In simple terms, if you want higher speeds, you need to provide more current per time unit to the transistors. Simplified, you want 6% higher speed, you need 6% more current (which is done by increasing the driving force, the voltage). Now if you looked into your old physics book, you would find that the heat to dissipate over your resistors goes like P = I^2 * R. So contrary of what you think, heat dissipation "goes worse" with clock rate increase. AMD stuff had a tendency to go through the roof rather early (compared to Intel chips) at some clock rate. This was improved and AMD can go around 1GHz without excessive heat production. Again the key word here is "excessive" (check a 7970 for heat dissipation), the 53MHz clock rate increase does increase the heat output noticably.

The second misconception is the assumption that "larger box = better cooling". This is wrong, as larger volume always means "possibility to have dead air space". So it is actually more difficult to cool stuff in a large box than to cool the same stuff in a small box (if done correctly). What always wins is good airflow within your box, irregardless of size. With "simple" consoles like XBox One/PS4 (it basically has only one major, but very localized, heat source, the APU), so designing a reasonable cooling system wasn't that difficult for either company. MS chose to go large box because it wants it to look like a cool Hifi component, and could then incorporate a "huge" fan as a freebie, pretty much noiseless until your game goes into overdrive..

I wonder who figured out the 853MHz, though. Why that odd number? Or maybe the guy who posted it simply mistook an engineering test ("we could run the APU consistent at 853MHz" in the lab) as a sign that end user consoles run at 853MHz?

Notice that for a company that told everyone who wanted (or not wanted) to hear that "it is not about the specs, it is all about the software", it is rather strange that they are fiddling with exactly those specs so late in the development cycle. Actually, we might already in the production phase so I wonder who made the long term test with increased clock rates...

In simple terms, if you want higher speeds, you need to provide more current per time unit to the transistors. Simplified, you want 6% higher speed, you need 6% more current (which is done by increasing the driving force, the voltage). Now if you looked into your old physics book, you would find that the heat to dissipate over your resistors goes like P = I^2 * R. So contrary of what you think, heat dissipation "goes worse" with clock rate increase. AMD stuff had a tendency to go through the roof rather early (compared to Intel chips) at some clock rate. This was improved and AMD can go around 1GHz without excessive heat production. Again the key word here is "excessive" (check a 7970 for heat dissipation), the 53MHz clock rate increase does increase the heat output noticably.

drkohler said:

Notice that for a company that told everyone who wanted (or not wanted) to hear that "it is not about the specs, it is all about the software", it is rather strange that they are fiddling with exactly those specs so late in the development cycle. Actually, we might already in the production phase so I wonder who made the long term test with increased clock rates...

Pemalite said:

Rubbish.
You don't need to increase the current (voltage) to increase clock speeds, have you learnt nothing about overclocking on the PC? Hint: People have been doing it for decades, even reducing voltages and increasing clock speeds and thus reducing heat and power consumption.

No offense, but I'll take the proof that is decades worth of overclocking over your opinion.

I agree with @drkohler.

The example you give is with multipliers... so you didn't changed the base clock to go with high multiples... like 10x200 = 2000Mhz or 11x200 = 2200Mhz... no change in voltage.

But to increase the base clock you have a limit... you start to increase it until you have to increase the voltage to get stable.

Now remember GPU only have the base clock (AMD GPU... nVidia have two clocks but I won't enter in this part).

Pemalite said:

I never mentioned anything about multipliers, nor do we have the option to play with multipliers on GPU's.

My point still stands, when you raise voltage, heat and power consumption increases exponentially. (I.E. The increase is the square of the voltage in a linear circuit.)

If you increase clockspeed and not the voltage, the increase in heat and power consumption isn't going to be an exponential increase, nor do you *have* to increase the voltage to run at a higher clock speed. (Untill you hit a wall and need more voltage, that is.) - Again, decades worth of proof via overclockers is available for you to peruse via google.

Also, nVidia moved away from having the shader clocks untied from the core clock some time ago.

RRod had more to to with solder failure than chip failure.

He did better...