Has my overclocked CPU "passed"?

A

AN O'Nymous

I overclocked my Athlon XP 2500+ Barton CPU on an nForce2 Abit NF7 mobo
to a 3200+ one by changing the FSB frequency from 166MHz -> 200MHz,
multiplier fixed at 11 and upping the CPU voltage from 1.65V to 1.8V. I
am using the retail 2500+ stock cooler, but have set my BIOS to sound
an alarm and eventually shut down the PC at 75 Celcius.

I did a 6 hour test with "Hot CPU Tester Pro (Lite Edition)". It
involved 28 repeated runs of "Complex matrix, Calculating Pi, Sorting
algorithms, Prime test, Memory, HD, MMX" tests. No errors were
detected.

Can I safely assume my PC's overclock is working? If I got proper
cooling, how much higher can I overclock this CPU? Is temperature the
only limitation in overclocking, or are other factors important (e.g.
dielectric breakdown of the semiconductor)?

Thanks.
 
P

philo

Can I safely assume my PC's overclock is working? If I got proper
cooling, how much higher can I overclock this CPU? Is temperature the
only limitation in overclocking, or are other factors important (e.g.
dielectric breakdown of the semiconductor)?

the main factor is temperature...
but there is always some risk with overclocking...
is the relatively small improvement in performance really worth risking
your investment in the cpu?

if it's working fine the way you have it setup...
I'd not risk trying to squeeze the "last ounce" of performance out of it
 
S

Sjouke Burry

philo said:
the main factor is temperature...
but there is always some risk with overclocking...
is the relatively small improvement in performance really worth risking
your investment in the cpu?

if it's working fine the way you have it setup...
I'd not risk trying to squeeze the "last ounce" of performance out of it
Every 6 degr Celcius halves the expected livetime of
a silicon semiconductor, for example a power transistor
at full load will live only a few years.
At 25 to 50 percent expected live time is tens to
hundreds of years.
A processor with millions of components needs only
one or two weak parts,and overclocking will kill it
quickly.(but you can always hope to be lucky....)
 
K

kony

the main factor is temperature...
but there is always some risk with overclocking...
is the relatively small improvement in performance really worth risking
your investment in the cpu?

It's all relative- why do people buy newer systems when they
only get relatively small improvements in many areas? Why
do higher priced CPUs sell for multiple times as much when
there is only a relatively small improvement?

Overclocking that chip is a good idea, there is every reason
to believe it will be a free performance gain. That is,
except for the part of more thorough testing and determining
if 1.8V is really necessary or if it might run lower.
Higher voltage also puts strain on the motherboard, if the
capacitors in particular feel more than warm during this
stress-testing and if the system is expected to be under
fair load (continuous processing types of apps like games or
video encoding), then the lifespan of the board could become
unacceptible, or at least further measures to cool it might
help.

if it's working fine the way you have it setup...
I'd not risk trying to squeeze the "last ounce" of performance out of it

That is a reasonable and conservative suggestion that is
best for mission-critical systems. Even so, these same
systems should be stress-tested before being assumed any
better than an overclocked one.
 
K

kony

Every 6 degr Celcius halves the expected livetime of
a silicon semiconductor, for example a power transistor
at full load will live only a few years.
At 25 to 50 percent expected live time is tens to
hundreds of years.
A processor with millions of components needs only
one or two weak parts,and overclocking will kill it
quickly.(but you can always hope to be lucky....)


Generally, no. Reasonable levels of overclocking are not
expected to kill any CPU quickly. This has been proven by
millions of overclocks for years. The main issue is not
going overboard with voltage to the extent that it cannot be
cooled effectively.
 
A

AN O'Nymous

kony said:
It's all relative- why do people buy newer systems when they
only get relatively small improvements in many areas? Why
do higher priced CPUs sell for multiple times as much when
there is only a relatively small improvement?

