3 Replies Latest reply on Jul 10, 2012 10:54 PM by Mylenium

    Dual GPU Cards


      I have 2 GEforce GTX 680's installed with 2 GB of VRAM each.  In the Adobe docs, it says....


      "If you have multiple GPUs installed, the GPU-accelerated ray-traced 3D renderer will use the CUDA cores on all of them, as long as they are of the same CUDA compute level. (See the technical specifications of your GPU for its CUDA compute level.) After Effects will also use all of the VRAM on the installed GPUs, with the caveat that both cards will be treated as if they each have the amount of VRAM on the card with the lesser amount of VRAM."


      But in the After Effects preferences, it sees the 2 cards but only saying 2 GB of RAM is available.  Anyone have any idea?  I tried this with and without SLI being enabled in the NVIDIA control panel.

        • 1. Re: Dual GPU Cards
          BenjaminMarkus Level 4

          We need to know a lot more to help you. Please provide answers to the questions listed here: "FAQ: What information should I provide when asking a question on this forum?"

          • 2. Re: Dual GPU Cards
            bigskyedit1 Level 1

            Sure.  Sorry about leaving possible helpful information.

            I'm using CS6 latest update.

            Windows 7 latest updates.

            Everything is working fine, Im just curious as to why my GPU information is only showing 2GB available when it should be 4GB.  And if there is something that I should be doing to let After Effects use all available memory.  Here is what is copied from my GPU information:


            Fast Draft:Available
            Texture Memory:1024.00 MB




            Vendor:NVIDIA Corporation
            Device:GeForce GTX 680/PCIe/SSE2
            Total Memory:1.95 GB
            Shader Model:4.0 or later


            Driver Version:4.2
            Devices:2 (GeForce GTX 680, GeForce GTX 680)
            Current Usable Memory:1.79 GB (at application launch)
            Maximum Usable Memory:2.00 GB




            • 3. Re: Dual GPU Cards
              Mylenium Most Valuable Participant

              I think you misunderstand how this works. SLI doesn't double the memory as effectively each card still uses its own buffers to do it's thing and the same is true for most CUDA/OptiX operations. Each card still has its own processing to do, those frameworks are merely a way of coordinating them. So in effect there may be the same redundant data loaded on each card because it is required to achieve the performance. Hence, since this is doesn't do anything that would allow you to laod more, the driver returns the value as if this were just one card and handles the rest by himself. That's also excplained somewhere on the NVidia pages, BTW.