Thanks for bringing up these excellent points. There are forum readers with age, knowledges and experiences over a wide range. I think there are many young ages (under 10?
), existing and future scientists, engineers, professionals around, .... Some of the things I put down are HS and/or college level in computer science, engineering and physics, ..., knowing these things better should help overclocking and designing systems, ...
I try to give summary and conclusion in the technical posts so that one can skip through the details to get something (hopefully), whether agree or disagree. Sometime, I may have skipped some details about how things are derived (to make the posts shorter), but the missing details can be found from links to other posts at the
links listed in first post of the thread, and I assume who are interested in more details would read those posts also. Any questions and comments about clarity, correctness, completeness, ..., are welcome.
Back to your specific questions, here are some answers (not in order):
V^2 means V to the power of 2 = V V (V times V)
V^3 means V to the power of 3 = V V V (V times V times V)
I try to avoid using superscripts (to represent power of 2, power of 3, ...), and subscripts which may not be unverisally displayable.
I, V, R, C, L, P, f, ... are current, voltage and resistance, capacitance, inductance, power, frequency respectively. I think they have been defined in the text. More basic concepts about I, V, R, C, L, P, f and their relationship requires some understanding about principle of electricity. I will give some links below about them.
For a brief summary,
V = I R (voltage V across a resistor with resistance R and current I)
P = V I (power dissipation P in an element with current I and voltage V across)
Q = C V (charge Q stored in a capaictor with capacitance C and voltage V across)
I = Q / t (current I equals to the flow of charge Q per unit time t)
If P is power and V is Vcore, dP/dV is rate of change of power with respect to V (or in terms of calculus, it is a derivative). It is used to represent how power is changed (dP) when Vcore is changed (dV).
So dP is a
small change of power corresponding to dV, a
small change of Vcore, at a certain operating point of a CPU. There are useful for analysing small changes
without the need to know the actual C and R representing the CPU for power calculation. This may answer your question about what are the values of C and R in the model that I used. Both C and R can be calculated/estiamted (plan to discuss it in another post), but it is NOT necessary to know their values for estimating power and current changes.
For example, since
P = C V^2 f (see link below for explanation), assuming CPU clock frequency f is constant,
dP/dV = 2 C V f
dP/P = 2 (dV/V)
(so the equivalent C and f are no more in the equation for calculating change in power correponding to change in voltage)
If a CPU consumes power P of 50W at Vcore 1.5V, then a small change of Vcore of 1% (15 mV), would give a change of power of 2% (1 W).
To answer your question about P = aV^3 (where a is a constant).
dp/P = 3 (dV/V)
Actually there was a typo in the original text "dp/P = 1/3 (dV/V)", it should be dp/P = 3 dV/V, the rest of the text remains unchanged.
It is for analysing the power change when both voltage (Vcore) and frequency (CPU clock) are changing. In that analysis, the frequency f is approximted by a piece-wise linear function as Vcore. What it means is that the frequency f (for max overclocking) varies linearly with Vcore over a small range, but may not be linear over a wider range, i.e. the proportional constant varies as Vcore. E.g. at 1.5V, it is 130 MHz/100 mV, at 1.7V, it is 100 MHz/100 mV, at 1.9V, it is 50 MHz/100 mV. Since P = C V^2 f and f = k V (where k is a piece-wise linear constant), so P = a V^3, a = k C.
The other piece-wise linear constant S referred to is to describe how max clock frequency varies with temperature. Electrons move slower as temperature rises. So S = 1/f df/dT = - 0.4%/C (rule of thumb estimation). I used an estimate for every degree C rise in die temperature, there is a loss of 0.4% of CPU frequency in the example.
You are correct that C/W=(Max_Die_Temperature-Ambient_Temperature)/CPU_POWER
Here are some links to some basic terms involved in this post (also some more in the first post of this thread):
What is cycle time and frequency
What is the active power of a CPU at frequency f and voltage V
Relationship of clock, die temperature and Vcore (page 2)
) finding one rated at the very least at 20A on the 12V rail. The 12V is clearly the most heavily taxed in most of today's system. If you keep those two factors in mind, its hard to go wrong.