Yes, completely agree. I was about to buy a cheap AMD64 754-pin Sempron
because I just need a decent number cruncher for the next two years or
so. In the end I decided that I might as well take a risk overclocking
my present 2500+ Barton. The app I run it with has parallel processing
& 64-bit capability, so the additional Sempron would have fit in fine,
in addition to my Barton and another AMD64 3400+ CPU (and another
24-node Beowulf coming soon).
Overclocking that chip is a good idea, there is every reason
to believe it will be a free performance gain. That is,
except for the part of more thorough testing and determining
if 1.8V is really necessary or if it might run lower.

I already did tests. The stock voltage for an AMD XP 2500+ Barton is
1.65V. I started at 1.7V because that's what a review said worked for
them. My PC still spontaneously rebooted. 1.775V was OK in Windows,
except in Linux where it froze (strange how stability wrt
voltage/overclocking is voltage dependent!).

1.8V seems OK ATM, I will need to find CPU testers in Linux for that
and haven't gotten around to it yet.
Higher voltage also puts strain on the motherboard, if the
capacitors in particular feel more than warm during this
stress-testing and if the system is expected to be under
fair load (continuous processing types of apps like games or
video encoding), then the lifespan of the board could become
unacceptible, or at least further measures to cool it might
help.

I've bought 3 80cm fans for my case, in addition to the PSU's fan.

With a badly designed cheap case (a mistake...I thought the case wasn't
important), do you suggest getting all the fans to blow out or...? ATM
I've got 1 in the front blowing in, 2 at the rear blowing out.
That is a reasonable and conservative suggestion that is
best for mission-critical systems. Even so, these same
systems should be stress-tested before being assumed any
better than an overclocked one.

Yes, I do run the CPU back at the stock speeds if I go near data
critical apps (don't even access the files on the HDD otherwise).

It is the long overnight runs where I start overclocking.
 
M

~misfit~

AN said:
Yes, completely agree. I was about to buy a cheap AMD64 754-pin
Sempron because I just need a decent number cruncher for the next two
years or so. In the end I decided that I might as well take a risk
overclocking my present 2500+ Barton. The app I run it with has
parallel processing & 64-bit capability, so the additional Sempron
would have fit in fine, in addition to my Barton and another AMD64
3400+ CPU (and another 24-node Beowulf coming soon).


I already did tests. The stock voltage for an AMD XP 2500+ Barton is
1.65V. I started at 1.7V because that's what a review said worked for
them. My PC still spontaneously rebooted. 1.775V was OK in Windows,
except in Linux where it froze (strange how stability wrt
voltage/overclocking is voltage dependent!).

1.8V seems OK ATM, I will need to find CPU testers in Linux for that
and haven't gotten around to it yet.

Strange how CPUs can vary. I'm currently using an XP2500+ at 11 x 200MHz at
stock vcore, 1.65V. It runs Prime95 all night error-free and is rock-solid.
(However, my PSU is excellent)

I look at 1.8V as the absolute maximum for this family of CPUs, especially
air-cooled.

One thing I noticed first-hand in my early overclocking days, you can often
get by with a lower vcore with a better PSU. If you're using a cheapie or
underpowered one they can fluctuate and not only make your O/C unstable but
put extra strain on the voltage regulatory circuitry on the mobo.

Moral: Have a good name-brand PSU of sufficient capacity.
I've bought 3 80cm fans for my case, in addition to the PSU's fan.

With a badly designed cheap case (a mistake...I thought the case
wasn't important), do you suggest getting all the fans to blow out
or...? ATM I've got 1 in the front blowing in, 2 at the rear blowing
out.

