I had this idea today while I was thinking about SGI and their last major feature: NUMALink (Non Uniform Memory Access). Essentially a way of allowing a computer to access a neighboring computer’s memory as if it were it’s own, with a slight performance penalty. This is what keeps SGI afloat: Large shared memory spaces.

Now I’m going to dazzle you with my faint knowledge of Electronics & Computers. Think about it for a minute: What is memory? Just a collection of Memory chips with some IO and Clock lines. You set the Address Lines, hit the clock, and the desired value appears on the output lines (Probably simplifying, but I think it’s close to the truth). So… Why couldn’t you build a device with a large Ribbon-cable type connection that could plug into a standard DIMM slot. This could then be connected to an external brick of DIMM’s with some basic hardware to connect them together & relay the addressing information. To your computer it would simply appear as a massive 128G DIMM or whatever, when in fact it was an array of smaller DIMM’s.

Granted, access times would be slower. But this could be overcome easily I would think, either through delaying the clock signal or by setting wait states that the CPU already understands. This would make large addressable memory spaces a nearly Consumer item.

Just something to think about I guess.

Be Sociable, Share!