Sign in with
Sign up | Sign in
Your question

How to run AMD K6-2 500mHz on an ECS P5VP-A+ board?

Last response: in Motherboards
Share
February 10, 2005 1:32:52 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hi folks,

I hope someone around here has a board as old as mine (1999).

The subject of this message about says it all. I understand that ECS
added support for multipliers above 4.5x after the manual was printed
out and processors became faster.

I searched Google Groups and a fellow named Steve Moores gave a link
to ECS's website which had details about how to set the jumpers, but
unfortunately, ECS has removed all support for this board from their
website. Creepy... like the board no longer exists.

Does anyone know what jumper settings I need to use for a 5.0x
multiplier?

I don't want to use the existing 4.5x multiplier with a higher system
bus, since that will move my AGP speed from 66 to 75, which my video
card manufacturer says is too high.

Anyone? Please? Thanks!

gene
February 11, 2005 2:47:18 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Thu, 10 Feb 2005 10:32:52 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hi folks,
>
>I hope someone around here has a board as old as mine (1999).
>
>The subject of this message about says it all. I understand that ECS
>added support for multipliers above 4.5x after the manual was printed
>out and processors became faster.
>
>I searched Google Groups and a fellow named Steve Moores gave a link
>to ECS's website which had details about how to set the jumpers, but
>unfortunately, ECS has removed all support for this board from their
>website. Creepy... like the board no longer exists.
>
>Does anyone know what jumper settings I need to use for a 5.0x
>multiplier?
>
>I don't want to use the existing 4.5x multiplier with a higher system
>bus, since that will move my AGP speed from 66 to 75, which my video
>card manufacturer says is too high.
>
>Anyone? Please? Thanks!
>
>gene

Actually, 75 vs 66 is not a high overclock and would give better
performance than the 5x multiplier... but

There isn't anything 'additional' needed to 'support' a 5x multiplier
since what the multiplier jumpers 'mean' is determined by the CPU.
E.g. it's the K6-2 that knows what 5x is, not the motherboard.

The original Pentium multiplier settings, were 2 jumpers (4 possible
codes)

BF1 BF0 multiplier

0 0 2.5x
0 1 3.0x
1 0 2.0x
1 1 1.5x

The MMX processor interprets the original 1.5x setting to be 3.5x,
since 1.5 was no longer needed, and is an example of the processor
being the determining factor. So the exact same jumpers became:


BF1 BF0 multiplier

0 0 2.5x
0 1 3.0x
1 0 2.0x
1 1 3.5x


AMD added a third jumper for super socket 7, adding more possible
codes, for

BF2 BF1 BF0 multiplier
1 0 0 3.5x
1 0 1 2.0x
1 1 0 2.5x
1 1 1 3.0x

0 0 0 4.5x
0 0 1 5.0x
0 1 0 4.0x
0 1 1 5.5x

Later K6-2 processors changed 2.0x to be 6.0x (so 400MHz could be
reached with a 66 MHz FSB)

Now, the BF lines are the signals on the processor pins and jumpers
usually tie a line to '0', with open being a 1. Some use two position
jumpers with one position being 1 and the other being 0. You'll have
to look at the jumper positions you know, compare to the tables above,
and determine which is BF0, BF1, BF2, then extrapolate to the
undocumented settings.
February 11, 2005 2:58:14 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Thu, 10 Feb 2005 23:47:18 -0600, flipper <flipper@fish.net> wrote:

Hi flipper,

First off, thank you so much for taking the time to explain all this
to me. It was very helpful for me to figure out all the jumper
settings that I needed.

>
>Actually, 75 vs 66 is not a high overclock and would give better
>performance than the 5x multiplier... but

Yes, I've read that some place, but wasn't sure. I checked with
Matrox to see if I could run the card at that speed, and they said no,
so I'll leave it at 66. I don't use the computer much for graphics
anyway. I have a newer computer that I use for most everything.

