I came across your website at https://docs.nvidia.com/cuda/gpudirect-rdma/index.html
and I have a question which I hope you will be able to answer.
We want to achieve things like GPUDirect RDMA support for Linux on Windows 10 by combining your company’s GPU and our own PCIe device + driver.
In order to do so, I would like to obtain the physical address on the GPU.
Is there any way to obtain it with the current driver on Windows 10 ?
We recognize that currently CUDA’s data transfer API is via handles and it does not allow to pin the destination address like Linux’s GPUDirect RDMA support.
It may be difficult to provide an API to obtain directly in the kernel like Linux, but I would appreciate if you could tell me how to obtain the physical or virtual address of the GPU memory that has been pinned by existing APIs or IOCTLs.
If only the virtual address is used, I think it can be transferred by converting it to a physical address in the kernel mode driver.