Has either intel or AMD yet created any 128-bit instruction set? Even if so, I don't think using it anytime soon would be a good idea. As you go up in bit depth, sure you allow more to be done per clock cycle, but you dramatically increase data bus bandwidth requirements and can thus actually slow things down.
I don't think 128-bit in the general computer space (e.g. everyday applications, integer mathematics, etc) would ever be a good idea until we start seeing computers with more than 16 EB of memory (which we are nowhere near right now.) For everything else (e.g. vector math, high precision floating points, etc) we already have a nice GPU architecture for that which already handles data structures way larger than 128-bits even, and there's no need to change OS kernels for that beyond what we already have.
Originally Posted by K007
i guess what i am trying to say is...i wish we still didn't hang onto old things when new things are out...
I think this whole XP mode thing will solve that problem going forward. Right now they reserve it for people running the pro version (and the pro version is aimed at those who would care about it the most anyways,) but I think as they learn to integrate VM's with the host OS better and with faster speeds (and given that newer x86 CPU's have been made more efficient at running hypervisors) it will become something that is used more regularly as opposed to a proof of concept.
Once you can just throw everything that isn't compatible into a VM, you can start ditching legacy support.
In fact, I think it would be a good idea for Microsoft to completely drop 32-bit support in the next version of windows, and if anybody needs to run an older app, do so under a hypervisor.