Since libcudart is less than a megabyte would I be right in suspecting that on a
host with more than a gigabyte of RAM the savings of linking a CUDA program to
the shared library libcudart.so.7.0 are negligable?
My reason for asking is I had a potential user give up because his LD_LIBRARY_PATH
did not include /usr/local/cuda/lib64/ and so the application failed with something like
error while loading shared libraries: libcudart.so.7.0: cannot open shared object file: No such file or directory
Would it be better to change the Makefile to build the image to include
the CUDA run time library?
Can this be easily done?
Is there a down side?
Is there a better solution?
As always and help or guideance would be most welcome