Is the FB-DIMM as dead as a dodo? No it's kicking, Pat

The added latency, with its awful effect on Sandra and Everest benchmark results, as well as the AMB chips heat & power intake, do seem to take priority over FB's separate read and write ports, or the ability to put 16 DIMMs on a dual-socket Xeon 1U gizmo. Well OK, 16 DIMMs is 200W for the memory alone.

While I don't exactly see much in the future for the current FB-DIMM platform beyond 2009, several vendors have already started producing new DIMMs meant for the dual-FSB1600 Stoakley/Seaburg dual-socket platform later this year - a platform which you could see all over the show.

Nanya, one of the less covered memory vendors here, has shown a little gem. The new lower voltage 1.5V DDR2-800 CL5 FB-DIMMs consume a bit less power, some 30%, and reduces heat by up to 30 C, which is a big deal, yet offers increased speed to provide in-sync operation for the new FSB. The capacities will ultimately be the same as for the current batch of FB-DIMMs: up to 4 GB per module, later possibly 8 GB.

Coupled with the Seaburg chipset's supposedly vastly improved latency, this should somewhat improve the overall memory performance on the dual Xeon - at least till Intel and/or Nvidia come out with high-speed, overclockable "enthusiast" dual-socket dual-FSB Xeon chipset using normal DDR2/3 memory. The grapevine says that either of these vendors should have such a chipset before the year's end. µ