used for ray tracing, and other professional applications.
performence wise its not so amazing anymore.
http://www.nvidia.com/object/product..._c1060_us.html
less then 1 Tflop single precision and just 78Mflop double precision.
ati and nvidia cards that cost 200 euro's are faster.
but drivers are designed to help with raytracing, GPGPU, and more
its a bit like firegl/quadro VGA cards, performence wise there not super, there extremely pricey, up to 5000 euro's for a card.
but the drivers are designed for CAD drawings and stuff, and support is a lot better.
and if you use a normal ati/nvidia card for cad, even if its performence wise a better card, firegl/quadro cards are better for cad drawings.
but useless for gaming.
there was a company that made a PC with 4 times a 9800*2 VGA card, and used cuda to control the cards.
the 4000 euro PC they made performed faster than a 512 CPU render cluster.
but only for certain applications, as you need to rewrite your application with cuda, else it wont use your VGA cards
and that is the idea of GPGPU, use your GPU to do more tasks, thats why native C++ support from nvidia a,d openCL from nvidia/ati is so interesting.
raw performence of a GPU is a lot more than hat of a CPU, we just need to use it