>
>AMD added a third jumper for super socket 7, adding more possible
>codes, for
>
>BF2 BF1 BF0 multiplier
>1 0 0 3.5x
>1 0 1 2.0x
>1 1 0 2.5x
>1 1 1 3.0x
>
>0 0 0 4.5x
>0 0 1 5.0x
>0 1 0 4.0x
>0 1 1 5.5x
>
>Later K6-2 processors changed 2.0x to be 6.0x (so 400MHz could be
>reached with a 66 MHz FSB)


This chart was extremely useful for me. Thanks for taking the time to
format and post it. I grabbed a pen and paper thinking it would be an
easy job to extrapolate the additional settings.

Really, it wasn't too difficult, but it was trickier than I thought.
The ECS P5VP-A+ board uses a three-pin set-up that you referred to vs.
a linked/open jumper system.

It was a matter of determining the BF0, BF1, and BF2 configuration and
then figuring out where the jumper should be placed for the zero and
one settings in your chart. Then it was just a matter of confirming
that all the settings jived with the chart. And, lo and behold, they
do. Yippee!

So, just in case somebody ever wants to figure out the settings for
this board again, I'll post the settings for any google groups
searchers.

1.5x 1-3 2-4 10-12
2.0x 1-3 4-6 10-12
2.5x 3-5 4-6 10-12
3.0x 3-5 2-4 10-12
3.5x 1-3 2-4 10-12
4.0x 1-3 4-6 12-14
4.5x 3-5 4-6 12-14
5.0x 3-5 2-4 12-14
5.5x 1-3 2-4 12-14

And, I guess, like you say, if someone needs 6.0x, it would be the
same as 2.0x. So:

6.0x 1-3 4-6 10-12

I also notice that 1.5x is the same as 3.5x. I guess that's the
result of the multiplier being controlled by the CPU, like you said,
not the motherboard.

Once again, thank you very much flipper. I'll try this out when I get
my "new cutting-edge" AMD K6-2 500 MHz chip. By the way, I only
bought it because I needed a new cpu fan, and the cpu with fan was
about $10 more than a fan alone.

gene
February 11, 2005 9:15:41 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Fri, 11 Feb 2005 11:58:14 -0600, gene <noemailplease@nowhere.com>
wrote:

>On Thu, 10 Feb 2005 23:47:18 -0600, flipper <flipper@fish.net> wrote:
>
>Hi flipper,
>
>First off, thank you so much for taking the time to explain all this
>to me. It was very helpful for me to figure out all the jumper
>settings that I needed.
>
>>
>>Actually, 75 vs 66 is not a high overclock and would give better
>>performance than the 5x multiplier... but
>
>Yes, I've read that some place, but wasn't sure. I checked with
>Matrox to see if I could run the card at that speed, and they said no,

Of course they will. Manufacturers aren't going to advise using their
equipment other than as specified.

>so I'll leave it at 66. I don't use the computer much for graphics
>anyway. I have a newer computer that I use for most everything.

It wouldn't improve the graphics to take note of. Where it would make
a modicum of different is in memory speed, FSB, and the PCI bus. FSB
is particularly important with processors that use motherboard cache
(an in all Pentium classic, MMX, K6, and K6-2 processors) because
that limits the cache speed to the processor.


>>
>>AMD added a third jumper for super socket 7, adding more possible
>>codes, for
>>
>>BF2 BF1 BF0 multiplier
>>1 0 0 3.5x
>>1 0 1 2.0x
>>1 1 0 2.5x
>>1 1 1 3.0x
>>
>>0 0 0 4.5x
>>0 0 1 5.0x
>>0 1 0 4.0x
>>0 1 1 5.5x
>>
>>Later K6-2 processors changed 2.0x to be 6.0x (so 400MHz could be
>>reached with a 66 MHz FSB)
>
>
>This chart was extremely useful for me. Thanks for taking the time to
>format and post it. I grabbed a pen and paper thinking it would be an
>easy job to extrapolate the additional settings.
>
>Really, it wasn't too difficult, but it was trickier than I thought.
>The ECS P5VP-A+ board uses a three-pin set-up that you referred to vs.
>a linked/open jumper system.

