Preferred operating system for Cuda projects?

Just wanted some opinions on the better choice of operating system for development in Cuda?
Assume 64 bit Windows vs Ubuntu 64 bit.

I have had all kinds of problems using a Tesla S2050 under Ubuntu 12.04, and wonder if it might be easier to migrate to Windows?

If there is the same architecture is there a significant penalty for using Windows?

I know about the limit on the number of the GPU’s, but besides that does anybody have any experience working with both operating systems? If so what do you recommend?


Please note that Ubuntu 12.04 is not supported by CUDA at this time, which explains the trouble you have had. Given that Ubuntu appears to be your preferred Linux distribution, why not switch to a version of Ubuntu supported by CUDA? A list of supported platforms can be found in the CUDA 5.0 release notes.

The CUDA 5.0 release notes can be found here:

Well, I’m an old linux fan, so I would stick with Ubuntu.
What I actually use is a Debian Lenny, took me a little trouble to install the drivers because of the “default” space on /tmp, but easy enough to go around, either on the OS (increasing the size), or by telling the installer to use another place as temp space.
And I have to tell you, I’m loving it.
Specially now with nSight, I can always ‘ssh -X’ on it, and use the GUI debugger, profiller, etc etc etc.
My 5 cents is the same as njuffa, if you prefer Linux, just stick with a version that works normally and have fun with it ;).

Since you have a Tesla card, there should be no more driver overhead in Windows (with the special Tesla driver) than in Linux, so use whatever you are comfortable with. Linux has major advantages if you are using a GeForce card because you can turn off X to disable the watchdog timer, and you avoid the overhead of the Windows WDDM driver interface.