Some questions on distributing CUDA-based APs


I m a newbee here. I would like to develop some Windows application where I
want to use CUDA to improve execution speed and distribute the application.
I have following questions relating to distribution:

Q1. Do I need to distribute any DLL or similar module along with the application?
If so, then what are they, and where can I find information licencing policies
regarding redistributing them?

Q2. The target machine might have no CUDA-supported GPU at all. I am planning
to develop the application so that it uses machine CPU when it cannot detect any
CUDA GPU using cudaGetDeviceCount(). What is the minimal set of components
I need to distribute so that cudaGetDeviceCount() (or cuDeviceGetCount() )
executes properly?

Q3. It seems Windows Vista is still not supported. Do you know anything about
supporting Vista in any future release of CUDA?

Thanks in advance.

:thumbup: First of all WELCOME :wave:


Check out this thread: Redistribute DLL

I think a lot of your questions will be answered here. and thode methods are part of the CUDA runtime


check out this thread: Mac before Vista



Thanks Jordy. I went through the said threads.

But I still have the following question

Q. It seems redistributing the set of 5 Dlls(cublas.dll cublasemu.dll cudart.dll

 cufft.dll cufftemu.dll) by copying them in the application directory is permitted. 

 However, the EULA says about the object code that "Developers ....expressly 

 agrees not to attempt to modify or distribute such Materials without first 

 receiving consent from NVIDIA.". Am I missing anything in the EULA?



I’m not sure about that question but I read somewhere that the EULA was a liitle bit old the last time… and they needed to update it. But I don’t know what version that was.

So maybe some1 of NV can help you with that.