I’m designing an FPGA device that need to feed video into the GPU memory with as low latency as possible.
According to GPU Direct for video, on copy of video data into the host frame is required. But according to GPU Direct for storage / networking it’s possible to copy data directly through PCIE switch from the storage device / network adapter into GPU memory, w/o host memory bounce buffer.
Is copying video data from FPGA PCIe device directly into GPU memory using memory mapping like described in GPU Direct RDMA supported?