agp 3.0 or 8x is just over 2GB/s theoretical, pci-e is 250MB/s per lane theoretical, making agp similar to pci-e x8. and pci-e x8 in sli only causes a few percent performance drop from pci-e x16 on top end cards, with some of the bandwidth used to transfer information between the cards/coordinate the cards. so i think the agp bus could have been kept in service for quite a while longer.

just like ddr ram, the price for agp parts rose simply due to new standards being quickly adopted by both intel and amd at the same time. but amd could have let intel do the hard work in the marketing sector (which intel could actually afford to do), and left adopting ddr2 and PCI-e till later on... although in the end whether to adopt PCI-e or not was somewhat up to motherboard and chipset manufacturers.

its like amd caved in to the suggestions of their marketing department, going to ddr2 and pci-e x16 so as to not apper 'behind' intel, along with other gimmics like labelling their cpus with numbers that were unrelated to the clockspeed (damn i hated that) when people who really cared (and would have made the bulk of the purchasing decisions) would have known full well that clockspeed was an unreasonable indicator of performance. and ddr2 turned out to have excrutiatingly little advantage over ddr2 given how much it cost, and the agp bus could handle today's top end dx10 cards with little or no performance hit.

all not good!