I have created a kernel on a 32-bit Windows-XP (32bit) system and moved the executable to a Windows Vista (64bit) system.
While the application works fine on the 32bit system, it doesn’t work on the 64bit system.
What I do is to compile the kernel via NVCC and create a cubin file.
In my code I load the CUBIN file (which I embedded as a resource in my executable) via cuModuleLoadData.
If I switch everything and force my executable to run in a 32-bit context everything works fine.
Both GPUs are Tesla C1060. By the way: Is it possible to also run it on different GPUs other than the compiled one (=> cuModuleLoadDataEx?)
What could be the problem?