as my topic title says, we want to transfer data, e.g. some preprocessed image data, directly from a FPGA-based PCIe frame grabber to
a CUDA graphics card. Since we have some hard real-time restrictions, one of our goals is to minimize the transfer latencies in our system
design. Transferring data to the system memory and then transferring it to the GPU is therefore not the best solution.
So, we think a good idea will be to transfer data directly from the FPGA frame grabber to the GPU using some DMA mechanism. But for this we need to know the memory mapping of the GPU. In the forum I’ve found several threads which deals with this topic (e.g. , , ) - but no solution.
So, our question is:
a) how to get the (physical) base address of the CUDA graphics card?