It would be useful for me to be able to share a device pointer between multiple processes, both on Linux and Windows. My understanding from the Cuda documentation is that since Cuda 4 there exist some “Ipc” API functions, such as cuIpcGetMemHandle, which enable sharing a device pointer between processes. However, this mechanism seems to be limited to Linux (and to devices with support for unified addressing). My question is: is there equivalent functionality for Windows? What, if possible, would be the best way to store data on a GPU device in one process and access it from another process?
Frans van der Have