Yes, I see. A two position, 1 and 0, kind of jumper.

>
>It was a matter of determining the BF0, BF1, and BF2 configuration and
>then figuring out where the jumper should be placed for the zero and
>one settings in your chart. Then it was just a matter of confirming
>that all the settings jived with the chart. And, lo and behold, they
>do. Yippee!
>
>So, just in case somebody ever wants to figure out the settings for
>this board again, I'll post the settings for any google groups
>searchers.
>
>1.5x 1-3 2-4 10-12
>2.0x 1-3 4-6 10-12
>2.5x 3-5 4-6 10-12
>3.0x 3-5 2-4 10-12
>3.5x 1-3 2-4 10-12
>4.0x 1-3 4-6 12-14
>4.5x 3-5 4-6 12-14
>5.0x 3-5 2-4 12-14
>5.5x 1-3 2-4 12-14
>
>And, I guess, like you say, if someone needs 6.0x, it would be the
>same as 2.0x. So:
>
>6.0x 1-3 4-6 10-12

Exactly, if they processor supports it.

>
>I also notice that 1.5x is the same as 3.5x. I guess that's the
>result of the multiplier being controlled by the CPU, like you said,
>not the motherboard.

Yep. And that was why I included it.

>
>Once again, thank you very much flipper. I'll try this out when I get
>my "new cutting-edge" AMD K6-2 500 MHz chip. By the way, I only
>bought it because I needed a new cpu fan, and the cpu with fan was
>about $10 more than a fan alone.

If you're lucky you could try overclocking the processor a bit.
Increasing the FSB being preferred to increasing the multiplier.

The problem with increasing the FSB is increasing the PCI bus, which
overclocks the IDE controller, which overclocks the hard drive. And if
the hard drive is overclocked too much then it will corrupt data. That
was a problem with more than about 12% when using UDMA33 hard drives
on UDMA controllers but it generally isn't a problem with UDMA66, or
higher, drives on UDMA33 controllers because it's already able to run
faster than the controller to begin with, so 'overclocking' doesn't
exceed the drive's capability.

When overclocking the processor it's often necessary to increase the
core voltage a bit and up to 2.4 is ok for a 2.2 volt K6-2, if the
heatsink is a good one (heat generation increases with the square of
the voltage)

>
>gene
>
February 12, 2005 1:13:05 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hello Flipper,

On Fri, 11 Feb 2005 18:15:41 -0600, flipper <flipper@fish.net> wrote:

>It wouldn't improve the graphics to take note of. Where it would make
>a modicum of different is in memory speed, FSB, and the PCI bus. FSB
>is particularly important with processors that use motherboard cache
>(an in all Pentium classic, MMX, K6, and K6-2 processors) because
>that limits the cache speed to the processor.
>
>If you're lucky you could try overclocking the processor a bit.
>Increasing the FSB being preferred to increasing the multiplier.
>
>The problem with increasing the FSB is increasing the PCI bus, which
>overclocks the IDE controller, which overclocks the hard drive. And if
>the hard drive is overclocked too much then it will corrupt data. That
>was a problem with more than about 12% when using UDMA33 hard drives
>on UDMA controllers but it generally isn't a problem with UDMA66, or
>higher, drives on UDMA33 controllers because it's already able to run
>faster than the controller to begin with, so 'overclocking' doesn't
>exceed the drive's capability.
>
>When overclocking the processor it's often necessary to increase the
>core voltage a bit and up to 2.4 is ok for a 2.2 volt K6-2, if the
>heatsink is a good one (heat generation increases with the square of
>the voltage)

Ah, once again, thanks for the tips. I'm learning a lot here. Ok,
the machine has an 80 gig Maxtor with UDMA133 as far as I can tell
from googling around.

