>> Welcome to the world of diminishing returns. When is the last time you saw a 20% performance gain in, say, a Pentium 4? <<
TMTA will realize a greater improvement than in the typical hardware-only product cycle because the CMS can handle a good deal more ILP than what has been achieved with Crusoe’s 128-bit VLIW. I think a further 50% improvement in ILP is more likely than the meager 20% which is the midpoint of your expectation.
One pitfall some people may fall into in this analysis is including NOP instructions in their thinking for the 256-bit implementation without taking into account that there are many NOP’s in the existing 128-bit implementation. Thus, to attain a 50% improvement relative to the existing architecture, the 256-bit CMS does not have to produce 6:1 ILP but rather a much smaller figure.