I am brand new to computing with CUDA SDK and with MPP in general. I was wondering which versions of linux supports the Tesla C870. I have been having trouble finding which driver versions if any support it. I know that Windows XP’s 169.21 CUDA SDK supports it but I was hoping to take advantage of Linux’s apparent benefits.
Can anyone debunk the claims that Linux is better for parallel programming then Windows?
If not what Linux version is suggested and which SDK should I install?
RHEL4, RHEL5, OpenSuSE 10.3 and 11, Ubuntu 7.10 and 8.04, and Fedora 8 and 9 are officially supported. Others will probably work if you’re willing to experiment. At this point, if you want to try it, just use the 2.1 beta (which means 180.06).
As far as it being better… well, it’s probably easier to do (you don’t have to deal with the stupid Win32 APIs), OpenMPI is pretty nice as well, and valgrind is super great. I don’t know if that makes it “better,” though–there’s no perf difference as far as I know (outside of worst case scenarios with particular thread schedulers).
Linux has had a lot of attention for cluster parallel programming. There’s a ton of software written for it and a large community, although you can of course build clusters with Windows.
For the kind of “parallel programming” that is CUDA, there’s no advantage, really.
Basically, the question of which platform you need is about what your application is going to do, what it’s going to talk to, what libraries does it need. If these questions don’t determine which platform you need (and you’ll know if they do), then either one is fine. Pick what is familiar to you, although do research on both.