I also used a cheap case at first. The one thing I did that made all the
difference was to put a small mark on the case side directly in-line with
the CPU heatsink centre. Then I took the side to a local engineering shop
and got them to drill a 3" hole. I then marked it out for the four
screw-holes for an 80mm fan and drilled it and fitted the fan and a finger
guard. Now, the sneaky bit; I carefully cut a section of plastic cup (after
looking at all sorts of bottles and containers) that was 80mm at the wide
end, around 60mm at the narrow end and just the right length (55mm in my
case) so that when I hot-glued it to the 80mm fan it had around 5mm
clearance to the CPU fan. You can see it here:

http://test.internet-webmaster.de/upload/1138958407.jpg

That reduced my CPU core temp by around 10°C. Then I had two fans blowing
in, that one and one cooling the HDDs at the front, and one fan blowing out
below the PSU. With the PSU fan blowing out as well it worked out that the
case was under slight positive pressure. That's good as, if you have
negavtive pressure, you often get dust sucked into drives etc.

Good luck with your CPU. That case the picture is of now has a Tbred B JUIHB
11/'03 XP1800+ in it running at stock vcore (another good PSU) at 10 x
200MHz, 2GHz (up from standard 1.53GHz). It benchmarks as well as an
XP2800+. The faster FSB makes up for the smaller L2 cache. It never reaches
50°C (core temp), even after encoding an XviD all day in summer.
 
K

kony

I've bought 3 80cm fans for my case, in addition to the PSU's fan.

With a badly designed cheap case (a mistake...I thought the case wasn't
important), do you suggest getting all the fans to blow out or...? ATM
I've got 1 in the front blowing in, 2 at the rear blowing out.

I seldom advocate a brute-force add-more-fans-to-case
strategy. I have, actually I forget the count but at least
3 Mobile Bartons o'c past 2.3Ghz. None have excessively
loud case fans but to combat capacitor heating they have
either large (92 x 25mm) fans on the CPU heatsink which
blow-by the CPU regulation area on the board (where the
susceptible caps are) or an additional fan devoted to
cooling them. For example,
http://69.36.189.159/usr_1034/nb_sinks/1.jpg

Ultimately, you may not need such a fan, but to make the
determination you need to check the temp of the caps during
an extended full load "stress test" type of application.


Yes, I do run the CPU back at the stock speeds if I go near data
critical apps (don't even access the files on the HDD otherwise).

It is the long overnight runs where I start overclocking.

part of the key in overclocking is knowing what the ceiling
is for a particular specimen. I don't just crank up voltage
and speed till I hit that highest # possible, rather
finding where the break-points are, voltage vs speed...
towards the end there's a lot of voltage needed for minor
additional speed increase, then having the desired
compromise in mind, I'll reduce the o'c a little and bump
the voltage up another 0.025V for extra margin. IE- AFTER
confirming the system was completely stable this is
additional margin. However, this is for non-critical
purposes, systems devoted to gaming, video capture, etc- for
serious work I'm more interested in absolute minimum heat
so the system is completely inaudible.
 
M

~misfit~

~misfit~ wrote:

Strange how CPUs can vary. I'm currently using an XP2500+ at 11 x
200MHz at stock vcore, 1.65V. It runs Prime95 all night error-free
and is rock-solid. (However, my PSU is excellent)

And, just for the hell of it, (My DSL was down) I changed the settings this
morning to 210MHz FSB x 10.5, still default vcore (and ~2.2GHz) and same RAM
timings. Ran Prime 95 for 3 hours fine, I stopped it then as I need the PC,
will leave it running all night tonight.

It is my considered opinion that faster FSB is always better at a given CPU
speed. I didn't benchmark this PC yesterday and today but I know from
experience that it'll be capable of doing mork work on the faster FSB. It'll
probably bench 5% faster. Not a lot but every bit helps when you consider
that this PC will likely have to last me several years yet.
 
A

AN O'Nymous

~misfit~ said:
Strange how CPUs can vary. I'm currently using an XP2500+ at 11 x 200MHz at
stock vcore, 1.65V. It runs Prime95 all night error-free and is rock-solid.
(However, my PSU is excellent)

Yes, strange indeed.
I look at 1.8V as the absolute maximum for this family of CPUs, especially
air-cooled.

