Why diode temp jumps around?

[Bitsbucket]

Hi all,
First I'd like to thank everyone here for past help, the newsgroups are a
wonderful thing in my opinion, and I know so many that don't even know they
exist! (newbie's that just surf and get email......I've been using them
since the 386 days, when it was allot more trouble to read them,)
anyway..........on to the questions.

My diode temps jump all around, even when I'm at idle? I do not understand
why that is, I have tried several heatsink/fan combos and its always the
same. Actually I am not at idle but just downloading files, which should not
take much work from the CPU.....any ideas? It is overclocked but how would
that affect diode temps constantly fluctuating? (45C to 47C)
Thanks
Bitsbucket.
Also in an earlier post I asked why do my temps steadily rise after a
heatsink change, it always starts out cooler and then rises to a temp, that
appears fairly steady (except for the fluctuations mentioned above) and
stays there, its almost like the thermal grease is making a good contact at
first and then degrades.......I am using Artic SilverII, is there something
better now? I lapped the bottom surface of the heatsink to a mirror finish,
literally you could have used it as a mirror!

 

[Kyle Brant]

| Hi all,
| First I'd like to thank everyone here for past help, the newsgroups
are a
| wonderful thing in my opinion, and I know so many that don't even
know they
| exist! (newbie's that just surf and get email......I've been using
them
| since the 386 days, when it was allot more trouble to read them,)
| anyway..........on to the questions.
|
| My diode temps jump all around, even when I'm at idle? I do not
understand
| why that is, I have tried several heatsink/fan combos and its always
the
| same. Actually I am not at idle but just downloading files, which
should not
| take much work from the CPU.....any ideas? It is overclocked but how
would
| that affect diode temps constantly fluctuating? (45C to 47C)
| Thanks
| Bitsbucket.
| Also in an earlier post I asked why do my temps steadily rise after
a
| heatsink change, it always starts out cooler and then rises to a
temp, that
| appears fairly steady (except for the fluctuations mentioned above)
and
| stays there, its almost like the thermal grease is making a good
contact at
| first and then degrades.......I am using Artic SilverII, is there
something
| better now? I lapped the bottom surface of the heatsink to a mirror
finish,
| literally you could have used it as a mirror!
|

I stopped using/endorsing ACII when I removed my HSF one day from my
Tbird 1.4 and noticed what appeared to be a brownish "burnt"
appearance to the exterior of the paste in the area adjacent the cpu
die on the HS where it had "squeezed out", and also that the ACII
paste had lost some of its "wetness" if you will. I reapplied a new
coat of ACII and some months later removed the HSF only to see the
same result again. I had not previously noted any problems with ACII
and its application/usage with older K6-2/3 AMD CPUs. It might have
just been airborne contaminants for all I know (I do smoke, so the
discoloration might be cigarette smoke particles).

I have experienced the exact same temperature "creep" symptoms you
describe above, at first the typical temps were lower, and slowly
increasing over time and stabilizing at a higher typical temperature
as if the paste was being "squeezed" out or perhaps flowing out from
the die-HS mechanical interface. I am using el-cheapo silicon thermal
grease now with good results for temps, though there is a slight
upward creep of temps with the silicone thermal paste also. I have
not noticed any brownish appearance to the white thermal grease as
seen with the ACII. I've read where others have used anti-seize
compound as a thermal "gap-filler" compound, but have not experimented
with this substance, other than on the bolts and screws of my car,
hehe. All of this suggested to me that a temperature versus viscosity
test of the ACII paste might be worthwhile, raising the temps to say
80-100C and looking at its "flow" characteristics and at the
temp-versus-time viscosity stability of the compound.

As to the diode temps moving around quite a bit, I think such is not
abnormal, and is very dependent upon the location of the diode w/r/t
the other circuitry of the CPU die. Keep in mind that even at idle,
your cpu IS working.

Finally, there are some who have suggested that lapping your HS is
counter-productive as the total surface area for the cpu-HS interface
is then reduced (with pits and lans removed, the surface area of the
HS adjacent the CPU die is marginally smaller thus reducing total
surface area and likewise lowering thermal transfer efficiency). The
thermal conductivity of the "filler" material (thermal paste) plays a
major role in the thermal transfer efficiency also. Lots of good
questions to prognosticate upon eh?

