advice on underclocking a Pentium 4 processor

[David Madore]

Hi!

I have severe overheating problems with my Pentium IV 550 processor
(at the normal frequency of 3.40GHz) so I'm looking for advice on how
to best underclock it (to diminish heat production as much as possible
with no degradation of system stability and only a relatively small
performance loss). Can someone help me with this?

One of the difficult parts is that I cannot actually measure CPU
temperature (apparently my kernel - it's Linux - doesn't handle the
motherboard features well enough yet, or something, so I have no
temperature reading). All I know is that when I start something too
intensive (like a big compilation) the emergency thermal regulation
steps in and downclocks the processor. The BIOS thermal monitor reads
a temperature of around 75 Celsius, which is more than the Intel spec
says it should be (72.8), though not much more, but then, that's not
in normal operating circumstances.

Anyway, I tried lowering the clock frequency in the BIOS (from 200MHz
down to 195MHz - the multiplicator being 17) and decreasing Vcore from
the nominal 1.4V to 1.325V, and I still have to test how stable this
is and whether it still overheats, but I took those figures more or
less at random. Is there some rule of thumb that says how much one
can decrease Vcore without compromising system operation if frequency
is decreased by that much? If my Vcore were too low, would I know
this at once (like the CPU not booting at all) or could it lead to
subtle errors in intensive computations?

There are many sites out there on the Web dealing with overclocking,
but only a very few pages deal with underclocking (and those which do
are about removing fans altogether, something that I'm afraid I can
hardly consider doing), so I thought I'd ask on this newsgroup.

Thanks for any thoughts!

 

[Michael W. Ryder]

Hi!

I have severe overheating problems with my Pentium IV 550 processor
(at the normal frequency of 3.40GHz) so I'm looking for advice on how
to best underclock it (to diminish heat production as much as possible
with no degradation of system stability and only a relatively small
performance loss). Can someone help me with this?

One of the difficult parts is that I cannot actually measure CPU
temperature (apparently my kernel - it's Linux - doesn't handle the
motherboard features well enough yet, or something, so I have no
temperature reading). All I know is that when I start something too
intensive (like a big compilation) the emergency thermal regulation
steps in and downclocks the processor. The BIOS thermal monitor reads
a temperature of around 75 Celsius, which is more than the Intel spec
says it should be (72.8), though not much more, but then, that's not
in normal operating circumstances.

Anyway, I tried lowering the clock frequency in the BIOS (from 200MHz
down to 195MHz - the multiplicator being 17) and decreasing Vcore from
the nominal 1.4V to 1.325V, and I still have to test how stable this
is and whether it still overheats, but I took those figures more or
less at random. Is there some rule of thumb that says how much one
can decrease Vcore without compromising system operation if frequency
is decreased by that much? If my Vcore were too low, would I know
this at once (like the CPU not booting at all) or could it lead to
subtle errors in intensive computations?

There are many sites out there on the Web dealing with overclocking,
but only a very few pages deal with underclocking (and those which do
are about removing fans altogether, something that I'm afraid I can
hardly consider doing), so I thought I'd ask on this newsgroup.

Thanks for any thoughts!

Are you sure that the heatsink is properly attached? I had similar
problems with this processor. It took several tries to get all the hold
downs properly installed. Before I got them right the idle temperature
was over 60 degree Celsius, afterwards they are under 40. My high
temperature, even at 100% utilization is still under 60.

 

[kony]

On Wed, 10 Aug 2005 22:14:53 +0000 (UTC),

Hi!

I have severe overheating problems with my Pentium IV 550 processor
(at the normal frequency of 3.40GHz)

Is this an unusual environment it's in? They're hot running
chips BUT given a decent heatsink, fan, and chassis cooling
they aren't usually "severely overheating". Is the system
instable? Shutting down? What temp is it reaching?

so I'm looking for advice on how
to best underclock it (to diminish heat production as much as possible
with no degradation of system stability and only a relatively small
performance loss). Can someone help me with this?

No, you can't just reduce heat a lot without a significant
performance loss. However, you might not need 3.4GHz worth
of performance- we can't determine that need for you.