Out of curiosity, how did you arrive at this figure? I was thinking
about going a bit higher with proper cooling and I'm glad I read your
post :-O
One thing I noticed first-hand in my early overclocking days, you can often
get by with a lower vcore with a better PSU. If you're using a cheapie or
underpowered one they can fluctuate and not only make your O/C unstable but
put extra strain on the voltage regulatory circuitry on the mobo.
Moral: Have a good name-brand PSU of sufficient capacity.

My PSU is a Antec SL350PGB 350W ATX P4 Smartpower Dual Fan Power
Supply.

Is this considered a good PSU?

I think my CPU's heat problems were because of me not putting on the
thermal grease properly. I removed the stock pad and basically put a
whole syringe of the thermal paste on the CPU. The proper way to do it,
I've since read, is to clean the CPU from the old paste, lightly clean
& sand the heatsink and then put a *drop* of the stuff on the CPU.

I also used a cheap case at first. The one thing I did that made all the
difference was to put a small mark on the case side directly in-line with
the CPU heatsink centre. Then I took the side to a local engineering shop
and got them to drill a 3" hole. I then marked it out for the four
screw-holes for an 80mm fan and drilled it and fitted the fan and a finger
guard. Now, the sneaky bit; I carefully cut a section of plastic cup (after
looking at all sorts of bottles and containers) that was 80mm at the wide
end, around 60mm at the narrow end and just the right length (55mm in my
case) so that when I hot-glued it to the 80mm fan it had around 5mm
clearance to the CPU fan. You can see it here:

Thanks. Sometimes I use "natural" cooling, i.e. open my case up. It
lets dust in though :)
 
A

AN O'Nymous

kony said:
I seldom advocate a brute-force add-more-fans-to-case
strategy.

When my AMD 2500+ Barton isn't overclocked and idling, its temperature
according to Sisoft Sandra is 49.5 Celcius. My room temperature is 21
Celcius. Would you say this idle temperature is high?
 
M

~misfit~

AN said:
Yes, strange indeed.

Yep. My ex has an XP2500+ that won't do 200MHz FSB X 11 stably without 1.750
volts. (And around 8°C higher temps than mine) We bought them a week apart
at the same shop, from the same shipment. We decided to leave it on 200 X
10.5 when she moved out, (she's not as on-to-it when it comes to monitoring
and maintaining as I am) it'll do that at 1.675V.
Out of curiosity, how did you arrive at this figure? I was thinking
about going a bit higher with proper cooling and I'm glad I read your
post :-O

I merely relate what I've been told. :) I've spent a lot of time over the
years in the overclocking groups and have overclocked a few of the later
Socket A CPUs myself and accepted wisdom is that, if it ain't gonna do it at
1.8v, it ain't gonna do it without massive cooling. (Water cooling, Peltier,
Phase-change). Above 1.8v there is little improvement in stability and very
large improvement in heat production. Basically you're hitting the danger
area where you could damage your CPU if you're unlucky.
My PSU is a Antec SL350PGB 350W ATX P4 Smartpower Dual Fan Power
Supply.

Is this considered a good PSU?

I think it's OK if it hasn't got a beast of a graphics card running off it.
Antec have a good repuation for their cases. I'm not familiar with it
personally, maybe Kony will say, he knows his PSUs. Personally I use mainly
AcBels and regard 450W as minimum. Especially as this sort of PC is usually
running a reasonably hungry graphics card.
I think my CPU's heat problems were because of me not putting on the
thermal grease properly. I removed the stock pad and basically put a
whole syringe of the thermal paste on the CPU. The proper way to do
it, I've since read, is to clean the CPU from the old paste, lightly
clean & sand the heatsink and then put a *drop* of the stuff on the
CPU.

