1)Yes they are. By design is impossible for the board components to have the same tolerance level as the maximum power draw capacity of the main chips. 15-20 years ago that tolerance level was only near 10% beyond the max capacity, now-days it can easily go between 20-30%. Most manufacturers started using military grade components after the capacitor plague specially for products intended for the overclocking the crowd, now every manufacturer use them,  from the cheap Acer products to the luxury class of devices from Apple. If you don't know this, is because you're too young perhaps. You need to understand basic electric engineering, what a capacitor is, a resistor, and how the transistors unload and overloads these two constantly when you use your pc, to comprehend what really happens when you mess with the original clock frequency settings. Just a change of .1mv would alter the internal components bringing stability or havoc to the entire system. If you don't take into account the tolerance level, the system will shut down or crash. That is why Bios, usually disable some options or put limits to them. With your approach then Yes, you are controlling the power state from windows, but windows is making a predicted change to the bios settings on the fly. But, you are not doing anything out of the ordinary to what the chip was designed for. Except for what you said earlier about pushing TDP up, and for this is why I talked about tolerance levels, is a requirement and not a choice for proper functioning, specially when lowering voltage for higher frequencies which decreases amperage (if passive), and stresses capacitance; hence why capacitors are the first to be blown. (capacitor tolerance level was exceeded)

2)Power states are just ACPI profiles. Just because you can play with Windows settings doesn't mean you understand what you are doing. But you are not an idiot, I give you that.  While you can be clever sometimes, the tone you use in your writing...pufff, just blows your intended purpose.

3)Clearly the framing of my comments include my experience blowing stuff every time for testing purposes. After repairing lots of shit like this everyday, I know very well what I'm talking about. Computer enthusiasts like lots of people here enjoy the thrill of squeezing performance....until something goes wrong and burn their systems. But clearly you don't give a shit about that, so you don't know, but some other people do really. If by basic understanding you mean my Ph D, then yes, clearly is an issue you haven't found yet cause you are not that deep in to this as I wrongly assumed before.

4) I was being humble with you, because I don't like to use AMD APUs. You can not judge everything in life based only on your self experiences; if something that makes you a naive person, something I highly doubt. You understood right away what I meant when I said VRAM, whether it was dedicated or shared ram, is still applied the context of my argument. Yet you decided to twist the argument to impose yourself with a deviated criteria.

APUs are just embedded processors within the same die, and the concept doesn't apply solely to AMD; Intel and Qualcomm also do this. But to answer you directly, all AMD APUs can have dedicated video memory if the target product requires it, just like with everything else.  Sharing the memory is just cheaper, and by design is the default option for building APUs boards, but sometimes an alternative is needed The are some brands like Intel, Acer, Zotac, Asus, etc that offer a wide range of products (APU style) with dedicated video memory modules for specific consumer needs.

Usually overclocking is limited in the bios. YES including those boards and cpus with unlocked potential. No matter how much you increase frequencies, the BIOS will not let you exceed some parameters and voltage, it can even default those values without letting you know if it finds instability. Whenever you computer restarts, crashes or shuts down, is a designed behavior from the Bios. If the bios lets you tweak something, then a tolerance level was already calculated. This is why sometimes you have to update your Bios, to have a better support and stability from new CPU or GPU units, etc. Most of the time with same model processors  will have multiple configuration sets based on the code of manufacturing, so behavior and performance outcome will vary even from a same model. If your system was reliable for years it was because it was only a marginal tweak, one already calculated and compensated in the BIOS internal code settings. But those changes will bite you back...sooner or later, depending on the quality of your build.

The only exception for this is ram overclocking, as ram modules do not have an ID tag for custom function sets based on the stick model and brand. Instead they use timings for compatibility, but if you mess with those, well unexpected behavior can occur, maybe a a random crash or a full corrupted hard drive. A very high overclocking on a ram stick would crack it, and I have seen it, when using nitrogen as cooling solution, it pops like pop corn out of the motherboard.

1) You will need to provide a citation of those systems with an AMD APU with dedicated graphics memory.

2) You can modify a BIOS.

3) RAM can have profiles.

1) I don't need to. You are purposely deviating from what I meant at the beginning, and going full circle. So there is not point. You are only focused in consumer platforms for PC. Enterprise platforms and custom platforms solutions are different.