So, yes, it does sound like I could bump up the FSB and the PCI.
Would I be risking the graphics card to move from a manufacturer
recommended 66 to say 83? That's about a 30% boost, I guess. Does it
matter that the computer is not generally hooked up to a monitor, and
is only checked on a couple times a day through radmin, which allows
me to view the desktop from my main computer? The graphics card isn't
called upon to do any gaming or videos or anything, just looking at
windows explorer mainly, or 2d applications. 99% of the time it will
be just generating the windows xp desktop.

I guess I should read some overclocking boards about this particular
graphics card, a matrox millenium G400 16MB.

My two options for overclocking are:

FSB AGP PCI
124 83 41
112 75 37

Sounds like the PCI would be no problem, but I would be moderately
concerned about over-doing the AGP and burning out the video card.

It would be nice to squeeze some extra performance out of this old dog
though, if only to see what kind of difference that would make.

Thanks for all the teaching flipper, you've been very generous.

gene
February 12, 2005 3:09:42 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Fri, 11 Feb 2005 22:13:05 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hello Flipper,
>
>On Fri, 11 Feb 2005 18:15:41 -0600, flipper <flipper@fish.net> wrote:
>
>>It wouldn't improve the graphics to take note of. Where it would make
>>a modicum of different is in memory speed, FSB, and the PCI bus. FSB
>>is particularly important with processors that use motherboard cache
>>(an in all Pentium classic, MMX, K6, and K6-2 processors) because
>>that limits the cache speed to the processor.
>>
>>If you're lucky you could try overclocking the processor a bit.
>>Increasing the FSB being preferred to increasing the multiplier.
>>
>>The problem with increasing the FSB is increasing the PCI bus, which
>>overclocks the IDE controller, which overclocks the hard drive. And if
>>the hard drive is overclocked too much then it will corrupt data. That
>>was a problem with more than about 12% when using UDMA33 hard drives
>>on UDMA controllers but it generally isn't a problem with UDMA66, or
>>higher, drives on UDMA33 controllers because it's already able to run
>>faster than the controller to begin with, so 'overclocking' doesn't
>>exceed the drive's capability.
>>
>>When overclocking the processor it's often necessary to increase the
>>core voltage a bit and up to 2.4 is ok for a 2.2 volt K6-2, if the
>>heatsink is a good one (heat generation increases with the square of
>>the voltage)
>
>Ah, once again, thanks for the tips. I'm learning a lot here. Ok,
>the machine has an 80 gig Maxtor with UDMA133 as far as I can tell
>from googling around.
>
>So, yes, it does sound like I could bump up the FSB and the PCI.
>Would I be risking the graphics card to move from a manufacturer
>recommended 66 to say 83? That's about a 30% boost, I guess. Does it
>matter that the computer is not generally hooked up to a monitor, and
>is only checked on a couple times a day through radmin, which allows
>me to view the desktop from my main computer?

No.

> The graphics card isn't
>called upon to do any gaming or videos or anything, just looking at
>windows explorer mainly, or 2d applications. 99% of the time it will
>be just generating the windows xp desktop.

Matrox are excellent 2d cards and that one is fine for the processor
you've got.

>
>I guess I should read some overclocking boards about this particular
>graphics card, a matrox millenium G400 16MB.
>
>My two options for overclocking are:
>
>FSB AGP PCI
>124 83 41
>112 75 37
>
>Sounds like the PCI would be no problem, but I would be moderately
>concerned about over-doing the AGP and burning out the video card.

Well, 124/83/41 isn't going to 'burn up' anything but it's high enough
an overclock that some things might not work. For one, if your memory
is PC100 then 24% over is pushing it. although some PC100 will make
it.

Some sound cards, including on-board sound, have problems with an
overlcock that high.

My guess is that the Matrox will work ok at 83Mhz but if it doesn't
then reduce the overlcock.

>
>It would be nice to squeeze some extra performance out of this old dog
>though, if only to see what kind of difference that would make.

It's a bit late since you've bought a K6-2 but a better processor
would be a K6-3 because it has 512K on-die cache and is easily 33%
faster than a K6-2 at the same clock speed. But they're not easy to
find, cost considerably more, and max out at 450 MHz (but that's
better than even if you got your K6-2 to 600MHz)..

