1 | initial version |
For OpenCV and related things? The 1080 is probably the best as far as speed. Some of the Tesla cards can match it, but they are way more expensive. Better for double precision stuff, if you're doing scientific computing on the side.
The CUDA portion of OpenCV almost exclusively uses 8 bit ints and single floats, with a few double floats. The limiting factor for a lot of the things you do will be memory bandwidth on the GPU and transferring to and from the GPU, for which newer GPUs are better. Check the specs, there are some pretty large differences between models.