lol... i know what TDP is.
The thing is, power consumption doesn't force a hard limit on what you can do. It's an optional limit. But TDP on the other hand forces a hard limit when considering the size of the console, size, type and cost of the cooling solution.
TDP is a direct function of peak power consumption. For any given processor with given clocks and voltage, the peak power consumption is essentially analogous to TDP, i.e. your thermal design limit used to design the cooling system.
One is a thermal quantity and the other a unit of electrical power consumed but the two are essentially analogous... i.e. you profile any given processor and determine a peak sustained power you design your cooling system for---you're essentially assuming all that electrical energy gets transformed into heat (which isn't far from the truth).
You are the one getting this wrong.
What specifically am I getting wrong?
Lets take say the 5700xt for instance. That's a 40CU chip and is capable of a sustained clock of around 1755Mhz at ~160W. Its possible to undervolt it and still hit those clocks but consume less power and in turn have a lower TDP. But if you choose to clock it higher by just 100Mhz or 200Mhz you could push your power consumption up by 20% and raise the TDP up by 10% if you were already at your efficiency threshold.
Nooooooo... if you push the power consumption up by 20%, the thermal output increases by 20%. Thus you need to design a cooling system with 20% more capacity, i.e. 20% higher TDP. They're directly proportional quantities.
The options are simple. If your power consumption is constant. Say 150W. When applying that 150W across two chips, one with 36CU and the other with 40CU, the 36CU chip will end up having a higher clock and a higher TDP while the 40CU chip will end up having a lower clock. You simply need more power to drive more transistors. This should be common sense.
No. If you have two chips both consuming a peak of 150W, the TDP will essentially have the same thermal cooling capacity.
What I think you're missing is that for any given fixed processor cores/clocks/voltages, actual power consumption will vary with processor load. Which is why when I've been discussing power consumption, I'm talking about the maximum which is essentially analogous to TDP.
You can tweak voltages all you want, but thats not really a viable option when considerring making 10s ofmilions of consoles each year and prioritizing yeailds. Are you going to tweak VRMs and cooling on each unit too? Nope. You will have a cooling solution and components that can handle you absolute worst performing chip. And that will be the baseline for your entire line.
The key thing you're not seeming to grasp is that clock-speed doesn't increase power consumption (/TDP) on its own. You can increase clocks by a nudge without any increase in voltage and the power consumption increase will be negligible. Increasing voltage increases power/TDP non-linearly (it's somewhere between the square and cubic power).
When console makers are manufacturing APU die, they will target voltage limits for a majority of the dies coming off the production line to be able to run at a given clockspeed stably. The voltage limit will imply a given TDP that they have already designed their console cooling system capacity too. So no there is no tweaking after the fact. But tweaking the voltage limits during production can maximise yields. So console designers will want to afford the most liberal voltage limits possible to remain within their console TDP design constraints.
On the other hand, the precise physical performance characteristics of the silicon coming out of the fab will depend on the manufacturing process. A more mature process that is producing a higher proportion of dies that hit target clocks well below the voltage limits will give the opportunity to raise the clock speed and voltage limits if desired and still be within the TDP constraints. Alternatively, they can keep the same clocks and voltage limit and enjoy higher yields.
Edit:
Referring back to my previous post where I gave the example of the PS4 Pro vs. X1X, the whole point is that clocks and die size aren't the only consideration.
You're tangent about vapour chamber cooling is irrelevant, because the cooling system only removes the heat a console APU will generate. I'm talking about the performance vs. power / thermal output characteristics of the APU itself.
The cooling system can be anything, as long as it is designed to dissipate your X Watts of TDP. That specific "anything" of your cooling system design merely implies a cost and will be more or less depending on what the X Watts TDP number is.