The K6-2+ is the fastest, even though with half the size 256K cache,
because it uses the smaller .18 micron process. Which means it comes
in up to 550 Mhz versions and often overclocks to 600MHz, maybe 650 if
you're extra lucky. They expect a 2.0 volt Vcore but, hey, you'd
overclock it anyway so 2.2 would be fine. Some motherboards, however,
have problems booting them because it's really a mobile chip. Whether
it would work in yours I do not know.

There's a K6-3+ too (the 'ultimate'), but those are even rarer than
regular K6-3s.

These people 'specialize' in procesors for the older socket 7 and
super socket 7 systems.

http://www.upgradeability.com/K6plus/index.htm?source=A...

>
>Thanks for all the teaching flipper, you've been very generous.
>
>gene
February 12, 2005 1:12:58 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hi flipper.

Ok, I think I understand most of the things you're saying now.


>Matrox are excellent 2d cards and that one is fine for the processor
>you've got.

I've done a little research on the video card and overclocking. Some
people were running the card a little higher than 83 with no extra
cooling. Besides, those guys were using it for gaming which must be a
lot more taxing on the card than staring at a desktop all day.

I also learned that AGP is expressed as a multiple of FSB, in this
case 2:3. I guess I can check the BIOS to see if I can change this
ratio, but it doesn't look like that's possible with this mainboard.


>
>Well, 124/83/41 isn't going to 'burn up' anything but it's high enough
>an overclock that some things might not work. For one, if your memory
>is PC100 then 24% over is pushing it. although some PC100 will make
>it.

The memory is PC133 if I recall correctly, so that should be all
right. I'll pull the sound card out, along with my old fax modem.
Don't need those in there at all.

I hadn't really considered the network card. I need that in there for
communication with the router. I'll look into that.

Ok, I just read some posts on google groups and it seems some people
can overclock this NIC without problems. I think I'll try it with the
PCI at 41 for awhile and see if everything runs stable. If it
doesn't, I'll drop it down to a 37. Shouldn't be any problems there.

>It's a bit late since you've bought a K6-2 but a better processor
>would be a K6-3 because it has 512K on-die cache and is easily 33%
>faster than a K6-2 at the same clock speed. But they're not easy to
>find, cost considerably more, and max out at 450 MHz (but that's
>better than even if you got your K6-2 to 600MHz)..
>
>The K6-2+ is the fastest, even though with half the size 256K cache,
>because it uses the smaller .18 micron process. Which means it comes
>in up to 550 Mhz versions and often overclocks to 600MHz, maybe 650 if
>you're extra lucky. They expect a 2.0 volt Vcore but, hey, you'd
>overclock it anyway so 2.2 would be fine. Some motherboards, however,
>have problems booting them because it's really a mobile chip. Whether
>it would work in yours I do not know.


Yes. Really, this machine wasn't worth upgrading, but the 500 MHz CPU
was just darned cheap, so I figured I would use it to replace my K6-2
350 MHz. Probably won't be much difference, but I'll be overclocking
the PCI now, so that will probably make more of a difference than the
additional CPU speed.

We'll see. It'll be fun trying this stuff out.

Thanks for your help and advice flipper,

gene
February 12, 2005 8:06:11 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Sat, 12 Feb 2005 10:12:58 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hi flipper.
>
>Ok, I think I understand most of the things you're saying now.
>
>
>>Matrox are excellent 2d cards and that one is fine for the processor
>>you've got.
>
>I've done a little research on the video card and overclocking. Some
>people were running the card a little higher than 83 with no extra
>cooling. Besides, those guys were using it for gaming which must be a
>lot more taxing on the card than staring at a desktop all day.
>
>I also learned that AGP is expressed as a multiple of FSB, in this
>case 2:3. I guess I can check the BIOS to see if I can change this
>ratio, but it doesn't look like that's possible with this mainboard.

