Early rumors about FB-DIMMs claimed that they were going to offer better bandwidth, power/heat efficiency, all adding minimal latency. Theory and reality don’t always end up being the same thing, and FB-DIMMs are great at some tasks, but there’s a fairly large latency penalty. Add to that the fact that the FB-DIMMs are significantly more expensive than existing DDR-2 DIMMs, and you can see why the technology has only taken off in the server/workstation markets.
AMD has developed a method to integrate a chip onto the motherboard that should offer many of the benefits of the FB-DIMMs with standard DDR-3 DIMMs. If the technology, dubbed G3MX, takes off, it could end up being used on normal desktops to multiple socket servers.
“This memory buffer may have the downside of boosting platform-level power consumption, but it does give FB-DIMM a host of advantages that vanilla DDR cannot match. If AMD has its way, though, FB-DIMM will have another year at most to enjoy those advantages before the company and its memory partners bring to market an FB-DIMM killer that will make DDR3 more of a fit for the server closet.
AMD's proposed solution, which it is developing in conjunction with memory makers IDT and Inphi, is called Socket G3 Memory Extender (G3MX). The new G3MX technology aims to give DDR3 the benefits of FB-DIMM's advanced memory buffer (AMB)—higher capacity, higher bandwidth, and RAS—without the power consumption headaches, and without the need for specialized DIMMs. The basic premise is simple enough: take the buffers off of the individual DIMMs and place a single buffer on the motherboard. So you get many of the benefits of FB-DIMM, but with a number of other advantages thrown in.”
Intel is expected to announce a competing technology shortly.