 

[Pief]

Hi all,
First I'd like to thank everyone here for past help, the newsgroups are a
wonderful thing in my opinion, and I know so many that don't even know they
exist! (newbie's that just surf and get email......I've been using them
since the 386 days, when it was allot more trouble to read them,)
anyway..........on to the questions.

I think theyd be better if Lurkers feel obliged to post more -
Every ten posts one reads, one should put in a little feedback.
I imagine AI 'dialogue digestors' will read through all this stuff in years
to come - perhaps just for fun

My diode temps jump all around, even when I'm at idle? I do not understand
why that is, I have tried several heatsink/fan combos and its always the
same. Actually I am not at idle but just downloading files, which should not
take much work from the CPU.....any ideas? It is overclocked but how would
that affect diode temps constantly fluctuating? (45C to 47C)

The fluctuations might just be down to a little random noise in its output,
the quicker the diode is sampled, presumably the more flucuations one might
detect, and there would be a sample rate thats just impractical with a
minimum realistic time period which the diode needs to collect enough
instantaneous impressions to calculate a probable temperature thats
generating its sensory events.... could work like that

Also in an earlier post I asked why do my temps steadily rise after a
heatsink change, it always starts out cooler and then rises to a temp, that
appears fairly steady (except for the fluctuations mentioned above) and
stays there, its almost like the thermal grease is making a good contact at
first and then degrades.......I am using Artic SilverII, is there something
better now? I lapped the bottom surface of the heatsink to a mirror finish,
literally you could have used it as a mirror!

With my machine, I put this behaviour down to to the air in the case, and
all the other compenents warming up -more slowly than the cpu, but
providing a slow feedback effect.

Im really itchin for a new nick soon, this one has been dragged through a
few too many embarassing excursions :]

 

[Ben Pope]

Thats fairly normal... the time constant of the CPU Diode is pretty low and
2°C isn;t a great deal.

Rises over what period of time? The time constant for your overall system
is probably quite high. It could easily take 20 minutes for your heatsink
to heat up and the rest of the case/components to reach a steady state.

Finally, there are some who have suggested that lapping your HS is
counter-productive as the total surface area for the cpu-HS interface
is then reduced (with pits and lans removed, the surface area of the
HS adjacent the CPU die is marginally smaller thus reducing total
surface area and likewise lowering thermal transfer efficiency).

Hmm, I'm not sure I'd go along with that... the heatsink to thermal paste
surface area is reduced, not the CPU to Heatsink area. Maybe if the thermal
paste had a lower thermal resistance than the heatsink... It's an
interesting one, but I think you can safely say that your thermal paste is
less good than your heatsink, therefore you want to minimise the amount of
paste, thus lapping your heatsink is beneficial... whether it's noticeable
or not is another matter.

Ben

 

[Kyle Brant]

--
Best regards,
Kyle
| Kyle Brant wrote:
| > | >> Hi all,
| >> First I'd like to thank everyone here for past help, the
newsgroups are a
| >> wonderful thing in my opinion, and I know so many that don't even
know
| >> they exist! (newbie's that just surf and get email......I've been
using
| >> them since the 386 days, when it was allot more trouble to read
them,)
| >> anyway..........on to the questions.
| >>
| >> My diode temps jump all around, even when I'm at idle? I do not
| >> understand why that is, I have tried several heatsink/fan combos
and its
| >> always the same. Actually I am not at idle but just downloading
files,
| >> which should not take much work from the CPU.....any ideas? It is
| >> overclocked but how would that affect diode temps constantly
| >> fluctuating? (45C to 47C)
|
| Thats fairly normal... the time constant of the CPU Diode is pretty
low and
| 2°C isn;t a great deal.
|
| >> Also in an earlier post I asked why do my temps steadily rise
after a
| >> heatsink change, it always starts out cooler and then rises to a
temp,
| >> that appears fairly steady (except for the fluctuations mentioned
above)
| >> and stays there, its almost like the thermal grease is making a
good
| >> contact at first and then degrades.......I am using Artic
SilverII, is
| >> there something better now? I lapped the bottom surface of the
heatsink
| >> to a mirror finish, literally you could have used it as a mirror!
|
| Rises over what period of time? The time constant for your overall
system
| is probably quite high. It could easily take 20 minutes for your
heatsink
| to heat up and the rest of the case/components to reach a steady
state.

