When connecting two GPUs in to cross-fire setup, how is that setup leveraged by the Open CL driver? Are the two devices visible as one? Do they share their GPU memory? Can I expect that when running a kernel on a cross-fire device, that the memory bandwidth will be doubled for that algorithm without thinking where and how to allocate the buffer?
Thanks, so what is then the point of connecting the two cards together? Why not leave them as is? Does that mean that Open CL is not able to make use of what is described as cross-fire?