What exactly is the SM_XX architecture version?

Feb 3, 2012 at 9:49 AM
Is this the Compute capability? I am trying to use atomicAdd (and eArchitecture.sm_20), but I get the error: CUDA.NET exception: ErrorNoBinaryForGPU (Ensure that compiled architecture version is suitable for device). I'm confused by this architecture version because I know my NVIDIA GTX 560Ti is a compute 2.1 device and that it is a fermi card, which I thought is what the SM20 was all about. If I use the emulator everything works fine, just not using the GPU. Any thoughts?
Coordinator
Feb 4, 2012 at 9:35 AM

Yes it is.  We also use a 2.1 Fermi card as part of the tests and there is no issue.  What happens if you try 1.2 for example?

Feb 4, 2012 at 12:30 PM

Thanks for the clarification.  So after installing about 10 months of updates on this workstation (it was isolated from all networks until now) including an NVIDIA driver update AND VS2010 SP1 everything works FINE.  Same code as yesterday no longer throws that exception.  I am set to move on with my research.  Blog posts and articles exolling the virtues of CUDAfy.NET to follow. 

If I use a 1.2 I still get: error: no instance of overloaded function "atomicAdd" matches the argument list

Which I expect because this is a 2.0 feature as far as I can tell. 

Thanks again, this has been a great and fun experience.