Dose the GTX660Ti card absolutely doesn't work with Nsight 3.0?

The project I am involving is new to me for GPU implementation for an existing Windows app. I have done a lot trying to setup a platform such as: Window 64-bit app with MS Visual Studio 2008 + Cuda 5.0 (latest)+ Nsight 3.0(latest) so can debug device/kernel code. As I realized only Nsight can help me to debug device code.

When I upgrade from Cuda 4.2 to Cuda 5.0, I also purchased a new graphic card from Cuda 5.0 supported list which is GTX660Ti. It works fine with Cuda 5.0 in my project if not debug device code. But after I installed Nsight 3.0, I found it is not in its support list, nor in Nsight 2.2 list, Just need to confirm this card is no good for Nsight GPU? If yes I need to replace it by another card exactly under Nsight GPU list. So far target on GTX580, 570 etc. 560, etc which are good for both Nsight 2.2 and 3.0 for sure.

Just curious on what happen on GTX660Ti card which is not supported for Nsight 3.0. I runs a sample project on it with Nsight. When I start the Nsight debug from Visual Studio by put a break point in the kernel code (in a .cu file) the computer was blacked or crashed! Have to reboot. If I don’t insert any break point in the kernel code, it still runs fine like Nsight debug is not involved. Is this because this card/GPU is not supported or should I at least get some warring or alert message before crash…

Chester

Hi Chester,

GTX 660 Ti is officially supported by Nsight 3.0, and it should work on Nsight 2.2

Are you doing a local debugging on that GTX 660 Ti? Meaning, you have built on the same machine as you are Nsight debugging on? And this GTX 660Ti is the GPU driving the windows display?

I would recommend checking the driver installed on the machine. It should match what the Nsight and CUDA toolkit requires.

Thanks

Hi Rafi,

If GTX660Ti is officially support by Nsignt 3.0, why it is not in the list of its documment? neighter not in Nsight 2.2 list? Please confirm this becuase i am on the way to purchase a new card (GTX560) to replace this one.

I am useing local debug, and the installation is doing fine following your instruction with the latest displaying driver came with Nsight 3.0 downloaded.

Thanks.

Rafi,

This is extra information, the GTX66oTi requires 450w power so there is a booster power module installed in my PC to supply power to it. it works fine in my Windows7 with all other applications.

Can you explain the symptoms in my previous message (crash)? first of all if can, lead me to a right local debug process I just followed with the installed Nsight 3.0 User Guide…

Thanks,
Chetser

Hi Chester,

The list of supported GPUs is updated at a less frequent pace than our development. The GTX 660Ti should work in Nsight 2.2 and is officially supported in 3.0.

Just curious, have you tried debugging one of the CUDA samples? Also, what are your CUDA Debugger preemption settings set to? You may find these under Nsight Menu–>Options–>CUDA.

-Jeff

Hi Jeff,

I believe you would receive the email with my long documented testing case I submitted. To answer your question here, the CUDA debugger premption in the sample project is defaut to “Prefer No Software Preemption”. Shoud I change to “Prefer Software Preemption”? what other options/settings under this page should change? Where to refer those settings as the User Guild does not show everything.

Chester

Jeff,

I tried set the premption to “Prefer Software Preemption”, it has the same problem when I insert a break point to debug.

Chester

I found your email and sent you a response. :)

-Jeff

Jeff, believe you received my individual email regarding my testing case. Wish you can find a time to take a look on it. If you still prefer I post individual question here so I can get a quick answer or support I will do so to move on. Here are some simple questionss I copied from my email so you may have simple answer on them.

  1. Besides what I learnt from User Guild, one thing I am still not very sure on what is the right way to debug host and device both? I knew I can debug host code without Nsight. I suppose to debug device code only with Nsight “Start DUDA debugging”? So what about I need to step in both break points?

  2. If this doesn’t work for any reason, is possible something still related to Nsight 3.0 which is still under Beta or so called preview release? So roll back to its 2.2 version would be a good test though of course I have some works to do?

Thanks.

Also found a similar thread posted under GEForce forums (why have seperate forum site?) by Patriot Infinity, copied here, wonder if this could be the similar problem of Nvidia card/driver?


GTX 460 (285.62) Black Screen / Crashing

Yes same problem here with GeForce GTX 460 + 285.62WHQL Drivers

Am on Win 7 - 64 Bit

I’ve never had to restart my system, When it crashes, the Screen goes black for bout 4 seconds after which it recovers

I’ve seen this Crash happen even when i just have a browser and chat window open

Nvidia needs to come with fixed drivers atleast this year !! /wallbash.gif’ class=‘bbc_emoticon’ alt=’:wallbash:’ />


I also found from other forums that that a gtx660 PC won’t boot. It is kind related to the BIOS and PCI version. Can you tell if this could related to my problem? or what is the requirement on BIOS/PCI for this card?

Chester

Jeff,
Further test and debug found the blackout/crash only happed on one of the sample project of Nsight 3.0 which is matrixMul. I tested others in the same CUDA solution file at least not crash. I also tested the sample projects came from the CUDA 5.0 toolkit, they are not crash either. So I can put that crash case behind at this time to move on other paths.

The problem now is how to debug with Nsight VS. So far for all projects (either from Nvidia or my own) after I insert break points in the device code, after I start either VS debug or Nsight debug, it disabled so never hit it. So there is still two possibilities 1.) There is still a problem in my Nsight 3.0 not fully function. 2.) what is the right procedure to debug device code .

Here is what I did,

  1. Insert break points in both host and device code, if start VS debug only, of course it only hit the host break points and disable the break points in the device code and skip the kernel function as expected, that is understood.
  2. Insert break points in both host and device code, if start Nsight CUDA debug, I expect all the break points in the device code still enabled and will hit but unfortunately they are not because they are disabled (grey out) after starting Nsight debug.
    When start the debug, the Nsight shows the message “Connect [computer name] to [computer name]” that means the Nsight Monitor is started. Also closed with message “Disconnected from [computer name]…” when I close the debug.
  3. When I start the VS debug and break at the line of entering the device code, I start Nsight debug, then try to step into the device code, Microsoft message shows “The debugger can’t continue running the process… ”.

I read the User Guide and thought I did right debug procedure. With all the information I provided, please review and help me in good time manner. Or please tell which document gives more details to follow.

Thanks,
Chester

Since this topic is about 660 model, I would like to state that I am also having problems, but with a 660M, the application crashes on launch when I’m doing a Graphics Debugging. I’m using NSight 2.2 and with a few applications the Monitor says “This device is not using a supported NVIDIA GPU”.

Which release of 2.2 are you using and which driver?

Visual Studio 2010
NVIDIA NSight 2.2.0.12313
Driver Version 306.97 and 310.70
Windows 8 x64

The GTX is using Optimus with a Intel HD 4000. I’ve read about incompatibility between Optimus and NSight but the Official requirements says that at least Graphics Analyzer is supported.

The culprit could be Windows 8. It’s not officially supported yet. For what it’s worth, the GTX 660 and GTX 660M are different chips.