Hi there @kodhel and welcome to the NVIDIA developer forums.
RDMA or GPU Direct is a feature that is targeted at CUDA on workstation (Quadro and Tesla) class GPUs. To use it starting with Kepler generation the GPU only needs to support BAR, which most of them do. Refer to the Accelerated Computing GPU Direct and RDMA CUDA documentation for reference.
The implementation for VULKAN was introduced much later than Kepler or even Pascal (like the P400) and as such only supports GPUs starting with the Ampere generation, for example RTX A4500 or higher or the Ada based RTX 6000.
I hope that helps.