Data Tansfer Speed From Host to GPU card

There is one parameter critical for us, the time to transfer video frame from host CPU to Nvidia card (Telsa or other). Do you know how many ms it takes to transfer HD video image data (appox 3.5 mb per frame in size)? My application is about video image processing in real time so the time to transfer data from host to coprocessor card and back is very important to us.

Does any one already have the benchmarking or please point me the direction to get it.

Transfer from memory to the GPU (pinned memory) ranges from 3-6 GiB/s based on your hardware platform. Just run the bandwidthTest sample in the CUDA sdk to benchmark it.