What is it, physically about a chip, be it a cpu or a gpu, coming off of the production line, that allows it to exceed the normal overclocking range of that chip line?
Lets say we've got a bunch of I5-2500ks default 3.3Ghz, most can OC to about 4-5Ghz, and then one random one can hit 5.8Ghz+ at unbelievably low voltages
What is it about that one chip that allows it to OC higher and/or being able to overclock at such low voltages?
While all CPUs produced basically come off of the same product line, not all are exactly the same due to micro variations as a result of the manufacturing process. For example, there is no specific production line for each model of the Ivy Bridge Core i5 CPU. There is one production line for all models. Based on testing a sample of each batch, the batch is binned based on the results into the various core i5 models.
There are literally about 1.4 billion transistors in an Ivy Bridge Core i5 CPU. The "true" golden sample would be if all 1.4 billion transistors worked flawlessly. Yeah... good luck with that. There will always be tens of thousands, maybe even hundreds of thousands of transistor that are either no functional or are not working efficiently. This leads to the binning of various models. For example:
i5-3570k - Most of the transistors are fine and these CPUs have no issues running at higher than the desired speed of a standard i5-3570. The multipliers are remains unlocked. Transistors related to vPro Technology, VT-d (virtualization), and Trusted Executions are disabled. Those are features for the non "k" models.
i5-3570 - Perhaps most of the transistors are fine for this model except the graphic core might have too many "bad" transistors so shaders and other segments of the graphic core is disabled. Therefore, instead of having an Intel HD 4000 graphic core, you have an Intel HD 2500. Perhaps there are enough "bad" transistors that prevents the CPU from overclocking to a clockspeed high enough to be worthy of being binned as Core i5-3750k. Thus this model will have it's multipliers locked and assuming the graphic core is perfect, certain segments of that graphic core is disabled to make it an Intel HD 2500 instead of the HD 4000. Transistors related to vPro Technology, VT-d (virtualization), and Trusted Executions are not disabled.
i5-3550 - If the test sample shows that the batch CPUs cannot run reliably at 3.8GHz, but they can at 3.7GHz, then they are binned as the i5-3550. Perhaps there are too many "bad " transistors, perhaps they need too much voltage to run at 3.8GHz. Who knows?
i5-3570S - If the test results show that the batch is capable of running at 3.8GHz, but do so with lower voltage so that the CPUs can be rated at 65w TDP instead of the standard 77w TDP, then they become the i5-3570S model which generally carries a premium similar to the i5-3570k models because they use less power.
Only the "k" models can be overclocked because it has unlocked multipliers. In the past you could also overclock Intel CPUs by increasing the FSB (Frontside Bus) which is linked to the RAM speed. Not so anymore. Starting with the 2nd gen Core i series (Sandy Bridge) Intel has switched to BCLK (Base Clock) which is set to run at 100MHz. Everything on the motherboard is tied to the BCLK like the USB ports and PCI-e ports. Both are sensitive to the BCLK frequency. Too high and they either stop functioning or worse, the are permanently damaged. Max recommended BCLK is probably 104MHz to 106MHz. Not worth damaging or making the PC unstable for that tiny increase in performance.
Therefore, you are limited to either the i5-3570k or the more expensive i7-3770k which has Hyper Threading (HT). Only useful if you have programs that can make use of HT. Games do not.
Golden Chips start with Sandy Bridge, Ivy Bridge is an overclocking let down. Sure Ivy has a high multiplier but it requires extreme cooling. Most of the Sandy Bridge K processors have no problem hitting 5.0Ghz @ 1.4-1.45v aircooled.