Even so, the CPU is multiplier locked so your alternatives
are reducing voltage (if motherboard bios allows it) and
FSB. Likely you'd set the FSB clock 33MHz lower, then drop
the voltage by about 10% and test stability. If it's not
stable then raise voltage a little, or if stable, lower
voltage a little more and retest, until you find the
threshold of instability... then add some margin of
stability.

Generally it's good to test CPU stability with Prime 95's
Torture Test. A very instable CPU will produce errors
within a few minutes, and the test can be stopped
immediately as soon as the first error appears so you can
reconfigure the system, but a thorough test must run for
several hours without any errors to be trusted.

One of the difficult parts is that I cannot actually measure CPU
temperature (apparently my kernel - it's Linux - doesn't handle the
motherboard features well enough yet, or something, so I have no
temperature reading). All I know is that when I start something too
intensive (like a big compilation) the emergency thermal regulation
steps in and downclocks the processor. The BIOS thermal monitor reads
a temperature of around 75 Celsius, which is more than the Intel spec
says it should be (72.8), though not much more, but then, that's not
in normal operating circumstances.

Yes that's too high. Where is the deficiency? Case
cooling? Ambient room temp? Fan speed? Heatsink itself?
Heatsink-CPU interface (poor mounting or thermal interface
pad/grease/etc) ?

Anyway, I tried lowering the clock frequency in the BIOS (from 200MHz
down to 195MHz - the multiplicator being 17) and decreasing Vcore from
the nominal 1.4V to 1.325V, and I still have to test how stable this
is and whether it still overheats, but I took those figures more or
less at random. Is there some rule of thumb that says how much one
can decrease Vcore without compromising system operation if frequency
is decreased by that much? If my Vcore were too low, would I know
this at once (like the CPU not booting at all) or could it lead to
subtle errors in intensive computations?

5MHz is a rather trivial FSB reduction. For significant
heat reduction you'll probably need far more FSB reduction.

If the vcore is too low it can cause the system to not POST
at all. It could post but crash during POST, or in bios
menus, or have trouble finding suitable boot devices, or
booting the OS, etc, etc. In other words, there is no "one"
indication of instability, due to the role of the CPU an
instability can be manifested almost anywhere.

IF you find no apparent problems, including no windows
errors reported, it can still easily be possible that data,
calculations are erroneous. That is one reason to test with
Prime 95's Torture Test, because it checks the calculations.

There are many sites out there on the Web dealing with overclocking,
but only a very few pages deal with underclocking (and those which do
are about removing fans altogether, something that I'm afraid I can
hardly consider doing), so I thought I'd ask on this newsgroup.

Do not consider underclocking to be any different than
overclocking. Either is merely changing the clock rate.
The original clock rate is somewhat an arbitrary speed,
merely what the manufacturer guarantees to be stable but
once a core is mature, it's mostly just how many MHz you
paid for, the price-tiering of any given technology at any
point in time. Putting that in context, there certainly are
still limits to what any core can do and some specimens are
better than others, but towards the end of adjusting FSB,
voltage, cooling, etc, there is little difference between
overclocking and underclocking.

Also keep in mind that when changing bus speeds, it may
result in different memory speeds & timings too. For that
reason you should also test memory with Memtest86+ before
ever booting the OS.

You need find a way to monitor the temp though, it's quite a
handicap not being able to determine what the full load
(maximum heat) is during testing. If all else fails,
consider getting an external temp probe, boring a hole in
the heatsink and mounting the thermal probe in it. A more
sophisticated method would be soldering up a circuit to the
CPU socket pins to take temp readings but that is a bit more
difficult and I have no suggestions on how you might do it
on your board.

 

[Paul]

Hi!

I have severe overheating problems with my Pentium IV 550 processor
(at the normal frequency of 3.40GHz) so I'm looking for advice on how
to best underclock it (to diminish heat production as much as possible
with no degradation of system stability and only a relatively small
performance loss). Can someone help me with this?