The max temp measured under full CPU load rises over a period of
months. My comments were not clear on this point, but I suspect the
OP intended the same meaning.

|
| > Finally, there are some who have suggested that lapping your HS is
| > counter-productive as the total surface area for the cpu-HS
interface
| > is then reduced (with pits and lans removed, the surface area of
the
| > HS adjacent the CPU die is marginally smaller thus reducing total
| > surface area and likewise lowering thermal transfer efficiency).
|
| Hmm, I'm not sure I'd go along with that... the heatsink to thermal
paste
| surface area is reduced, not the CPU to Heatsink area. Maybe if the
thermal
| paste had a lower thermal resistance than the heatsink... It's an
| interesting one, but I think you can safely say that your thermal
paste is
| less good than your heatsink, therefore you want to minimise the
amount of
| paste, thus lapping your heatsink is beneficial... whether it's
noticeable
| or not is another matter.
|

Then have a look at this article, food for thought:

http://www.overclockers.com/tips458/

 

[Paul]

Hi all,
First I'd like to thank everyone here for past help, the newsgroups are a
wonderful thing in my opinion, and I know so many that don't even know they
exist! (newbie's that just surf and get email......I've been using them
since the 386 days, when it was allot more trouble to read them,)
anyway..........on to the questions.

My diode temps jump all around, even when I'm at idle? I do not understand
why that is, I have tried several heatsink/fan combos and its always the
same. Actually I am not at idle but just downloading files, which should not
take much work from the CPU.....any ideas? It is overclocked but how would
that affect diode temps constantly fluctuating? (45C to 47C)
Thanks
Bitsbucket.
Also in an earlier post I asked why do my temps steadily rise after a
heatsink change, it always starts out cooler and then rises to a temp, that
appears fairly steady (except for the fluctuations mentioned above) and
stays there, its almost like the thermal grease is making a good contact at
first and then degrades.......I am using Artic SilverII, is there something
better now? I lapped the bottom surface of the heatsink to a mirror finish,
literally you could have used it as a mirror!

If the diode is truly on the die of the processor, the thermal time constant
is very quick. A socket based sensor has to wait for the socket to pass
the heat to the sensor, so a socket sensor will give the long term average
temperature.

To see what is going on, you need to observe the %CPU and the temperature
at the same time. See if there is a correlation between the two. If the
%CPU is constant, the temperature should also be constant. Otherwise,
the observed variation could be measurement noise from the monitor chip.

With respect to the thermal paste question, at least the AS III I use
changes consistency as time passes. The thickening of the paste is
supposed to help it stay in place between die and heatsink. No matter
what kind of paste material you use, there is a "pumping effect", where
the pressure and heat will cause the material to flow. With AS III,
what you are supposed to see is the temperature drop a couple of degrees
over the first two or three days of constant operation, followed by a
rise over the next year, at which time it is time to reapply it. Since
I use a very thin layer, I never see any squeeze out - it just seems to
dry out.

A more permanent thermal interface material (TIM) would be a phase change
type. This material is a solid at room temperature and is plastic at
operating temperature. It should be more resistant to the pumping effect
and last longer than a year. (How much longer it remains effective, I have
no idea.) The Intel HSF for example, has a metal foil with what looks
like carbon black (graphite) on it. That isn't likely to go anywhere.

HTH,
Paul

 

[Bill]

My diode temps jump all around, even when I'm at idle? I do not understand
why that is, I have tried several heatsink/fan combos and its always the
same. Actually I am not at idle but just downloading files, which should not
take much work from the CPU.....any ideas? It is overclocked but how would
that affect diode temps constantly fluctuating? (45C to 47C)

How do you mean by constantly, every second, every hour ? Don't you even allow for
reading tolerance ?

 

[w_tom]

The erroneous assumption (too often by those who don't learn
how microprocessors work) is that amount of work output is
somehow related to how hard a CPU works. Bad assumption. CPU
works full speed no matter what it is doing. But CPU power
consumption varies with different machine code instructions.
You do not know how hard the CPU is working unless you first
list the machine instructions being executed. And don't
forget - the CPU is working on many more processes besides
what you have told it to do. Typically OS is running more
than 20 processes besides your program.

