funky i dont understand one thing... Why are we thinking in terms of card itself.... Cant we have PCB where we can plug the GPU of our own choice keeping RAm and other stuff being the same...
Thats a concept what was thought of by Voodoo earlier.
Replacable Cores and Memories.
Thus just have a daughter board on the PCI-e slot, allowing us to plug in The Core (Like we do our CPU onto the mobo) and also Plug in the Memory (Like we do our RAM in the RAM Slots.)
Well its simply not fisible. We dont get as many models of perticular core as we get in CPU.
GPu architecture is specific to that core and whole PCB layout depends on that perticualr core.
GPU is far more complex than out current CPUs. The RAM can be efficiently accessed upto certain distance only from the core.
GPU PCB needs to be optimised for that perticular GPU.
Its not that noone thought about this. The decision was made with these things in mind. Its not fisible to go and devlope such system.