One of the difficult parts is that I cannot actually measure CPU
temperature (apparently my kernel - it's Linux - doesn't handle the
motherboard features well enough yet, or something, so I have no
temperature reading). All I know is that when I start something too
intensive (like a big compilation) the emergency thermal regulation
steps in and downclocks the processor. The BIOS thermal monitor reads
a temperature of around 75 Celsius, which is more than the Intel spec
says it should be (72.8), though not much more, but then, that's not
in normal operating circumstances.

Anyway, I tried lowering the clock frequency in the BIOS (from 200MHz
down to 195MHz - the multiplicator being 17) and decreasing Vcore from
the nominal 1.4V to 1.325V, and I still have to test how stable this
is and whether it still overheats, but I took those figures more or
less at random. Is there some rule of thumb that says how much one
can decrease Vcore without compromising system operation if frequency
is decreased by that much? If my Vcore were too low, would I know
this at once (like the CPU not booting at all) or could it lead to
subtle errors in intensive computations?

There are many sites out there on the Web dealing with overclocking,
but only a very few pages deal with underclocking (and those which do
are about removing fans altogether, something that I'm afraid I can
hardly consider doing), so I thought I'd ask on this newsgroup.

Thanks for any thoughts!

The best way to attack this problem, is with a good heatsink/fan.
You might even need to use a larger computer case, with more
room for air circulation around the CPU area. Some people have
reported a 10C reduction in temperature, by using a larger case.

Possible candidate heatsinks are Zalman CNPS7700, CNPS7000 (using
an adapter kit for LGA775 socket), Thermalright XP-90 and XP-120
(both of these don't ship with a fan, and the fan is purchased
separately).

http://www.zalman.co.kr/eng/product/view.asp?idx=145&code=005009
http://www.zalman.co.kr/eng/product/view.asp?idx=141&code=005009
http://www.thermalright.com/a_news/main_news_xp120.htm
http://www.thermalright.com/a_page/main_product_xp90C_775.htm

You can also consider installing a duct assembly, to bring cool
air directly to the CPU area.

90nm processors have a fairly large leakage-current-based thermal
component, so underclocking will not be as effective as you would
hope. I have used this site, to get voltage versus frequency
information, but this site no longer contains enough data to
make good quality scatter plots:

http://cpudatabase.com/index.cfm?action=search

I don't use Linux too much, but when I did, I used mbmon / xmbmon
for temperature readout. A search shows this as one site for the
program. There may not be support for your motherboard, but it
is worth a try.

http://www.nt.phys.kyushu-u.ac.jp/shimizu/download/download.html

Hope that helps,
Paul

 

[Robert Hancock]

Hi!

I have severe overheating problems with my Pentium IV 550 processor
(at the normal frequency of 3.40GHz) so I'm looking for advice on how
to best underclock it (to diminish heat production as much as possible
with no degradation of system stability and only a relatively small
performance loss). Can someone help me with this?

One of the difficult parts is that I cannot actually measure CPU
temperature (apparently my kernel - it's Linux - doesn't handle the
motherboard features well enough yet, or something, so I have no
temperature reading). All I know is that when I start something too
intensive (like a big compilation) the emergency thermal regulation
steps in and downclocks the processor. The BIOS thermal monitor reads
a temperature of around 75 Celsius, which is more than the Intel spec
says it should be (72.8), though not much more, but then, that's not
in normal operating circumstances.

Anyway, I tried lowering the clock frequency in the BIOS (from 200MHz
down to 195MHz - the multiplicator being 17) and decreasing Vcore from
the nominal 1.4V to 1.325V, and I still have to test how stable this
is and whether it still overheats, but I took those figures more or
less at random. Is there some rule of thumb that says how much one
can decrease Vcore without compromising system operation if frequency
is decreased by that much? If my Vcore were too low, would I know
this at once (like the CPU not booting at all) or could it lead to
subtle errors in intensive computations?

There are many sites out there on the Web dealing with overclocking,
but only a very few pages deal with underclocking (and those which do
are about removing fans altogether, something that I'm afraid I can
hardly consider doing), so I thought I'd ask on this newsgroup.

Thanks for any thoughts!