Intel engineering papers discuss a program that does 'no'
work but that maximizes heat inside the CPU. They simply
optimize the program to generate the most heat - and do
nothing. They organize the order of how instructions are
executed to also maximize heat - and do no work. Remember,
CPU is working full speed constantly. Just because you are
not taxing its output does not mean CPU is working less.
Unfortunately, too many "knowledgeable" computer users make
that erroneous assumption.

 

[JBM]

--
Best regards,
Kyle
| Kyle Brant wrote:
| > | >> Hi all,
| >> First I'd like to thank everyone here for past help, the
newsgroups are a
| >> wonderful thing in my opinion, and I know so many that don't even
know
| >> they exist! (newbie's that just surf and get email......I've been
using
| >> them since the 386 days, when it was allot more trouble to read
them,)
| >> anyway..........on to the questions.
| >>
| >> My diode temps jump all around, even when I'm at idle? I do not
| >> understand why that is, I have tried several heatsink/fan combos
and its
| >> always the same. Actually I am not at idle but just downloading
files,
| >> which should not take much work from the CPU.....any ideas? It is
| >> overclocked but how would that affect diode temps constantly
| >> fluctuating? (45C to 47C)
|
| Thats fairly normal... the time constant of the CPU Diode is pretty
low and
| 2°C isn;t a great deal.
|
| >> Also in an earlier post I asked why do my temps steadily rise
after a
| >> heatsink change, it always starts out cooler and then rises to a
temp,
| >> that appears fairly steady (except for the fluctuations mentioned
above)
| >> and stays there, its almost like the thermal grease is making a
good
| >> contact at first and then degrades.......I am using Artic
SilverII, is
| >> there something better now? I lapped the bottom surface of the
heatsink
| >> to a mirror finish, literally you could have used it as a mirror!

Then have a look at this article, food for thought:

http://www.overclockers.com/tips458/

Interesting article, When I first install my 2500+ I was able to over clock
my FSB to 200Mhz now the best I can do is 190Mhz. My heatsink is
very shiny. When I take it off to redo the thermal paste I'll take some
sandpaper and roughen it up to see if it makes a difference.

 

[Paul]

The erroneous assumption (too often by those who don't learn
how microprocessors work) is that amount of work output is
somehow related to how hard a CPU works. Bad assumption. CPU
works full speed no matter what it is doing. But CPU power
consumption varies with different machine code instructions.
You do not know how hard the CPU is working unless you first
list the machine instructions being executed. And don't
forget - the CPU is working on many more processes besides
what you have told it to do. Typically OS is running more
than 20 processes besides your program.

Intel engineering papers discuss a program that does 'no'
work but that maximizes heat inside the CPU. They simply
optimize the program to generate the most heat - and do
nothing. They organize the order of how instructions are
executed to also maximize heat - and do no work. Remember,
CPU is working full speed constantly. Just because you are
not taxing its output does not mean CPU is working less.
Unfortunately, too many "knowledgeable" computer users make
that erroneous assumption.

Ever heard of the HALT (HLT) instruction ? By using this in the
idle loop of your favorite OS, power is indeed saved. I can
even prove the power dissipation in the computer is not constant
by using my clamp-on ammeter. In my latest computer, the difference
between running MEMTEST and idling in WinXP is ~1 amp and 0.55 amps
respectively, from the wall socket.

The idle loop runs, whenever there is no useful work to do by the
OS. If the idle loop is coded with a HLT instruction, power is
saved. If you use a really old OS without a HLT instruction, then
power dissipation will be higher.

http://www.heatsink-guide.com/cpuidle.htm (and many other sites...)

HTH,
Paul

 

[w_tom]

As posted previously:

Typically OS is running more than 20 processes besides your program.

New work is constantly being processed even when operator is
not executing program. Why would temperatures vary? Only
some relationship between amount of work the operator assigns
and how hot a CPU is executing. Temperature of CPU is not
determined by how 'hard' a CPU is working - from perspective
of operator. Temperature is determined by what machine
instructions are unique to that process. And so there are
programs that accomplish no work but make the CPU very hot.
Other programs can execute that are constantly outputting work
- and cause little CPU heating.

A CPU can do much work and yet not become very hot. Or a
CPU can become very hot and not do any productive work. More
is determined by which machine instructions and in which order
they execute. That can explain temperature variations even
when operator is not executing a program. Halt instruction is
simply another instruction that consumes less power. It just
also happens to be an instruction that also does no work. But
CPU executes many other instructions besides Halt. And CPU is
typically running more than 20 separate processes.

