tacgnol06 :
Forget performance. How long is that CPU going to survive if it's constantly thermal throttling?
Since I haven't seen any previous descriptions of how the physics works before, I will attempt to lay out an explanation that covers your question, and more:
It should easily be understood that without metallic solder under the lid that the thermals will build up on the die. So far so good.
However, that also provides a singular benefit... the average temperature of the die will quickly be pretty close to temperatures measured at any given point on the die. This is helped by how modern OSes move threads from core to core. But, we should not expect temperature sensors all over the die to confirm that what I have said or to micro-manage the throttling behavior.
Therefore, the throttling will always happen in a reasonably narrow thermal envelope across the entire die. This throttle point is based on careful testing of the die to determine the required throttle-point to guarantee the CPU integrity for years. This should even be good for a thermal overload situation (e.g. CPU fan stopped). Again, so far so good, I hope.
Now enter the normal OCer who is not going to de-lid the CPU. They are already covered, as they haven't brought anything special to the table that would interfere with the throttle-point, except making it harder to reach (e.g. higher GHz).
Finally, we have the hard-core OCer who is going to de-lid the CPU and extract heat directly from the die or add metal paste. This is the special case, as now the temperature measured at different points on the die (if we could measure it) at extreme OC is entirely dependent on the vertical thermal extraction at each point across die. This presents no issue as long as the thermal contact is perfectly even. A good OCer takes care in making sure this is the case.
However, the average person (you and I both) may make mistakes in thermal compound or metallic paste/solder application (de-lidded or not) that would create an uneven extraction of heat into the heatsink. When there is a more direct but uneven thermal path to the CPU die, we can get hot-spots that 'might' cause the die temperatures most remote from the thermal sensor(s) to push the die beyond its safe limits in those locations when a 'seemingly' good OC has been achieved. Increase the voltage too far and now we have a problem, as the transistor gates/junctions begin to break down at an accelerated rate.
So, in effect, we might (jokingly) suggest the current Intel thermal design to be: CPU with very-difficult to remove training-wheels / airbags.
It seems that AMD has taken a different bent, and attempts to put more-than-adequate cooling solutions in the hands of its customers to head off issues that might occur, but still has to carefully manage throttling to prevent damage to the die.
Facebook
Twitter
Instagram