There is another reason for reducing the voltage; but it is not usually a user
determined function.
If you are trying to switch a transistor from "on" to "off" or vice versa, the
time taken is finite. But the rate of change of voltage is constant across most
chips. In other words they are using the same substance for the substrate, give
or take. So the lower the voltage change, the faster the switch. This is why
we have seen a steady reduction in the operating voltage of processors since the
8088.
There is, naturally, a problem with lowering the voltage: it is harder for the
circuit to discriminate between the two states. Worst of all is the use of
Emitter Coupled Logic - it switches across about .5V! That was used in the old
DEC Alpha RISC processors. Those technologies were fast, but expensive. Intel
et al use low voltage TTL... it is cheaper, and as we can see, it does work.
Note that none of this has anything much to do with heat or power per se. These
are almost side benefits. They are exploited more for these ends in, for
instance, the Intel Mobile series of processors, and to a lesser extent the AMD
equivalents.
Via C3 series use an additional, different technique to cut power: all the
instructions are run in microcode, like the old 8088-80386. This reduces the
transistor count inside the processor considerably, but also reduces the
performance for any given clock speed. The C3 is typically about 1/3rd the
performance of a Intel pentium series processor at the same clock. But then
they use much less than 1/3 the power and produce less than 1/3 the heat. The
old Transmeta Crusoe processors were different again... shame that processor has
died really.
As Chris points out, underclocking at the user level tends to walk straight into
the discrimination issue - we are not the manufacturers and so cannot re-work
the dies to correct any problems therein. Furthermore the results achieved by
one user are only a general guide to the results that anyone else might manage
due to variations in the production process, and indeed the power supplies (both
the main system PS and the low voltage, motherboard, DC-DC processor supply).
Overvoltage is a more complex issue... on one hand the switch time between
states is greater, but we still see that this generates a successful outcome;
probably due to a cleaner switch. It the voltage is too high though the
transistors that comprise the circuits con no longer cope... and a failure
usually follows. Of course the power required to run the processor and heat
emitted from it are rising too, hence the need for a low voltage power source
that has good excess power capability, and a wonderful heat sink on the
processor. All this costs money at some stage. The same statement about the
reliability of the exact voltage of the power, and the variability of the
processor in your motherboard apply as much at high voltages as low voltages.
I trust that you understand that all this is not exactly a simple process of
adding or subtracting voltage as you change the processor clock in order to
achieve your particular goal? Yes, you can vary these voltages - within limits
- as the chip manufacturers have to design with a tolerance to allow for
variations experienced in service when using a range of motherboards. Also that
the ever-faster processors of the future will use ever lower core voltages. So
will the memory, and North Bridge.
John.
----- Original Message -----
From: Chris Thomson
To: Silent-PC@yahoogroups.com
Sent: Monday, June 30, 2008 1:00 PM
Subject: RE: [Silent-PC] Re: Undervolting Intel Core2 Duo?
Well, that's backwards.
CPU (and any other semiconductor) power consumption is linear with the
formula V*V*F, that is, the square of the voltage times the frequency. So
undervolting is the best way to reduce power consumption. Of course at some
point the CPU becomes unreliable when the voltage is too low. Core2 CPUs are
designed to be reliable at a reduced clock speed (typically 6x) at 1.05V.
Many are reliable at full clock speed at this voltage, and some are reliable
at even lower voltage, but few can be run below 0.9V.
To answer the original question: the reason to undervolt is to reduce power
consumption, which in turn reduces cooling requirements, which in turn
reduces the needed fan speed, which reduces the sound level of the system.
-----Original Message-----
From: Silent-PC@yahoogroups.com [mailto:Silent-PC@yahoogroups.com] On Behalf
Of terramir
Sent: Sunday, June 29, 2008 6:59 PM
To: Silent-PC@yahoogroups.com
Subject: [Silent-PC] Re: Undervolting Intel Core2 Duo?
Q:> I only have one question - why undervolting?
>
A: Because heat generated is a function of clockspeed and voltage, the
less voltage you feed a processor the less heat it will generate, so if
a cpu is stock speed of let's say 1800 Mhz like the core2duo 4300 and
you run it at let's say 1200 mhz you have a slower computer that gives
off less heat, however the heat reduction is not linear with clock
speed, the reduction of watts drawn(heat produced) at the same voltage
is actually not 66% like that 66% of clock speed but more like a 80 to
90% of the original heat produced, however on voltage the reductions
are almost linear, so at the same clockspeed if you can reduce the
voltage by 25% you'll get like a 20 to 25% reduction in heat generated
there are some fudge factors there.
and with less heat generated you need less cooling ergo you generate
less noise.
terramir
BTW the guy with that board at stock speed I doubt you could go below
1V and the absolut minimum will probably be around 0.85V @ 33% under
stock speed. lower than that I can't tell you because I can't
manipulate my board that much.
However I have run a C2D 4300 @ 2700 Mhz (50% overclock) with about
1.175V- 1.212V so they are3 good undervolters on prinicipal, but
becareful if you undervolt you want to make sure that you have disabled
the power management options or fine tuned them because that's how my
c2d crapped out a few times when at idle the power mangement options
could reduce your voltage even further hence putting you under a
working limit so use a utility where you can specify the voltages
during power management. (remember if the bios overrides the voltage
you'll not have true vaules neccesarily in the utilities so go higher
first and see if it's stable by locking the power mangemnt options and
toasting it with a burn-in utility.
terramir
[Non-text portions of this message have been removed]
[Non-text portions of this message have been removed]