2 Replies Latest reply on Feb 21, 2008 8:17 AM by bayoumi

    Using same HD3870 for both Stream Computing & Display output

      What is the penalty for using the same GPU card such as HD3870 for both Stream Computing (Such as ACML matrix operations), and still use its output for LCD monitor control (regular apps, not games for display output).

      The question arises when you want to use a PCI-E Gen2 /HT3.0 for a motherboard such as 790FX chipset. If the penalty is high, then the second PCI-E G2 16X slot is useless when using two parallel card for computing (the second will be used for normal desktop LCD display).

      Again, a one PCI-E G2 x16 slot on the 770 chipset makes it impossible to use stream computing on it. The other option is to wait for the RS780?
      Best Regards
        • Using same HD3870 for both Stream Computing & Display output
          Hi Amr,

          I asked our engineers about this and they said that, if you are doing 2D display, it doesn't use 3D resources (unless you are using Vista). The 3D resources are what are repurposed for stream computations. Most of the performance effect from using the card for displaying as well will likely be with the communication over the PCIe bus, however the commands sent for 2D are relatively small. If you were running video decoding through the GPU at the same time, then it will start using more of that bandwidth as well as the DMA engine.

          In our experience, doing even simple 3D or the Folding@Home (which does frequent readbacks from the GPU) don't usually cause much of a slow down in performance on the GPU.

          The stream computations that will be least affected will be the computationally denser kernels that have minimal CPU communication dependencies.