Is GPU Device Memory accessible by other applications

If I write an application that allocates memory and copies some data into device memory, can another application be written that can access the data at that address ?

Whether using runtime API, driver API or somehow directly through pci configuration space…?

In the general case, no.

On linux, you can use the CUDA IPC mechanism to make a CUDA device pointer visible in another application.

There is a relevant sample code, in addition to the documentation for the IPC APIs.