 

[Paul]

As posted previously:

New work is constantly being processed even when operator is
not executing program. Why would temperatures vary? Only
some relationship between amount of work the operator assigns
and how hot a CPU is executing. Temperature of CPU is not
determined by how 'hard' a CPU is working - from perspective
of operator. Temperature is determined by what machine
instructions are unique to that process. And so there are
programs that accomplish no work but make the CPU very hot.
Other programs can execute that are constantly outputting work
- and cause little CPU heating.

A CPU can do much work and yet not become very hot. Or a
CPU can become very hot and not do any productive work. More
is determined by which machine instructions and in which order
they execute. That can explain temperature variations even
when operator is not executing a program. Halt instruction is
simply another instruction that consumes less power. It just
also happens to be an instruction that also does no work. But
CPU executes many other instructions besides Halt. And CPU is
typically running more than 20 separate processes.

This seems to be a pretty convoluted description of where power
dissipation comes from. In a CMOS circuit, power is dissipated
whenever a gate changes state. Power consumption is proportional
to FCV**2, where F is the toggle frequency, C is the capacitance
of the node being charged or discharged, V is the voltage of the
circuit.

When portions of a processor are not used on a given clock cycle,
their power is reduced significantly (but is still not zero).
You would be correct to say that the power consumed on each clock
cycle is _somehow_ related to the operation that is being done.

If you step back a bit from the gate level, and look at the
architecture of the chip, its power saving (ACPI) features, and
how an operating system works, there are other aspects to note.
The purpose of the HLT instruction is to stop execution until the
processor is awakened by an event. The event can be an interrupt
from a PCI/AGP card or it could be an interrupt from an internal
resource like a timer (used by the timer queue) or the system
clock.

The way that modern OS work, is they have a task scheduler inside
them, that at a minimum is executed for each system clock tick. To
give the illusion of multitasking (many programs sharing one processor),
the scheduler gives slices of time to programs. On some computers
this happens 100 times a second, so an individual slice is 10
milliseconds.

If the scheduler discovers there is no process that wants to do any
work, then the idle process can be run for a slice. If the HLT
instruction is executed in the idle process, then the processor will
be asleep for up to the next 10 milliseconds. During this time, power
consumption might be reduced to 25% of max power, as many portions of
the chip are not toggling (changing state). The only memory bus
activity will be refresh cycles done every 15.625uS by the Northbridge,
so the memory runs cool as well.

At a gross level, that is where the big power change comes from.
That is why there is such a difference between my 1.0 amp of wall
current when running MEMTEST (which runs flat out and has no scheduler
or power saving features - MEMTEST86 is a single dedicated program
without an OS) versus idling in WinXP only drawing 0.55 amps. The 0.45
amp difference in wall current represents all the intervals where the
processor is stopped, waiting for a clock tick/time slice interrupt
or another hardware event to wake it up. Note that, for this to work,
there must be a regularly arriving hardware event to awaken the
processor, otherwise it will become a zombie.

As time goes by, the design of processors is encouraging ever large
power dissipation swings between 100% processor utilization (no HLTs
being executed) compared to what is done when the processor is idle.
For example, I believe both AMD and Intel are introducing a feature
on their newest processors, where the VID code (the logic signals that
say what voltage to use for Vcore) is actually changed on the fly.
The new voltage regulators will be able to reduce or increase the
voltage in 200 microseconds, when instructed to save power by the OS.
(this is reducing the value of V in the FCV**2 equation, whereas the
HLT method is reducing F, the frequency that data bits change state.)
This will happen in a similar way to the HLT instruction - when the
OS is eventually coded to use this feature, the OS will reduce the
processor voltage and even the frequency (Power_Step) when no process
wants a time slice. This should go a long way to making desktop
motherboards conserve power just like their laptop brothers.
(This will also confuse the hell out of people when they read
the Vcore voltage in MBM )

HTH,
Paul

 

[Creeping Stone]

The erroneous assumption (too often by those who don't learn
how microprocessors work) is that amount of work output is
somehow related to how hard a CPU works. CPU
works full speed no matter what it is doing. But CPU power
consumption varies with different machine code instructions.

