NVIDIA® Nsight™ Development Platform, Visual Studio Edition 2.2 User Guide
On a computer that has two GPUs, you can designate one GPU as the "headless GPU." In this case, the term "headless" refers to a configuration in which the additional GPU does not send display information to a display monitor.
The reason that you may choose to use headless GPU debugging is because the Windows operating system normally requires a desktop to be assigned to every usable GPU. This can cause problems when trying to debug a CUDA-based application locally. If a desktop is running on the same GPU that is running a CUDA-based application that is being debugged, the CUDA Debugger can have problems communicating with the application that is being debugged. The CUDA Debugger disables breakpoints automatically when attempting to debug locally on a GPU that has displays attached.
On Windows Vista and Windows 7, it's recommended that users run their CUDA applications on a headless GPU.
The following procedure allows you to work around the desktop requirement.
To run headless:
|For a local debugging configuration, make sure that you first disable D3D acceleration for WPF. For assistance, see Setup Local Debugging.
|NVIDIA® Nsight™ Development Platform, Visual Studio Edition User Guide Rev. 2.2.120522 ©2009-2012. NVIDIA Corporation. All Rights Reserved.|