Joel Hruska

This isn't about being a fan of any particular company. It's a discussion of what is and isn't possible within digital technology.

Go back to 1999 and 2000, and you could buy a Duron or Celeron and reliably expect a 50-60% overclock. That's gone today.

AMD's FX-9590 can hit 5GHz, but only by nearly doubling its TDP. Think about that. Increasing its clock sp eed by 20% increases TDP by 80%.

Maybe you think that's a problem unique to AMD -- but it isn't. CPU clock speeds no longer scale according to frequency because Dennard scaling hasn't really worked since 2004-2005.

http://www.extremetech.com/wp-...

Look at the dark blue line. You think that's due to Intel alone?

Take a look at ARM: http://images.anandtech.com/do...

ARM predicts that it's high performance (meaning high power consumption) cores will double in performance from 2012 to 2016, yes -- but it's high *efficiency* cores will show much smaller gains. Why? Because they don't have enough therma l headro om to gain more.

(That's also a marketing slide, not an independent evaluation, so take its projections with a grain of salt in any case).

ARM's entire big.LITTLE strategy is designed to give better flexibility to customers specifically because frequency scaling and parasitic leakage have become so difficult.

http://www.extremetech.com/wp-...

Additional evidence abounds. If you water cool an R9 290 from AMD, you can crank an additional 30% out of the GPU core. It's not a clock speed-limited architecture, it's a *heat* limited architecture. This is also true with CPUs. That's why pouring LN2 over chips will double or triple their maximum frequency.

The problem isn't that chips can't run faster. The problem is that chips can't run faster and dissipate their own in ternal h eat. And that, more than any other factor, is the reason clock speeds are stuck.