Strange slow down from Linux to Windows

Accelerated Computing CUDA CUDA Programming and Performance
1

Hey all,

So, recently I changed my operating system from Ubuntu to Win7, and I am using Visual Studio 2008. But I noticed that the Cublas functions experience a performance drop in win7. For example, using cublasSgemm, for 2000x2000 matrix multiplications, it used to take ~.2seconds on Ubuntu, now it takes ~.98seconds.

Any ideas why? I had a lot of trouble trying to set up the environment in Visual Studio (never used it before and I also just learned CUDA myself), so I was wondering if it’s because I didn’t set up the environment correctly, or something…

My GPU device is GeForce GTX280.

Thanks in advance!

2

WDDM (eg: Vista/Win7) platforms have a lot of extra overhead for any CUDA calls that call into the driver/kernel…

I’d suggest profiling the time for almost all of our CUDA function calls (especially memory copies, synchronization points, and kernel launches) (note: just the calls, don’t worry about how long the actual copy/kernel takes) - which should hopefully give you a better idea of what exactly is taking longer between the two platforms. :)

Good luck!