Installation nightmares Can't compile SDK samples

I’ve been round and round in circles trying to work out what I’m supposed to download and install to get CUDA working on my machine. I have a Dell XPS M1530 laptop, running Windows Vista (32bit), and with a GeForce 8600M GT, and my compiler is MSVC 9.0.

Initially, when I tried to compile SDK code samples, the compiler said something about not being able to find nvcc.exe. I forget the exact message and I didn’t write it down, but it smelled of environment variables not being properly set, or something along those lines, so I decided to try reinstalling.

The CUDA 2.1 driver installer (181.20) popped up an error saying “The NVIDIA Setup program could not locate any drivers that are compatible with your current hardware. Setup will now exit.”, so I think okay, perhaps my system won’t support 2.1, so I should try 2.0.

The CUDA 2.0 driver installer (178.08) said “The drivers you are installing are older than the drivers currently installed on your system. Do you wish to continue and install the older drivers?”. That seemed a bit suspect to me, but there are worse things than not-brand-new drivers, so I agreed to install the older ones. Not a good idea, my monitor reset itself to 800 * 600 display and refused to be sized back up to anything sensible. I went to the NVIDIA site to get the latest drivers again to fix the problem (it sent me to the Dell site to get drivers), installed those, restarted, and everything seemed fine. I assumed at this point that I had “some” version of the CUDA drivers, although I didn’t know if it was the 2.0 or the 2.1, so I tried installing the 2.0 toolkit and SDK.

When I tried to build one of the code samples (fluidsGL), I got the following compiler error:

So I guessed I needed 2.1 after all. Installed the toolkit and SDK over the top of the 2.0 versions (the 2.1 installers spotted and uninstalled the 2.0 versions first). I get the following output before Visual Studio just hangs and needs to have its process killed via the Task Manager.

What do I need to do to get these SDK samples compiling and running on my machine? Have I missed something in terms of trying to work out which version of the software I need to install, or is the install process this confusing for everyone?


Even if I’m the only person having this problem, can someone give me some idea of how long I should expect compilation of one of the SDK samples to take (are we talking seconds here, or hours?), and if there’s anything I can do to make nvcc be a little bit more verbose in what it’s doing so I can get closer to diagnose why it’s hanging?

Any help at all is appreciated, I’ve been plugging away at this for days now and getting nowhere. I’m out of ideas.

If you’re using VS2008, you need CUDA 2.1.

There were some problems a while back where there were differences between the desktop drivers (provided by nVidia) and laptop drivers (provided by your laptop vendor). I think they were solved, but I’d search around to make sure that Dell is giving you the proper 2.1 drivers if that’s the toolkit and SDK you’ve got installed.

The error you’re getting about not being able to write to the build log (in the second output message)…if you’ve got weird permissions set up on your machine where the build log can’t be written, maybe that’s causing things to hang? It’s hard to say without more info…

The compilation should only take a few seconds…most of the projects are pretty basic examples of what you can do with CUDA.

If you’re using VS2008 Express, that may be the problem itself…CUDA 2.1 was the first release to add support for VS2008, and while it works fine with the full versions of VS, it seems there are some differences in the express version which are causing some problems. I don’t know what nVidia is planning to do about that, if anything.

I’d try uninstalling the toolkit, SDK, and Visual Studio (if you can), then re-install the newest drivers, VS2008, the 2.1 toolkit, then the 2.1 SDK. Note that after installing all that, you need to build a couple of small runtime libraries that are only used by the SDK examples (if you intend to compile them, that is). I don’t remember the names of them off the top of my head, but I wrote about it in an older post once (in this Vista forum, I believe), so if you search for it, there are fairly complete directions on how to do that.

