I looked at this diagram and I am wondering me that the increasing of the CPU performance seems to stop although the transistor count is still increasing. I mean what are the profts of the increasing transistor count after 2005? All other specs stoped to increase.
-
Clock speed doesn't equal performance. See megahertz myth.– gronostajMay 23, 2016 at 15:33
-
You start to see multi-core processors enter the market at the time the clock speed levels off. More transistors, more cores, hyper threading etc... but similar clock speeds– Joe TaylorMay 23, 2016 at 15:34
-
There are many other factors that must be accounted for other than those featured in this graph.– Ctrl-alt-dltMay 23, 2016 at 15:34
-
1Yes. But why ILP is not increasing anymore, too?– MichaelMay 23, 2016 at 17:55
1 Answer
Your example graphic is kind of outdated, but kinda answers the question at hand. Transistor count increases with the tech-ups (getting to a lower nm fab process) but once certain caps are hit, the increase in performance is no longer in-pair with the transistor count increase and that's the sign that a new architecture is needed. The story itself is quite long. An example of what I'm saying was the Athlon architecture performance jump or the i-core one. In both cases, new archtecture designed prevailed vs. increase in frequency (in 1st case) or number of cores and frequency (in second case). Your graph shows exactly a cap being hit, where the architecture becomes inefficient to keep-alive / maxed out.
So basically, more bullets in a gun are good, as long as the enemy gun does not out-range yours. When that happens, it's time that you re-design your gun.