IPC Mechanisims for MPS/CUDA

Hello all,

If I would have several kernels running in different programs, would it be possible for them to have some resources shared?

I can see in this image that the newer Volta MPS actually avoids having a shared memory space:

But for our use case it would be useful to be able to access the memory space of another service.

For example, something like having Mapped Memory from another CUDA process.

CUDA has an IPC mechanism that will work with or without MPS. There are several relevant sample codes, here is one.