Thanks for the reply. I’m using VS2008 (full, not Express), and it became obvious that I needed CUDA 2.1. My graphics cards drivers are what NVidia seems to have recommended, namely that I download an install something called R200183.EXE from Dell, and then install this from NVidia: 179.48_notebook_winvista_32bit_beta.exe . So it goes like:

R200183.EXE (Dell prerequisites) - Fine
179.48_notebook_winvista_32bit_beta.exe (NVidia recommended latest driver for the GeForce 8600M GT) - Fine
181.20_geforce_winvista_32bit_english_whql.exe (the CUDA 2.1 driver) - Not fine, says “The NVIDIA Setup program could not locate any drivers that are compatible with your current hardware. Setup will now exit.”… I don’t know what to do except presume that the 179.48 drivers will do the job okay.
CudaSetup-2.1-win32.exe (Cuda toolkit) - Fine
NVIDIA_SDK10_CUDA_2.10.1215.2015.exe - Fine

Is it some combination of having the 8600M GT and VS2008 that’s the problem? Do I need an older compiler and/or a newer graphics card? Seems like something of an oversight if I do, since both things are ostensibly supported.

With regards to the inability to open the build log, I’m not sure what I can do to get/give more information. It’s not an error/warning I’ve experienced before. I did notice that VS2008 seemed unable to save over .vcproj files when I tried to change the project settings, even though none of the files involved in the sample code appear to have been flagged as Read-Only. I thought maybe that it was some weird side-effect of the SDK Browser shrug. I googled PRJ0009 and found this: which claims to relate to VS2008, but it wasn’t much help - When I go to Tools->Options… I don’t have a “Projects” folder (I’ve got a “Projects and Solutions”, but that doesn’t have a “VC++ Build” option), so I don’t know how I can find out where it’s trying to write BuildLog.htm

No, you should be fine with the 8600GT and VS2008, but since the 2.1 drivers aren’t working, I think you might be out of luck until some new drivers are released that do work. The only other option that I can think of would be for you to downgrade to a slightly older driver (whatever the last CUDA 2.0 release was) + the CUDA 2.0 tools + Visual Studio 2005, if you have access to it.

Sorry to hear that you’re having such problems with CUDA…it’s really just that the drivers for laptops (in general) are a bit funky. On the desktop, you could probably be up and running with your exact same setup in a matter of minutes.



    install step VS2008 -> CUDA2.1 ->CUDA Toolkit 2.1->CudaSDK 2.1
    Update Vga card  "LABTOP MOD DRIVER" google search  for your vga (it support)

    befor run project   
    bulid project  "C:\NVIDIA CUDA SDK\common\Release_vc90.sln"  important  then  "run sample code"
    or test  VGA with -> rendercheckgl_vc90.vcproj       

    read  Nvidia guide paper (install)
    sorry i can't speak English good <img src='http://hqnveipbwb20/public/style_emoticons/<#EMO_DIR#>/thumbup.gif' class='bbc_emoticon' alt=':thumbup:' />  i think you understand;

    try it

    if you use matlab  "nvmex"  install platform SDK too [/b]

I’ve bought a desktop which was basically custom-designed to be able to run CUDA (basically a slightly better setup than the laptop hardware I have now, but something which should have less wacky driver problems), and I hope that’ll work better for me. We’ll see how it goes when it arrives, but if what I’ve been led to believe about my current hardware setup is true (namely, that although everything should work, but the drivers just aren’t up to it because it’s a laptop), are there any channels I can go through to try to get things working on this laptop as well?

I suppose what I’d like to see is for NVidia to be able to test a Dell XPS M1530 running 32 bit Vista, a GeForce 8600M GT, and MSVC 9.0, and work out why these things don’t work together, and to offer either tweaked drivers, an updated SDK or toolkit, or some other suitable workaround. I know it’s a big ask, but I figure that NVidia are likely to have the resources to be able to look into this and find out why a setup which should work doesn’t. I don’t know whether the responsibility here lies with NVidia or with Dell (or if there’s just something stupid that I’m doing wrong), but in terms of getting CUDA working with all of the hardware/software combinations it says it works on, it seems like there’s been some kind of oversight here which would be good to get rectified.