This sounds to me a lot like your heatsink is not making proper contact
with the CPU, to have the CPU temperature get up to 75 degrees and start
throttling. Better to get that fixed, otherwise you'll likely have to
take the CPU speed way down to keep it from overheating.

 

[David Madore]

First, many thanks for your reply, and thanks to all others also.

Is this an unusual environment it's in? They're hot running
chips BUT given a decent heatsink, fan, and chassis cooling
they aren't usually "severely overheating". Is the system
instable? Shutting down? What temp is it reaching?

The environment isn't too unusual: room temperature is high (27C,
that's 81F - I also have an overheating problem there) but not that
high. Case ventilation seems adequate: I've just added a new fan at
the back, which improved things considerably as far as the graphics
chipset goes (went down from 70C to 63C when idle), but doesn't seem
to have changed much as far as the CPU goes. The system is rock
stable even when the overheating alarm triggers the throttle - I've
left it running that way for as long as I dared, performing very
stringent checks, and all checks passed. As for the actual
temperature reading, it's hard to say, because my motherboard sensors
aren't supported under Linux and that's all I've got. I can read the
temperature under the BIOS setup utility, and when I leave it running
for several minutes (the BIOS must be busy waiting on the CPU, because
it's definitely heating a lot) it runs up to ~80C.

The "bad or badly attached heatsink" is beginning to sound plausible,
so I'll try to buy a new one tomorrow, but even that is strange: the
heatsink came from Intel, together with the CPU, and the
whatever-you-call-those-little-plastic-things-which-hold-everything-in-place
seemed to fit in perfectly and make just the "right" sound when
locking in place.

I tried using "Arctic Silver 5" thermal material today, but it doesn't
seem to have helped any (in fact, it probably made things worse
because I couldn't follow the instructions to the letter).

No, you can't just reduce heat a lot without a significant
performance loss. However, you might not need 3.4GHz worth
of performance- we can't determine that need for you.

I certainly can't avoid having _some_ kind of performance loss, but
maybe I can try to minimize it. For instance: only the CPU is
overheating, memory is fine, so maybe there's a way to tune the
timings so as not to underclock the memory. I'm definitely not an
expert there

Even so, the CPU is multiplier locked so your alternatives
are reducing voltage (if motherboard bios allows it) and
FSB. Likely you'd set the FSB clock 33MHz lower, then drop
the voltage by about 10% and test stability. If it's not
stable then raise voltage a little, or if stable, lower
voltage a little more and retest, until you find the
threshold of instability... then add some margin of
stability.

Right now I lowered Vcore from the nominal 1.4V down to 1.2975V (the
lowest the motherboard would allow) and I downclocked the front-side
bus from 200MHz to 170MHz (so I'm running effectively at 2.9GHz
instead of 3.4GHz) - and still overheating if I push things far enough
(though now it takes minutes rather than seconds before the thermal
throttle sets in). The system is still perfectly stable, however, as
far as I could test it.

Yes that's too high. Where is the deficiency? Case
cooling? Ambient room temp? Fan speed? Heatsink itself?
Heatsink-CPU interface (poor mounting or thermal interface
pad/grease/etc) ?

I wish I knew! One possibility is that the internal temperature diode
is utterly wrong and is throttling for no reason at all: is that known
to happen? Is there some way to verify the internal temperature
reading? Another possibility is a bad or badly attached heatsink, so
I'll try another one, but I'm not sure I believe this.

IF you find no apparent problems, including no windows
errors reported, it can still easily be possible that data,
calculations are erroneous. That is one reason to test with
Prime 95's Torture Test, because it checks the calculations.

Not having Windows to run the test, my main check has been: generate
1Gb (that's how much RAM I have) of ARC4 stream, store it in memory,
then generate it again (64 times), and check that it's consistant
(then repeat ad lib). This doesn't check the FPU so I should devise
something else for that.

Do not consider underclocking to be any different than
overclocking. Either is merely changing the clock rate.
The original clock rate is somewhat an arbitrary speed,
merely what the manufacturer guarantees to be stable but
once a core is mature, it's mostly just how many MHz you
paid for, the price-tiering of any given technology at any
point in time. Putting that in context, there certainly are
still limits to what any core can do and some specimens are
better than others, but towards the end of adjusting FSB,
voltage, cooling, etc, there is little difference between
overclocking and underclocking.
Right.

