Effect of SLI, PhysX and Nvidia 3D settings on CUDA-based scientific computing performance

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hi,

I have 2 GPU’s (both GT 755M) on my laptop and I use them for scientific computing. Do SLI configuration settings (options: maximize 3D performance / Disable SLI), PhysX settings (options: gpu1, gpu2, auto, cpu) or 3D settings such as

  • Ambient Occlusion
  • Anisotropic filtering
  • Antialiasing stuff
  • Texture fltering stuff
  • Triple buffering and many more
    have any effect of GPU performance using CUDA applications? Should I disable/adjust some of these?
#2

SLI and the 3D settings are purely a graphics thing, they have no bearing on CUDA. PhysX uses CUDA under the hood, but unless you plan to run PhysX-enabled apps at the same time you run CUDA-enabled apps, there should be no connection.

Note that unless your CUDA apps are specifically written with multi-GPU support they will only use one of two GPUs in the system.

#3

Having SLI mode enabled can have slight behavioral impacts on CUDA programs, although I would agree with njuffa that SLI functionality is essentially orthogonal to CUDA functionality.

For more details, refer to the programming guide:

http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#sli-interoperability