<I think that rambus isn't cost competitive near term>
You may be right, but then again, I doubt that DDR SDRAM will be, either. Unlike the transition from PC100 to PC133, adding DDR to the equation isn't merely a case of low-hanging fruit. You must add pins to the SDRAM interface in order to stabilize the DDR signals. Also, to double-pump data, you have to add clock strobes to latch the data coming through. Those strobes sure won't come for free, especially for a wide 64-bit interface. And finally, because of the high pin-count, adding additional DDR SDRAM controllers to a chipset in the future will be much more expensive than adding additional DRDRAM controllers. Heck, even Intel's upcoming Carmel chipset will feature two DRDRAM controllers on the chipset. Because of the small pin-count, an additional controller can be considered "low-hanging fruit." Try designing a chipset that featured two DDR SDRAM controllers. I guarantee you that the dual DRDRAM solution will look much more attractive.
The only legitimate arguments against DRDRAM are short-term arguments. Those I will accept, because DRDRAM will initially have a lot of startup headaches. They'll no doubt bring up the old standby arguments, "It's too expensive" or "PC's don't need it yet." But considering what the alternatives are, I'll put my money on Rambus.
Tenchusatsu |