Also keep in mind that when changing bus speeds, it may
result in different memory speeds & timings too. For that
reason you should also test memory with Memtest86+ before
ever booting the OS.

I have ECC RAM, and my tests will check that no errors have been
detected/corrected by interrogating the northbridge. So far I haven't
found any problems.

 

[Michael W. Ryder]

First, many thanks for your reply, and thanks to all others also.




The environment isn't too unusual: room temperature is high (27C,
that's 81F - I also have an overheating problem there) but not that
high. Case ventilation seems adequate: I've just added a new fan at
the back, which improved things considerably as far as the graphics
chipset goes (went down from 70C to 63C when idle), but doesn't seem
to have changed much as far as the CPU goes. The system is rock
stable even when the overheating alarm triggers the throttle - I've
left it running that way for as long as I dared, performing very
stringent checks, and all checks passed. As for the actual
temperature reading, it's hard to say, because my motherboard sensors
aren't supported under Linux and that's all I've got. I can read the
temperature under the BIOS setup utility, and when I leave it running
for several minutes (the BIOS must be busy waiting on the CPU, because
it's definitely heating a lot) it runs up to ~80C.

The "bad or badly attached heatsink" is beginning to sound plausible,
so I'll try to buy a new one tomorrow, but even that is strange: the
heatsink came from Intel, together with the CPU, and the
whatever-you-call-those-little-plastic-things-which-hold-everything-in-place
seemed to fit in perfectly and make just the "right" sound when
locking in place.

I tried using "Arctic Silver 5" thermal material today, but it doesn't
seem to have helped any (in fact, it probably made things worse
because I couldn't follow the instructions to the letter).




I certainly can't avoid having _some_ kind of performance loss, but
maybe I can try to minimize it. For instance: only the CPU is
overheating, memory is fine, so maybe there's a way to tune the
timings so as not to underclock the memory. I'm definitely not an
expert there




Right now I lowered Vcore from the nominal 1.4V down to 1.2975V (the
lowest the motherboard would allow) and I downclocked the front-side
bus from 200MHz to 170MHz (so I'm running effectively at 2.9GHz
instead of 3.4GHz) - and still overheating if I push things far enough
(though now it takes minutes rather than seconds before the thermal
throttle sets in). The system is still perfectly stable, however, as
far as I could test it.




I wish I knew! One possibility is that the internal temperature diode
is utterly wrong and is throttling for no reason at all: is that known
to happen? Is there some way to verify the internal temperature
reading? Another possibility is a bad or badly attached heatsink, so
I'll try another one, but I'm not sure I believe this.

Another problem I had with this processor is that the new motherboard I
bought for it placed the CPU and heatsink too close to the power supply
in my previous case. There was less than an inch of clearance and I
could feel the heat being dumped into the power supply through the side
of the case. Replacing the case with one that gave me a couple of
inches of clearance made a marked difference in the temperatures.

 

[David Madore]

Another possibility is a bad or badly attached heatsink, so
I'll try another one,

Well, that worked. I bought new heatsink+fan combo, by Zalman, and I
gained (or lost, whatever I should say) nearly 15 degrees (Celsius).
I guess the plastic thingies which kept the previous heatsink in place
were somehow inadequate and it wasn't correctly attached to the CPU.

 

[George Macdonald]

Well, that worked. I bought new heatsink+fan combo, by Zalman, and I
gained (or lost, whatever I should say) nearly 15 degrees (Celsius).
I guess the plastic thingies which kept the previous heatsink in place
were somehow inadequate and it wasn't correctly attached to the CPU.

Hmm, I wonder if this could be anything to do with this:
http://www.theinquirer.net/?article=25267 ? Obviously the 3.73MHz is a
typo and not the same FSB as yours but.... makes ya wonder if the "hot
Prescott" got a bit of a bum rap because of a misdesigned heatsink
retention mechanism. Might be worth whinging at Intel on this.