Do people really do much heavy lifting with OpenCL/CUDA using cards that they are also using to actually drive browsers? Seems to me like if you are doing serious work with that sort of thing, you probably have dedicated cards for that (if not straight up using Nvidia Tesla cards or similar).
I don't use OpenCL often enough to justify the expense of a dedicated card, but, when I do use it, it's a life saver to be able to simply run the code on the CAD workstation.
as a developer, whose primary role is in GPGPU programming, i can't give you an accurate answer.
i can tell you, however, that the specialist cards typically are only in super-computers. consumer GPUs are typically faster, and significantly cheaper than their HPC counterparts.
Probably, but that's just one more reason for them not caring about 4 monitor support in non-Windows drivers. From what I've gathered from mailing lists chatter, the very reason Nvidia maintains Linux and FreeBSD drivers is OpenCL/CUDA related work, apparently some big customers demand this (but since I don't really have a source for that, take it with a grain of salt).
