My CPU's days are numbered...how long before it dies?

[Idontcare]

One other observation : I'm using a V-OFFSET of 0.225 volts. That means that rather than supplying the VID voltage of ~1.145 volts for my particular chip, the motherboard is supply 0.225 more.

While I understand what it is that you are aiming to communicate here, just be aware that the term "V-offset" has a specific meaning in regards to a motherboard unintentionally over-volting or under-volting a cpu.

There are two things that keep your cpu from "getting" the voltage you specify in the BIOS - the first is Voffset and the second is Vdroop.

Voffset would be something akin to voltage regulator error. You specify in the BIOS that you want the CPU to get 1.5V but the actual output by the VRMs is 1.4V or 1.6V (a 0.1V offset in the specified voltage).

What you are referring to as a "V-offset" is not really given a specific term, we just call it the over-voltage. You are over-volting your chip by 0.225V, presumably by specifying a voltage of 1.37V in the bios?

Or do you really mean to say your motherboard actually has a 0.225V Voffset? (that is insanely huge for a Voffset, RMA your board territory)

On your chart, it takes more like 0.45 volts of offset to go from 2.8 ghz to 4.0 ghz. That means that this same graph for my CPU has a shallower slope. (is the rate of rise for a complex polynomial graph called slope? don't remember) Why is that?

Mathematically speaking it is called slope. You are thinking of "linear slope" for which the answer would be no - we don't usually speak of linear slope with functions that have a non-zero second derivative (i.e. it curves, there is a slope to the slope )

The graph is CPU specific, my chip will have a different shmoo from yours and from anyone else. Like snowflakes, they will be similar and always have the same general form of increasing from left to right.

 

[Habeed]

It has a setting labeled V-offset. However the voltage regulation system works, I know that the processor has a secret "VID" which is the voltage the processor wants at full performance. Intel sets the VID empirically by testing each chip and somehow determining how much voltage a chip needs for maximum stability.

When the chip is idle or throttled through speedstep, it asks for less voltage, and the voltage as monitored by cpu-z goes down. I have noticed that if I set my cpu voltage directly using the motherboard option, this feature doesn't work : even at idle CPU-z says my chip is running at 1.35 volts or whatever.

If I use V-offset, at idle I CPU-z displays 1.15 volts and at load ~1.37 volts (with spikes to 1.39 volts when the load suddenly reduces). Hence I use offset. Also, I was having stability problems that instantly went away when I switched to offset mode.

Yeah I took calculus 3 once, I vaguely remember now that the slope or first derivative still has variables in it rather than being a constant for a polynomial graph like you have there.

 

[Habeed]

A much more interesting question I would like to ask you :

How do the IC manufacturers avoid rare CPU errors? I've noticed through my playing with overclocking that at a particular clock speed and voltage that is marginal : aka near the point that the CPU makes a lot of errors : then there will be rare errors.

For instance, when I had my CPU clocked to 4.2 and too little voltage, I would put the system on an 8 hour stability test run and it would work fine for hours, only failing when the temperature in my room rose a little bit in the morning. For the CPU, failures are abrupt and usually total : a BSOD or the stability test program crashes with an error message. I'm guessing that the mistakes the cpu are making have to do with memory address calculations rather than a error in an FPU calculation or an integer arithmetic calc.

The same effect with my graphics card overclocking. At 850 mhz, OCCT would report an artifact every other minute. At 840, every half hour. At 830 mhz, 1-2 artifacts over a 5 hour run. At 820 mhz, it was finally 'stable' with no artifacts no matter how long I ran it. This was with a 460 gtx. Yes, I boosted the voltage and learned that a little extra voltage helps, but too much and reported 'artifacts' actually happen faster.

So how do Intel/AMD decide where to clock their chips? How do they do speed-binning quickly when a rare error might only show up in a longer test run? And finally, how did you collect the data for that graph on page 1 to find 'stable' voltages when the chip might make a mistake every few hours even at a 'stable' setting?