Why has CPU's hit a wall in terms of single threaded performance? (Graphs inside)

[henrikrox]

I guess this is not macbook pro specific, so mods can move this thread if they want.

Today, i got to thinking about CPU performance the last 5-20 years, and it really got me thinking.

In 1993 the clock freq on a normal cpu at 60mhz. In 2003, 10 years later, we had clock freq at 3.0Ghz. Thats a 50 times increase in 10 years. Where are we now 9 years later? We are still around the 3.0ghz mark. Know, of course in this years we have gotten more threads/cores and improved cache, integrated graphics and other goodies. However, many many programs still don't take advantage of all this cores. From a gaming point of view, there are very few games that actually take advantages of 4 cores.

Example 1: (taken from anandtech)

Thats 21% increase.

Example 2: (also from anandtech) - with a even beefier current cpu

Thats a 14% increase.

I don't know that much about CPU's, but why are we going down this path? With improved thermals and TDP shouldn't higher clocks be easily achievable?

x86 is so dated, and there is a lot more interesting things happening elsewhere: http://en.wikipedia.org/wiki/TRIPS_architecture

Please note: That of course for some areas of work, for example: Folding, number crunching, more cores is desirable, I'm just talking about the average joe here.

Please note 2: There is a lot of interesting stuff happening in quantum mechanics which is going to translate into CPU's, however thats far away still.

 

[Slivortal]

From what I understand, single-CPU processor speeds are mainly determined by core temps (in overclocking, more coolant results in higher processor speeds). So wouldn't the bottleneck here be CPU temp? Of course, I don't know much about the science behind CPU speeds...

EDIT: Also, there's far too much precedent for x86 to see a move to TRIPS in the near future. Not saying it can't happen, but there's going to be a TON of resistance from a lot of places.

 

[leman]

This wikipedia article explains the problem rather well: http://en.wikipedia.org/wiki/CPU_power_dissipation

After a certain complexity threshold, increasing frequency on a chip is pretty much impossible due to heat issues. Parallelization is the obvious answer. Another approach would be dropping the centralized clock and moving to asynchronous CPUs, but this won't solve heat issues during peak requirements.

Fortunately, many demanding tasks are also highly parallelizable.

 

[corvus32]

Why? Because people shifted away from mostly buying big desktop computers in the early 2000's that could take the heat to buying increasingly thinner sexier mobile devices since then. The CPU industry simply went where the money was going.

If extreme single (or multi-) threaded performance is required, just over clock an i7-2600K processor.