CUDA 8 and VS 2010?

Because of a hardware upgrade (from Win 7 to Win 10, and GTX 1080 Ti), I have to upgrade from CUDA 6.5 to 8. I still use Visual Studio 2010, and I really, really do not want to do an expensive and complex upgrade to VS. Nvidia online documentation says that VS 2010 is ‘Deprecated’. But will they still work together?

Deprecated means they should still work together. However it’s an indication that support will be dropped in a future release. I think if you look at the CUDA 9 support, you would find that support for VS2010 is dropped.

Thanks! I’ll go ahead and install it, and post my experiences here.

AAGH! Right up front it told me that it was unable to find any CUDA hardware on this computer! I have a pair of GTX 1080Ti cards with the latest driver. I am letting it install anyway, as the driver it wanted to install is older than the driver I have. We’ll see.

I have been using CUDA 8 with MSVS 2010 ever since the CUDA 8 release version shipped. I recently upgraded to the latest version of CUDA 8, 8.0.61 plus the patch. I have not encountered any issues, but I do not use all of the functionality of CUDA and its associated libraries, so your mileage may vary. I am currently running with driver version 385.41 (this is on Windows 7).

Thanks! Their supplied driver was older than my installed driver, which caused their error message. I’m having some success getting it set up, but a ways to go yet.

CUDA releases get packaged with the driver available at the time of the release. This means that the packaged driver can be outdated a time of installation, especially so towards the end of a release’s lifespan. That also means that the packaged driver may not support hardware introduced after the release of a CUDA version (this would be the case for the GTX 1080 Ti with respect to CUDa 8).

I haven’t checked closely in recent times but if you choose custom install in the CUDA installation you should be able to disable the touching of the driver. What I typically do is install CUDA using defaults, then immediately download the latest driver package to overwrite whatever the CUDA installation gave me.

I’m so very close… almost working, but not quite.

The build goes perfectly, and the CUDA device computes correctly, so my CUDA apps run well.

The CUDA debugger appears to work perfectly. I can place a breakpoint in a kernel, single-step, look at variable values, et cetera.

If I select “Trace application” and then check the CUDA box, it runs perfectly and gives the usual table of performance statistics, all looking correct.

BUT… If I select “Profile CUDA Application” it never collects any kernel information. Depending on what experiments I select, my app either finishes ‘normally’ or it hangs and crashes. In either case, there’s no information to display because no kernels were captured.

I’d really appreciate it if anyone has a hint of what’s going on. So close!

If this refers to CUDA integration into the MSVS 2010 IDE, I am afraid I can’t assist you, because I never use it. I only use command line tools.