a GTX 285 used. It is the best bang for the buck, and avoid any expensive Nvidia hardware.

                   

                  If somebody wants to contact me about this, my email is venefax at gmail





                  • 6. Re: Tesla card in Maximus not being used
                    shooternz Community Member

                    I dont think you will have surprised anyone here by your findings.

                     

                    How did you dig yourself such a deep and expensive hole in the first place.?  I am assuming bad advice from an equipment retailer.

                     

                    Fact its a hobby is even more startling to me!

                     

                    BTW - I hope you are not pointing the finger at Adobe or Nvidia for this?

                     

                    Adobe / Nvidia put up the specs and you ran off with the ball seemingly without any foundation or research.

                    • 7. Re: Tesla card in Maximus not being used
                      philip_38 Community Member

                      Actually yes, I point the finger at Adobe support. I called them and thety told me that the Maximus configuration would make Premiere fly. It does not. However, I did not lose any money. I will return the Quadro and the Tesla tomorrow. The rest of the equipment will be used for its original purpose. I wil keep my desktop box (which I have owned for long) with an I7 and a EVGA Classified 3 board, plus one GTX 285, for Premiere.

                      I wish that Premiere Mac would work at the same performance level as the Windows version. I hate to keep two boxes for something that my Mac should do on its own.

                       

                      It is not only  a hobby. I help a political cause, not related to US politics though, but a certain country in Latin America where I was not even born, but that I feel the need to help.

                      • 8. Re: Tesla card in Maximus not being used
                        shooternz Community Member

                        If I can suggest a GTX 670 card (with 4GBs onboard)   plus the GTX285 may easily satisfy you once you lose the rest of the "stuff"

                         

                        Throw the two in together ...using the 285 for GUI (monitoring) and the 670 in for 1334 core  CUDA power (no monitors).

                         

                        IMHO : Best Adobe support will be found in this forum - including the Hardware Forum.

                         

                        We have Users plus the Adobe lead guys playing around here.

                         

                        These guys listen and contribute.

                         

                        Adobe Support is not highly regarded here at a solution level.