- Theoreticaly, but CPU instructions are bunched and resolved now in
parallel see ,so the CPU is constantly trying to employ all of its
mechanisms. From a simplified rendering of microprocessing Id expect the
variation in heatouput you imagine too, but in practice it boils down to
wether the CPU is explicitly being instructed to idle/power-save and wether
its not (blast ahead at full chat)
/Certain Pentium models inc P4 and, Ive read but dont know - the Barton,
slow down when they sense theyre too hot with a built in temperature
sensing diode.

You do not know how hard the CPU is working unless you first
list the machine instructions being executed. And don't
forget - the CPU is working on many more processes besides
what you have told it to do. Typically OS is running more
than 20 processes besides your program.

A process is an Operating system level handle for whats being done,
Processes group together threads of execution, so while you can see 20
processes running in task manager, each process can relate to multiple
threads so its normal for hundreds of separate threads to be running.
The time each individual thread gets to run is refered to as a quantum-
http://www.sysinternals.com/ntw2k/info/nt5.shtml

I suppose its a bit like molecules and atoms, at a certain level we feel in
the know because we know about molecules -before we find out about atoms
that is

Intel engineering papers discuss a program that does 'no'
work but that maximizes heat inside the CPU. They simply
optimize the program to generate the most heat - and do
nothing. They organize the order of how instructions are
executed to also maximize heat - and do no work.

Such heat producing code, does not produce a great amount more heat than
'normal' code though -they usualy specialise in thrashing the on chip
caches which are big heat producers in CPUs -thats why the Barton runs
hotter than Tbred, and why the Tbred can reach slighlty higher clock
speeds. Even when the CPU is waiting multiple cycles -up to 100, for data
to arrive from memory, its waiting in a state of readiness, refreshing
itself constantly and consuming almost as much power as it will need when
it gets to proceed. The difference in heat output is most significantly
effected by instructions which tell it to shut down units and buses
entirely.

CPU is working full speed constantly. Just because you are
not taxing its output does not mean CPU is working less.

Athlons min quoted power consumption (for idle) is around 50w and Max (for
readyness & thrashing) about 65w. I think Pauls reported power consumption
with memtest will have a lot to do with all the memory driving going on, in
the Northbridge and dimms, which can put out significant heat.

-Run a cpu thrasher and see what temp you reach, thats code designed to
produce heat, then reboot and go into your bios, Bios code is not designed
to produce heat, but does not explicitly idle either so if theres a
hardware section with cpu temperature readout, you can check how close it
is to the delibrately designed cpu thrasher programs - its pretty close for
me.

Unfortunately, too many "knowledgeable" computer users make
that erroneous assumption.

There is almost no end to this knowledge and more importantly
understanding. I can be a bad one for getting gleefull and glib with a
grasp on a few esotericals - its no harm to act young. Experts feilds of
reference can be very impressive in their prefered context but completely
miss points in others.

-Stay Curious

 

[Creeping Stone]

In my latest computer, the difference
between running MEMTEST and idling in WinXP is ~1 amp and 0.55 amps
respectively, from the wall socket.

Hi, Im intrested to know, what the wall socket reports for non-memtest 100%
cpu utilisation in winXP, and if that current is from US or UK mains
voltage -to estimate how much power consumption varies from driving the
memory.

tia,

 

[w_tom]

Yes the Halt instruction can reduce power. But the point is
this. How much 'work' is a computer doing? Does executing a
memory test program mean the CPU is working harder - or that
other parts of computer are working harder - compared to
what? What determines how 'hard' a CPU works are the
instructions it is processing - and not necessarily the task
assigned by a human. If continually processing a Halt
instruction, then CPU is consuming less electricity - while
working full speed from the perspective of CPU and working on
nothing from the perspective of the user. But computer never
stays in Halt. It has 20+ processes to service while, from
perspective of user, it is doing nothing - while it is not
working.

The point is that amount of heat produced - if that is the
measurement of working 'harder' - varies according to the
machine instructions processed AND not because the program
must accomplish more 'work' from the perspective of the user.
And as Creeping Stone notes, this is further complicated by
how multiple instructions are executed simultaneously. Even
in Halt, the CPU is working at full speed - just on a low
power instruction that stops some parts of the CPU.

