but thats really to much in price. really.
also step away from any hybried graphik solutions also on amd side.
one of the best solutions for opensource and openCL is the AMD fusion system.
the amd fusion supports 16gb of ram in 2 SO-dimm sockets and you can use the full 16gb for openCL apps on top of the GPU.
no other solution gives you more VRAM for OpenCL on the GPU.
if you run openCL apps with havy ram usage the fusion is the KING per WATT usage.
thats the point why nvidia sell Tesla cards with 12gb Vram.
thats because on openCL only RAM matters.
and thats not all they sell tesla cards with 12gb vram but the geforces only get 512/1gb/1,5gb.
you lose any openCL benchmark if 64bit is used in the calculation and if your calculation need much RAM.
and put an fast SSD in it and 16gb of fast ddr3 ram (8gb per module).