That's exactly right. Same with the PCI clock. And same with the CPU
clock: that's what the multiplier is for.

Rather simple concept, actually. You need different clock speeds for
the various things and the easiest way to get them is to
divide/multiply them from a common clock: by an appropriate ratio.


>>
>>Well, 124/83/41 isn't going to 'burn up' anything but it's high enough
>>an overclock that some things might not work. For one, if your memory
>>is PC100 then 24% over is pushing it. although some PC100 will make
>>it.
>
>The memory is PC133 if I recall correctly, so that should be all
>right. I'll pull the sound card out, along with my old fax modem.
>Don't need those in there at all.

If there's no need and you want to pull them for that reason then fine
but it doesn't hurt to try them.


>I hadn't really considered the network card. I need that in there for
>communication with the router. I'll look into that.

Mine have all worked on 44MHz PCI busses but, again, just try it. I
mentioned it so you don't panic if there's a problem, or spend hours
trying to 'troubleshoot' why it isn't working.


>Ok, I just read some posts on google groups and it seems some people
>can overclock this NIC without problems. I think I'll try it with the
>PCI at 41 for awhile and see if everything runs stable. If it
>doesn't, I'll drop it down to a 37. Shouldn't be any problems there.

right.

>
>>It's a bit late since you've bought a K6-2 but a better processor
>>would be a K6-3 because it has 512K on-die cache and is easily 33%
>>faster than a K6-2 at the same clock speed. But they're not easy to
>>find, cost considerably more, and max out at 450 MHz (but that's
>>better than even if you got your K6-2 to 600MHz)..
>>
>>The K6-2+ is the fastest, even though with half the size 256K cache,
>>because it uses the smaller .18 micron process. Which means it comes
>>in up to 550 Mhz versions and often overclocks to 600MHz, maybe 650 if
>>you're extra lucky. They expect a 2.0 volt Vcore but, hey, you'd
>>overclock it anyway so 2.2 would be fine. Some motherboards, however,
>>have problems booting them because it's really a mobile chip. Whether
>>it would work in yours I do not know.
>
>
>Yes. Really, this machine wasn't worth upgrading, but the 500 MHz CPU
>was just darned cheap, so I figured I would use it to replace my K6-2
>350 MHz.

I completely understand.

> Probably won't be much difference, but I'll be overclocking
>the PCI now, so that will probably make more of a difference than the
>additional CPU speed.

The 500 will probably run 550 and 500/550 vs 350 is a decent upgrade.


>We'll see. It'll be fun trying this stuff out.

Yep. It's fun.

I'll give you an example that isn't really worth doing but was 'fun'
and interesting at the time. I had an old Pentium 166MMX that was
multiplier limited so you couldn't overclock it on a regular socket 7
motherboard but, for fun and 'just to see', I tried it with a 1.5x
multiplier on a super socket 7 with 100MHz FSB instead of the
specified 66 MHz FSB. The performance increase was quite dramatic,
even though it was 150 MHz instead of 166.

Of course, there's no good reason to waste a super socket 7
motherboard with a Pentium MMX running 150MHz but it highlighted the
effect of increasing the FSB to performance in those older type
processors where L2 cache is on the motherboard (as is also the case
with your K6-2)

That is, of course, why super socket 7 increased the FSB from 66 to
100 MHz to begin with. Basically, it doesn't make much difference how
fast the processor *could* process instructions if the instructions
can't get to it that fast. And they can only get to it as fast as the
cache can send them and that is limited by the FSB speed.

As for the PCI bus, that should improve your hard drive speed because
your IDE controller is slow, by today's standards, and overclocking
will improve that.

>
>Thanks for your help and advice flipper,
>
>gene
February 14, 2005 12:30:26 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hi again Flipper,

Thanks for the comments regarding the NICs running at 41. The info
about people overclocking NICs was pretty sparse, so your first-hand
experience is useful.

I also found your experiment with the 166MX chip interesting. Neat.
I never really considered how the bus speed would affect things. It
was always just a matter of "well, let's see how I can get my
bus*multiplier to = my cpu speed. To learn that underclocking a
processor at a higher FSB was really illuminating.

