3 Replies Latest reply on Nov 6, 2011 3:18 AM by Flying_Kiwi

    Integrated Graphics + Dedicated Graphics for CUDA

    marbre

      Hi,

      I have an Intel DQ57TM Desktop Board, which I use together with an Intel Core i5-650. I am absolutely glad with the integrated graphics, but I had to install a dedicated graphic card to use CUDA.

       

      Is is possible to use the integrated graphics for video/display, but also to activate the dedicated graphic card to run CUDA applications on it? By default, it seems that if not chosen as primary device, the IGP is disabled if an additional video card is installed. If I choose the IGP for video, I am not able to see my nVidia card neither in the Windows devicemanager nor under Linux with 'lspci'.

       

      Does anybody has an idea how can use both?

      Best Regards

       

      Marius

        • 1. Re: Integrated Graphics + Dedicated Graphics for CUDA
          Flying_Kiwi

          From the top of page 16 in the product user guide for this board http://www.intel.com/support/motherboards/desktop/dq57tm/sb/CS-030950.htm

          "The board supports either integrated graphics (Intel Graphics Technology) or PCI Express 2.0 x16 graphics." ie one or the other but not both at the same time. This is common with other Intel boards so I doubt it'll change (or that it's even possible to change).

           

          Why not just ask your provider of CUDA compatible software for a version that'll run on your Intel built in graphics and then all will be fine (and it'll save you some money on extra graphics cards surplus to your needs - you said you were happy with the Intel graphics otherwise)?

          • 2. Re: Integrated Graphics + Dedicated Graphics for CUDA
            marbre

            Well, yes and now. First of all thanks a lot for your answer.

             

            I found the following on the Z68 chipset (http://www.intel.com/content/dam/doc/product-brief/z68-experss-chipset-brief.pdf):

            "Intel Z68 Express Chipset-based platforms provide support for third-party switchable graphics solutions, allowing end users to take advantage of the world-class media features delivered by 2nd generation Intel Core processors along with the gaming capabilities of discrete graphics products for the most demanding games."

            But if I take a look for example to on of the Z68 boards as DZ68DB (http://downloadmirror.intel.com/19926/eng/DZ68DB_ProductGuide01_English.pdf) the Product Guid Tells:

            "The board supports system graphics through either Intel HD Graphics or a PCI Express 2.0 x16 add-in graphics card."

             

            Thats why I am not realy sure if hybrid graphics were first introduced with Z68. I would be happy if I could use the Intel HD graphics, without the board disabling the the PCIe x16 slot. I don't want to use the nVidia as graphic device, but I cannot even see it at the Windows devicemanager or under Linux with 'lspci'.

             

             

            Running CUDA applications on other hardware than nVidia isn't possible, as CUDA is a parallel computing architecture developed by nVidia and only integrated on nVidia devices. Right now, OpenCL (for which also Intel realeased it's API this year) isn't a alternative for me.

            • 3. Re: Integrated Graphics + Dedicated Graphics for CUDA
              Flying_Kiwi

              marbre wrote:

               

              Well, yes and now. First of all thanks a lot for your answer.

               

              I found the following on the Z68 chipset (http://www.intel.com/content/dam/doc/product-brief/z68-experss-chipset-brief.pdf):

              "Intel Z68 Express Chipset-based platforms provide support for third-party switchable graphics solutions,


              I see what you're getting at now. The HP Touchsmart II PC that an employer provides has this functionality where you can use the higher perfoming ATI graphics for graphically demanding tasks or leave it on the Intel Graphics when this isn't needed and you want to conserve battery power (you can't use both at once though). That device doesn't have a Z68 chipset and the particular model I'm using, has been around since early 2010 so this is nothing new.

               

              As your motherboard isn't in a laptop or tablet PC, there isn't the same attention to battery power conservation so there's therefore no NEED to have switchable graphics. Therefore what the user guide (and document I earlier referred to) mention about it being an 'either / or' situation will stand.

               

              By all means, fire off an email to Intels marketing people / Customer Services asking them to consider a switchable graphics motherboard for the future but I imagine it'd be very unlikely that they'd be able to implement this due to the fact they can't control what discrete graphics card users would choose to fit. At least with a laptop, it's all able to be designed in by the manufacturer to work together in a controlled environment.

               

              I realise that your CUDA apps won't run on other than proprietary nVidia hardware but my point is if you approach the app publisher and ask for a version of the same app that'll work with 'regular' graphics solutions, they may be able to help (they may even already have such an app available). Sure the app may not have the same bells and whistles or run as fast but there may still be something available which meets your needs and enables usage of more general purpose Intel graphics hardware.