I've read about this overheating issue for the past few days, including this entire thread.
We've all read reviews comparing the 4790k to the 4770k (and others) relative to heat, with promises that the "new" thermal compound is doing better than the old material. I've yet to read a review, mostly dated several months back, which observed this temperature problem (100C at stock speeds under load, throttling, etc). This is also true of the heatsink / water cooling system reviews which include the i7 line in their comparisons. It just doesn't come up.
Yet, we are seeing it happen. Not every CPU, and it appears to be widely variant.
There are some unintentionally hilarious "tutorials" on Youtube and blogs involving hammers, wood, vices and drastic measures. The more tame versions use razors, but all revolve around one central theme. They theorize that the IHS is not touching the CPU die, and/or the thermal compound inside is junk.
Now, these are anecdotal at best, but taken as a whole, once the laughter subsides from watching someone kinetically launch their CPU toward a canopy, they do seem to demonstrate there's a problem of varying manufacture of the final assembly of the IHS, and that corrective measures (which are risky for anyone with sense) are actually simple in their basic form. If the IHS were to make better contact with the die, and/or the thermal material were more carefully chosen, there wouldn't be as many (or any) reports of such high temperatures.
Now, in reality, there's only one reason to purchase a "K" chip, and that reason theoretically voids the warranty, so these drastic measures aren't really as crazy as they appear, yet there IS a failure rate of about 25% for those attempting it. It would appear there are knowledgeable enthusiasts with 95%+ success at this, but we should really hope for a better solution from a multi-billion dollar company. This echoes the Pentium 3-600 (it was overclocked from the factory, infamous for failure), and the NVidia melting GPU's, or the XBox 360 melting GPU's from a few years ago. It seems there's a simple, preventable and easily corrected error in the choice of materials and/or fabrication.
For example, those intrepid, brave experimenters, after having sent their i7 through a guillotine, discovered the black glue holding the IHS onto the PC board is really thick. Thick enough, and likely of a type of adhesive that expands slightly when it cures, that it may be lifting the IHS upward, separating the plate from the die. Some report seeing evidence of an air pocket within the TIM, which would thoroughly explain heat transfer problems.
It is entirely possible that this glue settles. Some have reported odd and sudden changes, or improvements, which they may explain as adjustments in BIOS settings, but it could be that after some time, some heat and the constant pressure of a mounted heatsink, the IHS squeezes on the glue, moving the IHS toward the CPU die, some fraction of a mm, making the critical difference in physical contact and therefore thermal transfer.
I assume that big name makers are seeing this in their production lines, but notice the bad chips, replace them - and thus we have SOME reason to suggest that i7 based systems for consumers may well be best purchased from name brand assemblers like HP, Dell, Apple and the like.
That's not an i7 "K" customer. We're left with a problem. We're going to have to gripe, return, RMA and hope for a decent performer on the Nth attempt, and a correction to what appears to be genuinely a manufacturing problem. Or, we're going to take a hammer to the thing!
I doubt there will be a mae culpa published on a forum like this, until an official response can be designed.
The design problem, with less impact, apparently dates back through Sandy and Ivy Bridge parts. The Devil's Canyon parts are already a "correction" to the 4770K, and it's not working.
Temperatures in the 88W I7 are higher than we see in other 125 watt CPU's using the same HSF, which makes no sense - unless it is a fabrication problem.
For years the IHS was soldered onto the die. There are some current parts built that way, but not these. Maybe it really isn't necessary to solder, I don't know - but it is absolutely necessary to make CONTACT, because TIM just doesn't convey heat like contact does, and every microscopic amount of additional space between the die and the IHS gives dramatic climbs, and variances, in thermal transfer effectiveness.
It's just too obvious now at these speeds, power and generation.