Hi! My current programs that use CUDARC work currently only on machines that have CUDA (v13) installed, and have its bin folder on the system PATH. I would like this program to run on machines that don't have CUDA installed. How can I do this? Thank you! The target machine will have a nvidia GPU, but I don't want them to have to download and install the CUDA dev toolkit.

Maybe I can copy the relevant DLLs (or SO? etc on Linux) and package them with the program, or even bake them into the program's binary? (It's a standalone binary). How do y'all generally do this? (I have figured out that I need to package the .ptx with the binary))

This is the error message I get when trying to run on a machine without CUDA toolkit installed:

Unable to dynamically load the "cudart" shared library - searched for library names: ["cudart.dll", "cudart64.dll", "cudart64_13.dll", "cudart64_130.dll", "cudart64_130_0.dll", "cudart64_130_0.dll", "cudart64_10.dll", "cudart64_130_0.dll", "cudart64_9.dll", "cudart.dll.13", "cudart.dll.11", "cudart.dll.10", "cudart.dll.1"]. Ensure that `LD_LIBRARY_PATH` has the correct path to the installed library.

So, I was thinking maybe I package cudart64.dll with the executable (Or the Linux equiv for that version)?