Ok, that doesn't sound good. Basically, what you're aining for is to have as
much of the CPU core in contact with the HS as possible. I hope when you say
"sand" the heatsink you mean with very fine wet-and-dry paper, (400 grit
minimum, 800 is better) used wet. It's called "lapping" in the jargon and is
more polishing than sanding. The abrasive paper is layed flat on a piece of
glass to provide a flat surface, taped down and then moistened well. The
heatsink is then laid flat onto the paper and moved in smooth, regular
circles. Figure 8's are good. It's a PITA.

Yeah, remove all traces of the old TIM (Thermal Interface material), by
scraping, then by rubbing with a cloth soaked in meths or similar. Lapping
isn't often needed as most HS's are pretty good. Back in the day they were
rough with deep machining marks on them and lapping was considered de
riguer. Now, I don't bother. The purpose of the TIM or paste is only to fill
any air gaps there may be due to an imperfect contact. Therefore you want
the bare minimum. a blob the size of a grain of rice is plenty on these
bare-core CPUs, extravagant even. I put the TIM on the core and then, with
cling-film over my fingertip, spread it over the surface of the core as
evenly as I can, ideally leaving a layer about the thickness of the
cling-film. Then I rub what's on my finger into the area of the HS that the
core is going to contact. That's all you need. I place the first side of the
clip on the socket without letting the core touch the CPU, then sit it flat,
sliding it maybe a 16th of an inch back and forth to ensure the paste is
spread evenly. Then clip the other side down.

If you want a great, inexpensive alternative HSF, Thermaltake have a model,
TR2-M3. It's under $10 and (as long as you're sticking under the 1.8V
guideline) ample for any Socket A CPU. It has an 80mm fan so is very quiet
for the amount of air it moves in comparison to the AMD stock 60mm
screamers.
Thanks. Sometimes I use "natural" cooling, i.e. open my case up. It
lets dust in though :)

I find that the ducted system I use gets me lower temps than just having the
side off, by about 5°C. Yeah, you have to carefully remove the fan from the
HS every few months (without disturbing the TIM) and clean the dust from the
fan and fins. I use a large artist's brush for the job.

Good luck. Overclocking is fun once you know what you're doing. :)
 
M

~misfit~

AN said:
When my AMD 2500+ Barton isn't overclocked and idling, its temperature
according to Sisoft Sandra is 49.5 Celcius. My room temperature is 21
Celcius. Would you say this idle temperature is high?

Yes. Although if that's still at 1.8V it's understandable. Still too high
though. I strongly recommend you download and use Motherboard Monitor 5.
I've found it to be the most accurate of the monitoring programs, I have my
CPU core temp and load % constantly displayed in my systray.. Sandra can
often read the wrong sensor.

My 2500+ @ 3200+ speed idles at 10°C above room temp. Case shouldn't be much
more than 5° above room. You're talking an idle speed of 28.5°C above room,
that's insane. 15° would be considered acceptable in a case that isn't
ducted like the one I posted the pic of (although mine isn't ducted anymore,
I've used that case for my O/Ced XP1800+, I was given a Thermaltake Xaser
III case that my Barton lives in now).

I suggest you remove the HSF and clean and re-fit.

Good luck.

Heh. Sorry to butt in Dave.
 
K

kony

Yes, strange indeed.


At 11 x 200, 2.2GHz, a lot of cores would run ok at 1.65V.
It was borderline on average over the family, rasing the MHz
the next 300 or so required significant vcore increase.

It can depend on the motherboard too. 'Tis not the avg.
voltage that matters as much as the minimum. Suppose your
board is set for 1.65V, but it dips .09V. Suppose another
board only dipped .06V, and the threshold was .07V. The
former board will need at least 0.025V incremental (since
vcore setting is always in steps). Further, the more
marginal board may have even more deviation in vcore drop
the higher the vcore is set. You might find one board needs
1.85V for the same o'c another can do at 1.775V.

Out of curiosity, how did you arrive at this figure? I was thinking
about going a bit higher with proper cooling and I'm glad I read your
post :-O

That is a good ballpark figure, though if you had a really
good heatsink you can push a little further. On one box
I've an o'c Barton in, an Alpha Pal8045 (which was
supposedly great for it's era) can't cool sufficiently at
1.85V, 2.4GHz. However, this is also within context of a
reasonable (quiet) fan operation, I always determine the max
fan noise I would tolerate first, THEN see how high it'll
go.

On another system with a Thermaltake SLK-900U, very similar
CPU can run all day (and does, it was doing batch video
encoding), full load, without problem at 1.85V. Keep in
mind an SLK-900U was supposedly a $48 heatsink which is a
bit excessive in my mind but I'd happened upon a Newegg
Refurb'd specimen for $25. Now that the system isn't
encoding video anymore, a ~ 1200 RPM and lower o'c makes it
inaudible.

Seldom is it only about what voltage, but at what cost to
hear, pay, or lifespan of the motherboard.


My PSU is a Antec SL350PGB 350W ATX P4 Smartpower Dual Fan Power
Supply.

Is this considered a good PSU?

It's not bad but a bit marginal for high overclocking if you
have other fair loads on it such as a gaming video card or a
lot of HDDs.

I think my CPU's heat problems were because of me not putting on the
thermal grease properly. I removed the stock pad and basically put a
whole syringe of the thermal paste on the CPU. The proper way to do it,
I've since read, is to clean the CPU from the old paste, lightly clean
& sand the heatsink and then put a *drop* of the stuff on the CPU.

The need to lap a 'sink can depend a lot on how well it came
finished from the factory. I've lapped a 'sink till it was
so good the CPU was within 1 degree of same temp with no
thermal compound at all... and then decided it wasn't worth
the effort to lap it so well. IOW, getting the surface
fairly flat is the more important part, a mirror finish
isn't nearly as important as a minimal amount of good
medium-low viscosity synthetic thermal compound. The
synthetic compound is far more important on these open core
CPUs as the thermal density is much higher and will degrade
performance of std. silicone based componds over several
months. With a stock speed CPU, there's more margin and a
few degrees temp rise may not matter but when overclocking
it becomes more necessary to have a habit of using good
parts- not necessarily premium though, Arctic Silver 5 is
not really necessary, Arctic Alumina would work too or
another 3rd party compound that's reasonable quality, thin
enough, and synthetic based.

I never use thermal pads though- far easier to just clean it
off the 'sink before it ever melts onto the CPU.


Thanks. Sometimes I use "natural" cooling, i.e. open my case up. It
lets dust in though :)


If you have a spare system, or when you begin building your
next one, take the time to get creative with airflow and
filtering. Given ample exhaust fan(s), rate, or pusher fans
in front with a large filter area, there should be no need
to keep the case open. Just beware of altering airflow too
much, if you reduce the amount coming in through the case
front, it may cause the HDDs to run hotter.
 
K

kony

When my AMD 2500+ Barton isn't overclocked and idling, its temperature
according to Sisoft Sandra is 49.5 Celcius. My room temperature is 21
Celcius. Would you say this idle temperature is high?


"Most" socket A boards don't use HALT instructions to cool
the CPU when the system is idle. With poor case cooling, a
poor heatsink, or very slow heatsink fan your temp would be
expected. It is a little higher than ideal though,
particularly if it rises much under full load.

In other words, forget about the idle temp, the important
temp is that attained under extended full load operation.
Use something like Prime95's Torture Test, large in-place
FFTs setting, to heat it up for a good 30 minutes or more
then check the temp. Leave that running for a few hours to
see if there are any errors as it is a very useful test for
o'c systems. Many systems which are seemingly stable will
fail at the specific Prime95 test above within a few
minutes.

After getting a full load reading, you're faced with the
situation that some CPUs retain stability at higher temps
than others... can depend on how close they are to their
ceiling speed too. Generally anything over 60C is a sign
you need to improve the cooling starting at the weakest
areas first.
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Similar Threads


Top