OS dependence - Linux version not fully optimized?

Author	Message
bapu Send message Joined: 18 Dec 09 Posts: 4 Credit: 5,217,353 RAC: 0	Message 44883 - Posted: 10 Dec 2010, 13:00:18 UTC I just compared performance of the current CUDA applications under Linux and Windows as I was a bit disappointed by the Linux computation times. I have a system with GTX 295 cards that is booted either from a Windows or a Linux (Suse 11.2) disk. CUDA was installed using the latest Nvidia files under both systems, Boinc was freshly installed (as where the project files). Under Linux I observe computation times around 25 minutes, under Windows about half of that ... Could that point to the possibility of some serious improvement in the Linux code/app? ID: 44883 · Rating: 0 · rate: / Reply Quote

Matt Arsenault Volunteer moderator Project developer Project tester Project scientist Send message Joined: 8 May 10 Posts: 576 Credit: 15,979,383 RAC: 0	Message 44886 - Posted: 10 Dec 2010, 14:22:08 UTC - in response to Message 44883. I just compared performance of the current CUDA applications under Linux and Windows as I was a bit disappointed by the Linux computation times. I have a system with GTX 295 cards that is booted either from a Windows or a Linux (Suse 11.2) disk. CUDA was installed using the latest Nvidia files under both systems, Boinc was freshly installed (as where the project files). Under Linux I observe computation times around 25 minutes, under Windows about half of that ... Could that point to the possibility of some serious improvement in the Linux code/app? There isn't really any Linux specific code. The CUDA application is currently being replaced with the OpenCL one. You might want to try that. I'm not seeing much of a difference between Linux and Windows for that, though I didn't see one for the CUDA one either. ID: 44886 · Rating: 0 · rate: / Reply Quote