I'll probably get my new processor in a couple of days and will change
all the settings then. I guess maybe I should run some tests before
and after my tweaks to see how performance is affected.

RE: The L2 cache; yes, I've read that's a pretty major sticking point
of the K6-2 vs the K6-2+ processors and above. Before I bought my
current computer, I considered trying to upgrade the old machine by
putting in a K6-2+ 550 or something like that, but it turned out the
resale prices on those chips was pretty high, and I wasn't entirely
sure if it would work with this board.

I did a fair amount of research before buying my current machine,
which has a AMD Barton 2500+. Even though I'm running it at spec, I
liked the idea of getting it because of its promise of overclocking
potential. I bought the faster RAM in case I wanted to up the FSB at
some point.

Of course, modern boards have better versatility when it comes to
overclocking... at least mine does. However, I've owned this 2500+
machine for quite awhile now, so I can't remember all the stuff I
learned before buying it.

Back to the old machine, I've set the graphics at XP's lowest possible
settings to keep the graphics card working nice and cool. That should
ensure that the AGP overclock is no problem. Other than that, there's
little that can go wrong... at least until I actually try it! All
that can go wrong will, I suppose.

I'm a benchmarking novice. I'll have to check out some programs for
testing hard drive transfer speed etc. I see the program Sandra gets
lots of mention. I think I'll check out Tom's Hardware to see what
they like to use. I perused that site extensively before investing in
my 2500+ machine.

I'll see what kind of improvements I can make and post a follow up.
Thanks for the great exchange.

gene
February 14, 2005 4:38:34 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Sun, 13 Feb 2005 21:30:26 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hi again Flipper,
>
>Thanks for the comments regarding the NICs running at 41. The info
>about people overclocking NICs was pretty sparse, so your first-hand
>experience is useful.

That was just my card, of course. YMMV


>I also found your experiment with the 166MX chip interesting. Neat.
>I never really considered how the bus speed would affect things. It
>was always just a matter of "well, let's see how I can get my
>bus*multiplier to = my cpu speed. To learn that underclocking a
>processor at a higher FSB was really illuminating.

It sure surprised me. Of course, the improvement with 10% to 20% FSB
increase won't match a 50% increase.

>
>I'll probably get my new processor in a couple of days and will change
>all the settings then. I guess maybe I should run some tests before
>and after my tweaks to see how performance is affected.
>
>RE: The L2 cache; yes, I've read that's a pretty major sticking point
>of the K6-2 vs the K6-2+ processors and above. Before I bought my
>current computer, I considered trying to upgrade the old machine by
>putting in a K6-2+ 550 or something like that, but it turned out the
>resale prices on those chips was pretty high, and I wasn't entirely
>sure if it would work with this board.
>
>I did a fair amount of research before buying my current machine,
>which has a AMD Barton 2500+. Even though I'm running it at spec, I
>liked the idea of getting it because of its promise of overclocking
>potential. I bought the faster RAM in case I wanted to up the FSB at
>some point.

Ah hah. Well, my 'fast' machine is a mobile 2400+ Barton overclocked
to 2.2 gig at 200 FSB (400 DDR).

It will go higher on a lower FSB (multiplier choice problem), but I
went for the higher FSB.

You should have no problem at all overclocking the 2500+.


>Of course, modern boards have better versatility when it comes to
>overclocking... at least mine does. However, I've owned this 2500+
>machine for quite awhile now, so I can't remember all the stuff I
>learned before buying it.

Right. The biggest difference being the limited number of PCI and AGP
multiplier choices.

>
>Back to the old machine, I've set the graphics at XP's lowest possible
>settings to keep the graphics card working nice and cool. That should
>ensure that the AGP overclock is no problem. Other than that, there's
>little that can go wrong... at least until I actually try it! All
>that can go wrong will, I suppose.

The AGP speed is not going to affect the heat output to speak of.

