Parallel Nsight 1.5 and Visual Studio 2010 Cuda development

[Bill]

I'm going to work on a CUDA program this semester for my graduate level parallel programming class at UAH.

I finally got debugging running on my GTX 275 with Parallel Nsight 1.5 RC and Visual Studio Professional 2010.

I can run the desktop on one GPU, while running debug code on the other. (no monitor)

The speedup that I should be able to get is pretty big for floating point operations.

Nvidia explains it best. (excerpt from CUDA programming guide)

Here is a picture of my setup as it is right now. Local debugging like this requires a dual GPU system. The GTX 275 is halted in the picture on the hardware break point.

Before you ask running the debugger requires Aero to be turned off.

I'll update you guys on my experiences with this. Maybe I can make some performance charts.

To get the debugger to actually run locally like this, you need to do these things. (In Vista/7)

Install Visual Studio 2010 (and install VS2008 SP1 or have the compiler installed from the windows SDK)

Install Parallel Nsight 1.5 monitor and host. (either the 32 or 64 bit version) Right now you need to be in the beta program.

Install the required developer driver or newer. (in this case 260.61)

Setup your monitor on the GPU you aren't going to debug on. In this case I am running my desktop on the 9300, and the code on the GTX card.

Set PhysX in the Nvidia control panel to run on the debug card. (GTX card in this case)

Disable Aero

Run the Nsight monitor program, and turn off WDDM TDR. (set "WDDM TDR enabled" to false)

Run "NVIDIA Parallel Nsight 1.5\Common\DisableWpfHardwareAcceleration.reg"

Reboot your machine

Launch Visual Studio 2010/2008SP1 and load up the Cuda example project (matrixMul_vc100.vcxproj for VS 2010)

Build and start CUDA debugging from the Nsight drop down menu. Make sure you read the Nsight Users guide.

[Kamika007z]

Very cool bill!!

I'm excited to hear more about this! :)