I need to build a distributed computer system with as few nodes as possible (two is ideal - one primary and one backup) and the total amount of RAM for the boards must be 48Gb or more (half of this is real data and half is a replica for backup purpose).
What I want to optimize is the cost and size of the whole systems. This will be a rack-mounted computer.
The CPU must be Intel (do not blame me - this is a req from my customer) and probably any CPU of 2GHz or better would do - faster is no drawback of course! - even if only two boards are used.
The only board I have found so far that will allow 24Gb on a board (allowing the system to be built with only two boards)is the new Tyan Thunder GC-HE (not even released it seems with quad Xenon). I would then use this board with only one processor from the start to keep the initial cost down. The alternatives are to use about 12 more normal boards with about 4 Gb of RAM on each but the cost of the whole system is then likely to be much higher even if I use P3 or P4 CPUs because of the cost of the rack-mounting modules (with power-supplies, fans etc), larger cabinet required etc.
I have also considered "blade servers" but the ones I have seen do not allow more than about 2 Gb of RAM per "blade" and then the cost become to hight (and besides I have no use for 24 CPUs!).
My question is if anybody know of any othe (preferably available and cheaper) motherboard that the Tyan that can handle a lot of RAM (preferanbly 24Gb)? I know that there are some dual Xenon boards (also from Tyan) that handle 12Gb but since I then have to use four boards instead of two (and four expensive Xenon CPUs) I think this will be more expensive than using two of the Quad-boards capable of 24Gb RAM each...
That is the point that today you can build your own "big iron" with a bunch of 1U rack mounted computers, a switch and some network cabling! This all depends on what you are doing of course - if you intend to run big databases you probably have to buy a big (expensive) SMP server but if you are doing other kind of computing that ONLY require a fast CPUs and a lot of memory you get more "bang for the buck" with your own home made solution - at least if you intend to write the software yourself so you can divide the computations between the machins!