3 Replies Latest reply on Mar 18, 2009 7:33 AM by thesquiff

    4870X2

    thesquiff

      I've been using this card for development in the SDK1.3. I notice v1.4 now officially supports it.

      I'm guessing I may actuallty have only been using only half the card before as I wasn't using multiple contexts. Are multiple contexts still required to use both gpus, or is this done automatically now? I've recompiled and can't see any performance increase.

       

      TIA

        • 4870X2
          thesquiff

          Some hours of fiddling later and I have a program that creates a context on each device and launches the same kernels on the same data simultaneously. I was expecting the overall execution time to stay about the same + a bit of overhead. However the execution time has doubled....?

          Presumably the kernels are not running concurrently as they should be. Do I need two monitors attached to the card or something (I seem to remember something about this mentioned before)?

            • 4870X2
              thesquiff

              Any ideas anyone? I'm about to smash my head through the monitor...

              Thanks

                • 4870X2
                  thesquiff

                  May wasted hours later, I think I've found my problem. I needed to call calCtxIsEventDone() after calling the kernel for each GPU to allow the concurrency to occur. Seems like a messy trick - perhaps the multiGPU paragraph in the user guide could be expanded to mention this?