Because different instructions have different power
requirements, Intel's early Pentiums required the power supply
to go from less than 1 amp to tens of amps in microseconds.
There can be that much power difference between different
machine instructions. Even a massive change in power
consumption can mean 'harder' during that power change because
change of power and not the overall power consumed is
'harder'. Using electricity consumption as the benchmark of
work, then different instructions have a massive difference in
how much 'work' is done. But the CPU, in the meantime keeps
processing instructions at full speed and executing 20+ other
processes - regardless of whether the user is running a
program or not.

IOW the question is "what is working hard?". Define
"working hard". Somehow many users assume the CPU processes
no instructions when no programs are running. There are 20+
other processes to be serviced. CPU continues executing full
speed even when Halt instruction is implemented. Some
programs will cause CPU to run hotter because some programs
use more 'high electric consumption' instructions grouped in a
certain way. How hard is CPU working then? Depends on which
machine instructions are getting executed - and not
necessarily because the picture has many pixels. Electrical
consumption of a CPU can vary greatly even during the same
program execution which is why Intel required CPU power supply
to be able to change from one to ten amps - quickly.
Different machine instructions make CPU work harder or
easier. What instruction is executed determines CPU heat.

From the perspective of a user, the CPU is either running or
sprinting. So yes, sometimes it is working harder - but who
cares. Its working harder either way.

 

[Paul]

In my latest computer, the difference
between running MEMTEST and idling in WinXP is ~1 amp and 0.55 amps
respectively, from the wall socket.

Hi, Im intrested to know, what the wall socket reports for non-memtest 100%
cpu utilisation in winXP, and if that current is from US or UK mains
voltage -to estimate how much power consumption varies from driving the
memory.

tia,

' gathering moss,
andy

The current was measured on a 120V Canadian outlet. When the current
is measured in isolation, it doesn't take into account the power factor,
which is the phase angle between voltage and current, so it isn't
strictly accurate to multiply current by voltage to get power, but it is
close enough.

WinXP Idle = 0.55 amps
3DMark2001SE = approx 1 amp
Prime95 Small FFT = 0.961 amps (the torture test menu item)
Prime95 Large FFT = 1.03 amps (the torture test menu item)

The difference between the two currents, times 120V, gives about
8.4 watts for the memory + Northbridge interface. As a DDR memory
stick has a power rating of about 5W, and there are two in this
machine in dual channel mode, this 8.4W number sounds OK.

Paul

 

[Kyle Brant]

I suggest there are many ways to describe power consumption aspects of
a CPU, nevertheless, if power is applied then instructions are
executing and it is consuming power. A CPU will consume different
amounts of power based upon its operational state (the actual
instructions executing).

I do note that power is consumed whether a CMOS or MOSFET device is
conducting (gate enabled) or in the non-conducting state (gate
disabled). Of course, in the disabled state, the power consumption is
very low, but when there are millions of such gates with power
applied, the total leakage current becomes a significant value. Also,
switching a MOSFET device on and off at higher frequencies requires
more power than lower frequency switching assuming the same
operational voltage across the drain/source (IOW, a 50% duty cycle at
1MHz will require less energy or power than a 50% duty cycle at 100MHz
due to parasitic capacitance and inductance of the circuitry).

 

[Creeping Stone]

In my latest computer, the difference
between running MEMTEST and idling in WinXP is ~1 amp and 0.55 amps
respectively, from the wall socket.

Hi, Im intrested to know, what the wall socket reports for non-memtest 100%
cpu utilisation in winXP, and if that current is from US or UK mains
voltage -to estimate how much power consumption varies from driving the
memory.

tia,

' gathering moss,
andy

The current was measured on a 120V Canadian outlet. When the current
is measured in isolation, it doesn't take into account the power factor,
which is the phase angle between voltage and current, so it isn't
strictly accurate to multiply current by voltage to get power, but it is
close enough.

Im glad you worked that out for me then

WinXP Idle = 0.55 amps
3DMark2001SE = approx 1 amp
Prime95 Small FFT = 0.961 amps (the torture test menu item)
Prime95 Large FFT = 1.03 amps (the torture test menu item)

The difference between the two currents, times 120V, gives about
8.4 watts for the memory + Northbridge interface. As a DDR memory
stick has a power rating of about 5W, and there are two in this
machine in dual channel mode, this 8.4W number sounds OK.

Paul

Thanks for that - good secure info.
Thats 8.4 watts difference for driving the memory and Northbridge though
,no? -not huge, but intresting on top of the current already used for
constant refreshing.