Not of much interest to you since you don't plan on exercising the
graphics but there's an overclock program for that card to increase
the memory and GPU clock speed.


>I'm a benchmarking novice. I'll have to check out some programs for
>testing hard drive transfer speed etc. I see the program Sandra gets
>lots of mention.

Sandra is popular.

Also check the hard drive with the cache turned off so you get the raw
reading.

Since you're planning to run some benchmarks I might as well warn you
of what is usually a 'big surprise'. The performance of processors
using motherboard cache doesn't scale directly with speed. I.E.
doubling speed doesn't double performance because the *cache* speed
remains the same. It is usually around half to 1: doubling speed will
give you a, roughly, 50% performance boost.

Processors with on-die cache scale roughly 1 to 1, within reason.

> I think I'll check out Tom's Hardware to see what
>they like to use. I perused that site extensively before investing in
>my 2500+ machine.
>
>I'll see what kind of improvements I can make and post a follow up.
>Thanks for the great exchange.
>
>gene
February 15, 2005 10:59:14 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hello Flipper (and anyone else watching this thread),

Ok, I've got my 500 MHz K6-2 cpu.

I did some benchmarks before the upgrade, with FSB=100, PCI=33, and
AGP=66.

Then I overclocked to PCI=124, PCI=41, AGP=83. The multiplier was set
to 3 for a slight overclock of my 350MHz processor to 372 Mhz.

I did see some major improvements in the memory bandwidth benchmark
that Sandra does (about a 25% improvement). That was pretty amazing.
I think you aluded to this flipper, due to the lack of onboard L2
cache on the K6-2?

There was about a 17% improvement in the File System benchmark from 14
to 16 MB/s. Probably a rounding error in there, as it just uses
integers.

I found the system was unstable at this overclock for some reason. I
restored it to the way I had it. Could be as simple as needing to up
the voltage on the cpu for the modest 350 to 372 MHz. However, the
video card had some kind of hiccup at the AGP speed of 83. It wanted
a reboot, and was showing some color distortion on the desktop. It
had happened a couple times before with no overclock, so it could be a
coincidence, or a function of the cpu sputtering. I suspect the cpu
because another application crashed where I've had no problem before.

Then, I got my new cpu and installed it, again with the FSB et al
overclocked to 124/83/41. The CPU is running at 4x multiplier = 496
MHz. Ran the benchmarks again. The CPU-dependent tests... Drhystone
and Whetstone turned up about 40% better than the starting tests.
The multimedia tests came up 42% better, which is largely due to the
faster CPU clock speed.

Interestingly, there wasn't much improvement on the memory bandwidth
that could be attributed to the faster CPU. Most gains came from
overclocking the PCI Bus and the FSB speeds.

The time from power on to the windows desktop improved from 1:14
originally to 0:58.

Copying a 961 MB file from my other computer over the network improved
from 4:34 to 3:59; most of this improvement was from the faster bus
speeds again.

Copying the file on the same computer improved from 3:23 to 2:45,
again due to faster bus speeds.

Converting 10% of that 961MB wav to mp3 using cdex improved from a
time of 3:17 to 2:17. This was a mix of faster bus speeds and faster
overall cpu speed. Slightly more of this improvement was from faster
CPU.

Copying the file from the computer to the other computer over the
network (reverse trip) went from 3:34 to 3:15. Almost the entire
improvement was from faster bus speed.

I haven't overclocked the CPU because I don't have any thermal paste
on the heat sink, just the wax pad. Actually, the old heatsink didn't
even have a wax pad on it, so I guess this heatsink is a little
better.

I wondering if I should try overclocking the CPU, or leave it at 496.
I think I'll leave it there for several days to make sure it's running
stable at these elevated bus speeds. Then, I may just get some
thermal paste and knock up the cpu speed a notch... to 558MHz.

I'm just wondering if I need the paste and if I need to turn up the
voltage to 2.3 from 2.2 spec. If I decide to bump it up, I can look
into it a little further.

So... those are the results for what it's worth. