Re: basic queries about overclocking

On Sun, 24 Jul 2005 17:26:42 +0100, Jon D <jon_d@nomail.com>
wrote:

>I am running a Duron Applebred 1800 in a Via 266 mobo by Syntax
>with 768 KB of 133 SD-RAM. CPU temp under load = 44 degrees C.
>Phoenix BIOS.
>
>My crappy mobo doesn't permit very many setting. (EG can't set cpu
>multiplier.)
>
>---------
>
>Can I check with you guys about some basic stuff:
>
>If I slightly increase the cpu voltage from the 1.52 V at present
>(nominal voltage for my Duron 1800 is 1.50 V) then will that help
>increase cpu stabilty?

If it is stable at 1.5V and you don't overclock, it should
not need changed. If you do overclock by any significant
margin, yes you're likely to need raise the voltage some BUT
because you can't change the multiplier, you will soon reach
a stability limit of the northbridge and memory- typically
that limit on Via KT266 is around 150-odd MHz FSB, at which
point your IDE (hard drives and data) are at risk, as well
as USB and network adapters starting to act flaky. Other
parts might too, I mention only the most common. Of
particular concern is the hard drive data, since it may
become corrupt entirely and you have to format and start
over from scratch. For that reason it's a VERY good idea to
make a complete backup of all HDD data if you seek to push
the FSB past (roughly) 145MHz.


> I can increase by only 25 mV, 50 mV or 75
>mV. I figured that 75 mV feels safe - QUESTION ONE: is this
>right?

Frankly, those are too little increase to get you very much
benefit. To fully exploit your chip you'd probably need to
up the voltage to around 1.8V, a 300mV increase. That
increases heat a lot though, and your board limits you too
much to get benefit from doing so.

Generally, you might go ahead and raise it the 75mA, then
proceed to do whatever things you were to overclock, then
once you have it confirmed 100% stable, try lowering the
voltage again and retest stability- there is no need to
raise the voltage if CPU stays stable at stock voltage.


>
>In the past with a slower processor, I increased the "CPU Host
>Clock" (which sounds to me like the FSB) from a starting value of
>100 to 104 (an extra 4 %) or a slightly unstable 106. I think
>memory was the limiting factor for stability this time.

It it was KT133 (not KT133A) then it was the northbridge
that limited you. Generally though, such a minor increase
is hardly worth fooling with. Generally it's good to just
shoot for the moon, ie- use a board that supports the next
higher FSB range and see if it'll post at the max you hope
to use. For example, with a Duron 1.8GHz you would have
13.5X multiplier (IIRC), so you'd raise the voltage to 1.8V
or so and raise the FSB to 166MHz/DDR333.

Since your motherboard won't do that, the options you have
left that are of enough gain to bother with will revolve
around finding another way to increase the CPU multiplier.
Sometimes a bios mod will give you that feature. Sometimes
there's a hidden (undocumented) jumper on boards, or plated
holes (soldered) where there "could" be pins for a jumper
but the manufacturer didn't add the jumpers. On your
particular board I have no idea. Other times you might need
use a low voltage and needles to burn a CPU multipler bridge
on the CPU (risky unless you had another board you could use
to confirm whether the CPU was stable at the new multiplier
before permanently modifying it). last there is the
wire-wrap or wire-socket method where you insert a piece of
wire in the socket or wrap it around two pins to change the
multiplier. Google should turn up some info on this, I
don't recall the specifics for that CPU.


>
>I have currently got all the memory settings at what I believe is
>their most aggressive:
> SDRAM cycle length = 2,
> Trp = 2T,
> Tras = 5T,
> Trcd = 3T.

Have you confirmed they're stable at those settings? If you
haven't yet, run Memtest86 for several hours. Keep in mind
that IF you decide to increase the FSB any (which will
probably be useful at least up to 140MHz or a little higher)
you may easily loose stability with those memory timings, so
you might want to set them more conservatively and then
later, after you've removed that variable and have checked
stability with memtest86, THEN reduce the timings again and
retest with memtest86. That is the general procedure, but
frankly, these days it's not much return in performance for
all the time spent. A newer motherboard, CPU, memory would
run circles around it.... if you spend hours just to get
another 10% performance gain that's not much benefit
relative to the 80% gain you'd get from newer parts.


>
>QUESTION TWO: Should I be able to increase the FSB to about the
>same value as before because, presumably, the faster new processor
>(Duron 1800 versus old Duro 700) should have no effect on the FSB
>setting?

Your prior CPU used 100MHz FSB (DDR200). The Duron 1800
uses 133MHz. As mentioned previously, you can probably
raise that to at least 140MHz, IF your memory stays stable,
but if you go much beyond 145MHz be sure your data is backed
up. After overclocking it and testing with memtest86, it
would be good to check CPU stability with Prime95's Torture
Test. That should run for at least several hours, or until
the very first error- at which point you can stop the test
and need to take appropriate meaures to regain stability and
begin testing again. Again, a lot of work for minimal gain
these days. Some people might prefer to just leave the CPU
at stock speed, 1.5V, and see how quiet they can get it, as
a Duron 1800 can be the foundation for a silent system, as
it produces less than 1/2 the heat of a modern P4.


>
>QUESTION THREE: Would I be able to increase the FSB bus more if I
>made the above memory settings less aggressive? This would seem
>to be the only way to increase the CPU speed.

yes, it seems I answered your questions before I'd ever read
them.

>
>Finally, is "Vcore 2" (reported by Motherboard Monitor 5)
>essentially only used by the CPU for I/O operations?

You cannot assume what MBM5 is reporting unless you have
compared it to a bios reading. MBM5, being generic in it's
ability to run with different boards, may report different
things for different people/systems. I wouldn't worry
about that voltage at all, you really only need be concerned
about the CPU vcore, memory voltage (if applicable, it may
not be on your board), and the power supply voltages such as
3.3V, 5V, 5VSB, 12V. I doubt yours even has chipset voltage
levels or reporting but if it does you can usually ignore
those on your board too, as the chipset itself is the
biggest limitation and it just won't go very far using
reasonable measures.

Hi Kony, thank you very much for your extensive reply. Like Wes,
I am particularly grateful for the hard numbers you guys mention
as that gives me some useful info.

I have put my other comments inline below.


> On Sun, 24 Jul 2005 17:26:42 , Jon D <jon_d@nomail.com> wrote:
>>
>> I am running a Duron Applebred 1800 in a Via 266 mobo by Syntax
>> with 768 KB of 133 SD-RAM. CPU temp under load = 44 degrees C.
>> Phoenix BIOS.
>>
>> My crappy mobo doesn't permit very many setting. (EG can't set
>> cpu multiplier.) Can I check with you guys about some basic
>> stuff:
>>
>> If I slightly increase the cpu voltage from the 1.52 V at
>> present (nominal voltage for my Duron 1800 is 1.50 V) then will
>> that help increase cpu stability?


On Sun 24 Jul 2005 21:35:58, kony wrote:
>
> If it is stable at 1.5V and you don't overclock, it should not
> need changed. If you do overclock by any significant margin,
> yes you're likely to need raise the voltage some BUT because you
> can't change the multiplier, you will soon reach a stability
> limit of the northbridge and memory- typically that limit on Via
> KT266 is around 150-odd MHz FSB, at which point your IDE (hard
> drives and data) are at risk, as well as USB and network
> adapters starting to act flaky. Other parts might too, I
> mention only the most common. Of particular concern is the hard
> drive data, since it may become corrupt entirely and you have to
> format and start over from scratch. For that reason it's a VERY
> good idea to make a complete backup of all HDD data if you seek
> to push the FSB past (roughly) 145MHz.


Hey man, I never gave any thought to my hard drives and the data
passing thru the IDE adapters. That is really important. Hmm, I
got to give that more thought.


>> I can increase by only 25 mV, 50 mV or 75 mV. I figured
>> that 75 mV feels safe - QUESTION ONE: is this right?
>
> Frankly, those are too little increase to get you very much
> benefit. To fully exploit your chip you'd probably need to up
> the voltage to around 1.8V, a 300mV increase. That increases
> heat a lot though, and your board limits you too much to get
> benefit from doing so.
>
> Generally, you might go ahead and raise it the 75mA, then
> proceed to do whatever things you were to overclock, then once
> you have it confirmed 100% stable, try lowering the voltage
> again and retest stability- there is no need to raise the
> voltage if CPU stays stable at stock voltage.


>> In the past with a slower processor, I increased the "CPU Host
>> Clock" (which sounds to me like the FSB) from a starting value
>> of 100 to 104 (an extra 4 %) or a slightly unstable 106. I
>> think memory was the limiting factor for stability this time.
>
> It was KT133 (not KT133A) then it was the northbridge that
> limited you.


I may have confused with detail. In fact I have the Via 266 (I
think that is my VT8367 northbridge with 8235 southbridge). But
don't worry because your point makes me realize that it is too
simplistic for me to see only cpu or memory as the items which
might get pushed past their limit.


> Generally though, such a minor increase is hardly
> worth fooling with. Generally it's good to just shoot for the
> moon, ie- use a board that supports the next higher FSB range
> and see if it'll post at the max you hope to use. For example,
> with a Duron 1.8GHz you would have 13.5X multiplier (IIRC), so
> you'd raise the voltage to 1.8V or so and raise the FSB to
> 166MHz/DDR333.


I guess I know what you are saying. However I was just fooling
around with these settings (sorry, I mean I was making impromptu
empirical investigations, heh!) and I noticed that by pushing the
FSB speed from 100 MHz on the old Duron 700 to about 105 or 106
MHz I was able to make my system noticeably SNAPPIER in response.
This SNAPPINESS was very welcome!


> Since your motherboard won't do that, the options you have left
> that are of enough gain to bother with will revolve around
> finding another way to increase the CPU multiplier. Sometimes a
> bios mod will give you that feature. Sometimes there's a hidden
> (undocumented) jumper on boards, or plated holes (soldered)
> where there "could" be pins for a jumper but the manufacturer
> didn't add the jumpers. On your particular board I have no
> idea. Other times you might need use a low voltage and needles
> to burn a CPU multiplier bridge on the CPU (risky unless you had
> another board you could use to confirm whether the CPU was
> stable at the new multiplier before permanently modifying it).
> last there is the wire-wrap or wire-socket method where you
> insert a piece of wire in the socket or wrap it around two pins
> to change the multiplier. Google should turn up some info on
> this, I don't recall the specifics for that CPU.


This is getting me into deep stuff and to be honest it nearly
killed me (and the mobo) to get the big heatsink and fan on
without smearing the thermal compound all over the place. So I
don't really want to go down there and fiddle with the cpu
hardware again unless I have to.


>> I have currently got all the memory settings at what I believe
>> is their most aggressive:
>> SDRAM cycle length = 2, Trp = 2T, Tras = 5T, Trcd = 3T.

>
> Have you confirmed they're stable at those settings? If you
> haven't yet, run Memtest86 for several hours.


Kony, you touch on something I tend to struggle with. In the past
I have tried to use Memtest but couldn't make head or tail of it.
It was confusing enough to work out the difference between:
Memtest86 ver 3.2 http://www.memtest86.com/
Memtest86+ ver 1.60 http://www.memtest.org/

In the end I decided it made little difference at my level which
one I ran. But I found that actually running Memtest was a
mystery. It ran and then seemed to loop and it never seemed to
come up with any conclusions. I wasn't sure what to press next or
whether any given test was completed.

I also tried a different "Memtest" from HCI Design but it seemed a
bit unsatisfactory. http://hcidesign.com/memtest/

I then figured that maybe I should try Microsoft's Windows Memory
Diagnostic at http://oca.microsoft.com/en/windiag.asp. However I
never got as far as trying that out.

So, can you or anyone else suggest which I should use and if you
think one of the two Memtest86's is best then is there anywhere
which explains in *simpler* language and with simpler steps (than
the README.TXT) what options to take while it is running.


> Keep in mind that IF you decide to increase the FSB any (which
> will probably be useful at least up to 140MHz or a little
> higher) you may easily loose stability with those memory
> timings, so you might want to set them more conservatively and
> then later, after you've removed that variable and have checked
> stability with memtest86, THEN reduce the timings again and
> retest with memtest86. That is the general procedure, but
> frankly, these days it's not much return in performance for all
> the time spent. A newer motherboard, CPU, memory would run
> circles around it.... if you spend hours just to get another 10%
> performance gain that's not much benefit relative to the 80%
> gain you'd get from newer parts.

Yeah, I know what you mean. If I can get this box running sweetly
then I'll start building a more serious box. The cost of memory
is off-putting.


-------------


>>QUESTION TWO: Should I be able to increase the FSB to about the
>>same value as before because, presumably, the faster new
>>processor (Duron 1800 versus old Duron 700) should have no
>>effect on the FSB setting?
>
> Your prior CPU used 100MHz FSB (DDR200). The Duron 1800 uses
> 133MHz. As mentioned previously, you can probably raise that to
> at least 140MHz, IF your memory stays stable, but if you go much
> beyond 145MHz be sure your data is backed up. After
> overclocking it and testing with memtest86, it would be good to
> check CPU stability with Prime95's Torture Test. That should
> run for at least several hours, or until the very first error-
> at which point you can stop the test and need to take
> appropriate measures to regain stability and begin testing
> again.
> Again, a lot of work for minimal gain these days. Some people
> might prefer to just leave the CPU at stock speed, 1.5V, and see
> how quiet they can get it, as a Duron 1800 can be the foundation
> for a silent system, as it produces less than 1/2 the heat of a
> modern P4.

You are dead right. Mobo Monitor reports my Duron 1800 running at
46 degrees C with no real noise from the cpu fan. The core temp
is probably actually higher because the mobo sensor is no doubt
poorly placed but even so it run pretty much as you describe.


--- remainder snipped out ---

On Sun, 24 Jul 2005 22:28:03 +0100, Jon D <jon_d@nomail.com>
wrote:

>
>Hey man, I never gave any thought to my hard drives and the data
>passing thru the IDE adapters. That is really important. Hmm, I
>got to give that more thought.

Didn't mean to alarm you, but it is common. Just be aware
that raising FSB introduces the potential for instability in
more than just CPU and memory.



>I may have confused with detail. In fact I have the Via 266 (I
>think that is my VT8367 northbridge with 8235 southbridge). But
>don't worry because your point makes me realize that it is too
>simplistic for me to see only cpu or memory as the items which
>might get pushed past their limit.

The main issue is your specific motherboard. It's maximum
"in spec" FSB is that you're already using, 133MHz/DDR266.
That is assuming I'm remembering correctly that it's the
default FSB speed for a Duron 1.8. Either way, 133MHz is
still the max FSB clock rate for KT266, anything more is
technically overclocking the board itself, and more.



>I guess I know what you are saying. However I was just fooling
>around with these settings (sorry, I mean I was making impromptu
>empirical investigations, heh!) and I noticed that by pushing the
>FSB speed from 100 MHz on the old Duron 700 to about 105 or 106
>MHz I was able to make my system noticeably SNAPPIER in response.
>This SNAPPINESS was very welcome!

Well a 100MHz FSB and memory bus is pretty slow for a modern
system. Indeed you could simply raise your FSB by 6MHz,
from 133 to 139MHz. _IF_ it stays stable then you get a
4-5% performance increase, though it's really not much, many
people might find their overall performance increased more
from a faster HDD. Generally to perceive a speed difference
a person needs about 10%, or so I've heard- no proof of it.


>This is getting me into deep stuff and to be honest it nearly
>killed me (and the mobo) to get the big heatsink and fan on
>without smearing the thermal compound all over the place. So I
>don't really want to go down there and fiddle with the cpu
>hardware again unless I have to.

.... then don't. Just enjoy the system till you find it's
time to upgrade or replace it again. In the interim you
could start accumulating knowledge about how to overclock
your next system-technology-of-choice, for example an Athlon
64 or whatever.



>Kony, you touch on something I tend to struggle with. In the past
>I have tried to use Memtest but couldn't make head or tail of it.
>It was confusing enough to work out the difference between:
> Memtest86 ver 3.2 http://www.memtest86.com/
> Memtest86+ ver 1.60 http://www.memtest.org/

I used mostly memtest86, not memtest86+, but the + version
has more support for new boards. Either should work with
your system. They're meant to be easy to use, easiest way
being to simply make a floppy and boot to it. Just let it
run for several hours or until you see if listing any
errors. If it doesn't show any errors after several hours,
just turn the system off and boot windows. If it does show
errors, you need to readjust some setting to get the system
stable again BEFORE booting windows. By not booting windows
until you have some minimal assurance of memory stability,
you greatly decrease the chances of file/data corruption-
not due to the aforementioned hard drive/IDE issue but a
more general type of random corruption from CPU errors or
memory errors.


>
>In the end I decided it made little difference at my level which
>one I ran. But I found that actually running Memtest was a
>mystery. It ran and then seemed to loop and it never seemed to
>come up with any conclusions. I wasn't sure what to press next or
>whether any given test was completed.

That's exactly what it's supposed to do. The goal is to
just let it loop over and over. If it shows no errors
that's a "Pass". If it shows ANY errors, readjust the
system and rerun memtest86 till it doesn't show any errors.
This part of overclocking, the testing, can take a long time
and is among the reasons I suggest it may not be worth the
time to try to aggressively overclock the system, since
there may not be so much of a performance gain. Even if you
managed to change the CPU multiplier, the higher you
overclock the CPU the more the PC133 memory will be a
bottleneck.


>
>I also tried a different "Memtest" from HCI Design but it seemed a
>bit unsatisfactory. http://hcidesign.com/memtest/
>
>I then figured that maybe I should try Microsoft's Windows Memory
>Diagnostic at http://oca.microsoft.com/en/windiag.asp. However I
>never got as far as trying that out.

Memtest86+ seems the best alternative. Anything that
requires you to boot to windows to run it is flawed by that
very fact alone. Booting a single time with lots of memory
errors can trash your registry then you have a lot of work
to salvage your windows installation or have to reinstall
it.


>
>So, can you or anyone else suggest which I should use and if you
>think one of the two Memtest86's is best then is there anywhere
>which explains in *simpler* language and with simpler steps (than
>the README.TXT) what options to take while it is running.

Just let it run, it will do fine "hands off". All you have
to do is check it to see if any errors are reported. If
there's any errors you can go ahead and stop testing because
it's not stable. You might note how long it had ran before
any errors as a rough gauge of how instable it is. For
example, if you had 12,000 errors within 5 minutes, you'd
likely have to make more significant changes, slow the bus
speeds or raise the memory timings more than if you only had
5 errors showing up in 4 hours.



>Yeah, I know what you mean. If I can get this box running sweetly
>then I'll start building a more serious box. The cost of memory
>is off-putting.

I'd just make it as quiet as reasonably possible. Then
it'll have longer fan lifespan, slower dust buildup, in
addition to being less distracting.



>You are dead right. Mobo Monitor reports my Duron 1800 running at
>46 degrees C with no real noise from the cpu fan. The core temp
>is probably actually higher because the mobo sensor is no doubt
>poorly placed but even so it run pretty much as you describe.

Poor sensor placement could be too high OR too low, though
it likely is too low because your board is designed for
earlier CPUs that produced less heat. Even so, it "could"
be fairly accurate but only slower to respond to temp
changes. 46C sounds roughly right though, depending on the
other factors like the 'sink, fan speed, chassis cooling,
etc.