Im actualy just trying to work out, if my Climate Prediction App adds
anything to the 'lecy bill since it does heat the northbridge a bit - but I
think I can tolerate it now

 

[w_tom]

To return to the original point, the CPU is working hard -
generating much heat - whether or not the user is running an
application program. So much heat, that an AMD processor will
self destruct (in seconds) whether or not CPU is executing an
application program - because the CPU is working that 'hard'.
Therein lies an analogy of a CPU either running or sprinting;
it does not stand idle or sit in a comfy chair when not
executing an application program. It keeps working so hard as
to create significant heat.

The above being contrary to a layman's perception that a CPU
is only working hard when executing user's programs. CPU is
'working so hard' that AMD CPUs always require heatsink
dissipation and Intel CPUs will activate its emergency thermal
protection circuits. That is the point. Contrary to popular
myth, the CPU is always working hard.

Hopefully defined is a perspective to the concept of
"working hard".

In the meantime, the CPU's Halt instruction was (once) used
more liberally in laptops. Trade off was decreased power
consumption for less performance. Is Halt used equally in
both desktop and laptop systems now?

 

[Paul]

To return to the original point, the CPU is working hard -
generating much heat - whether or not the user is running an
application program. So much heat, that an AMD processor will
self destruct (in seconds) whether or not CPU is executing an
application program - because the CPU is working that 'hard'.
Therein lies an analogy of a CPU either running or sprinting;
it does not stand idle or sit in a comfy chair when not
executing an application program. It keeps working so hard as
to create significant heat.

The above being contrary to a layman's perception that a CPU
is only working hard when executing user's programs. CPU is
'working so hard' that AMD CPUs always require heatsink
dissipation and Intel CPUs will activate its emergency thermal
protection circuits. That is the point. Contrary to popular
myth, the CPU is always working hard.

Hopefully defined is a perspective to the concept of
"working hard".

In the meantime, the CPU's Halt instruction was (once) used
more liberally in laptops. Trade off was decreased power
consumption for less performance. Is Halt used equally in
both desktop and laptop systems now?

I was hoping that my quotation of the power figures on my new
computer would make that apparent. 1 amp when "working hard"
and 0.55 amps when not "working hard". You see, I am one of
those "laymen" who chooses to believe that when a peocessor
executes a HLT instruction, the processor "stops working hard".
It is sleeping. It will be awakened by the next hardware interrupt,
and when that happens, the power consumption will climb.

The power consumption is not constant, neither is the rate that
work is being done. There are a number of throttling mechanisms
at work. An example we haven't talked about yet, is how the
Pentium behaves when it overheats. It doesn't execute an instruction
on every clock tick. It has some clock cycles where it does nothing,
in an attempt to reduce the processor temperature to a safe point.
This is why it is possible to execute Prime95 on a Pentium4 and
rip the heatsink/fan off the processor while it is running. Even
though the Prime95 program wants to use every cycle, the throttling
feature inside the processor reduces the percentage of clock cycles
during which instructions are executed. This reduces the processing
rate and the heat dissipated. (No, I haven't read any Intel docs on
this, so I don't have any links to quote from. This is paraphasing
what I've read in this newsgroup.)

With the temperature controlled fans in the system, I don't need
to look at a "process viewer" to tell when the CPU is running
at 100%. The fans tell me when the CPU is really busy, as they
make an awful lot of noise when the CPU and PSU heat up.

What I'm objecting to is the notion that the CPU executes an
instruction on every clock cycle. For some periods of time, the
processor is asleep. For some percentage of clock cycles during
periods of overheat, the processor will not execute an instruction,
in an attempt to reduce the temperature. So, while in times past,
a processor used to execute an instruction per clock cycle (no
power reduction features in those processors), current processors
have a whole array of features that give variable performance.

Even the Northbridge has a throttling feature. Many Northbridge
chips have a % duty cycle counter in them, which watches how ofter
a DIMM is accessed in a given interval. If the duty cycle is too
high, the Northbridge will actually reduce the rate that it does
transactions on the DIMM. And if the memory stops responding, this
causes the processor to stall (which causes its power dissipation
to drop as well). This feature has to be supported in software, and
I don't know if modern OS enable it or not. Here is an example
in the 845PE datasheet:

http://developer.intel.com/design/chipsets/datashts/25192401.pdf (pg.98)

Paul