Apparently, average selling price of a current-generation high-end graphics processor over its lifetime is $18, so, either graphics card maker gets RADEON 9800 PRO or the GeForce FX 5800 Ultra, it pays roughly $18 per chip depending on the timeframe. I believe that high-end graphics processors from NVIDIA and ATI are priced more or less equally and do not account for a substantial part of graphics cards’ costs.
Given that in $399 per card there is retailer’s profit, taxes, transport fare as well as graphics card manufacturer’s margin, we can figure out that it does not really cost too much money to manufacture a high-end graphics card. Maybe about $120 or $150, but definitely not more than $200, at least, in case this is not something really exclusive or, maybe, the first batch of products with low yields and so on. Considering that graphics chips only cost about $18, we can figure out that there are two more components of a high-end graphics card that may cost substantial amount of money: these are PCB and memory. Based on various reports I can say that fast DDR SDRAM or DDR-II SDRAM cost more than a GPU or even PCB; that is basically why NVIDIA decided not to utilise DDR-II memory with its NV35 because of very high price on these advanced memory products.
It is interesting to note that the RADEON 9800 and the GeForce FX 5800 have nearly the same core-sizes of 190 square millimetres even despite of the fact that the chips are made using 0.15 and 0.13 micron fabrication processes respectively. Wafers made using 0.13 micron technology cost more than those made using less advanced 0.15 micron process; as a result, we may conclude that ATI may receive higher gross margins selling its high-end GPUs compared to NVIDIA at present.
Keep in mind that there is a bunch of expenses in the price of a graphics processor: the cost of actual semiconductor (includes chipmaker’s profit margin and other expenses, such as R&D or execution), packaging, fare, R&D expenses and developer’s margin (includes profit margin as well as other expenses like advertising or execution). So, in fact, production of GPUs is not really expensive, but should be a value that is nearly a constant. Taking this fact into account we can conclude why semiconductor manufacturers try to transit to more and more advanced fabrication processes and why it is so hard for some of them.
Source : X-Bit labs