3) Not profiles. ID tags so they can be identified just like CPU codes are verified during boot by the BIOS, currently this is non-existent. RAM sticks use a FPGA controller for compatibility based on the NAND speed and socket target. If given and id tag or code, it will probably provoke costs to go way higher with the testing across infinite devices, wil require new socket profiles/pinouts for the ID module integration and it will not provide any real benefit to the standard consumer. RAM sticks are built based on specifications. The motherboards(bios) usually just check if the ram complies with those, and adjust timing accordingly, after 1st boot.

Going back to the main topic, there were rumors about PS5 overheating. While the rumors are probably fake (Jeff Rickel) I will be neutral on this just to be fair for the sake of the argument. Accordingly to Cerny, the way PS5 is being built is with a specific power level that will be linear to the capabilities to the cooling solution, reducing drastically the thermal load spike under heavy processing that occurs in ordinary platforms, by changing the frequency and not the voltage. By having an specific power level, the cooling solution will work as intended in all game scenarios independently from workload they can provoke. At least he was very clear in that aspect and while he hasn't revealed yet the cooling solution, if memory serves me right he mentioned that it was also to maintain the system as silent possible preventing the fan speed from changing so much. In addition to that there were reports from the stock market analysts, stating that the cooling solution selected for the PS5 was more robust and costlier than previous consoles. With the PS5 specs already revealed, the cooling solution surely was designed entirely to manage such high frequencies in the RDNA2 architecture. The only aspect that could be giving Sony a problem is the additional heat coming from the SSD solution, while it can be true I don't really think it would be such hurdle. If SSD is really increasing heat inside  the console, then it just need to be relocated or isolated. Another way is to add heatsinks and use a separate airflow for it, nothing special or complicated for a company as Sony. So, in my opinion the rumor is just utterly nonsense to me.

1) You made a claim, you have failed to adhere to the burden of proof, thus said claim can be discarded.

2)RAM includes a Serial Presence Detect chip which has an ID tag so that it can communicate with the BIOS to set the appropriate defined memory timings and frequency.

The XMP profile is an extension to that which allows for the DRAM to operate with defined specifications outside of the JEDEC specs.

RAM doesn't include NAND. RAM is RAM. Ram is not NAND.

1) What I said was that overclocking can damage VRAM, yet you are focused and obsessed to make a point based on the generalization of AMD APUs North-bridge limitations, which has nothing to do with the point on hand I was discussing. But ironically I'm gonna put perhaps the gratest proof of all, for you. For example the PS4. In the case of the PS4 system is an AMD APU design, and yet it has been built with a dedicated video memory(GDDDR5), and in reversal it shares the VRAM Pool with the system memory. And before you rant about PS4 not being an APU, here read a little more. Perhaps now you're gonna say I didn't understand you or that im not interpreting correctly, and for that my friend mental hospitals do exist.

2) That is not what I talked about. And yes RAM has NAND. Read here. Also you can read more here. For a computer to be able start and boot properly a ram stick needs to have a SPD module. Basically what this module do is communicating the BIOS the safest timing and frequency under the proper DDR standard and start the PC. This has been like this for ages, and with it requiring manual tweaking in the BIOS for high performance sticks for extra performance if you need it. This changed a bit with the XMP profiles by Intel. With XMP the BIOS can access additional timings and frequencies stored in a 2 pre-arranged profiles within the SPD modules, so the bios can auto adjust these settings without further tweaking just by choosing which. For this to work, the SPD module need to have such data, if not, it will run in standard mode without any auto-tweak. If the extra data in these profiles were to be used as the default SPD info, RAM compatibility will be broken creating a huge problem between motherboard makers and ram manufacturers.

About the PS5

Mark Cerny said he was saving more details for the teardown of the system. Probably if any surprise, it will announced at that moment. Any chance you guys see the launch date being pushed back a few months(4-6) for whatever reason? 

1) Granted I am playing with semantics, obviously we are talking about dedicated graphics memory which is of the GDDR variant.

2) The framed context of your statements made it sound like you were asserting that RAM is NAND. They are separate entities... Different technologies.

Those sticks are server products, not consumer commodity sticks anyway. Again, redundant.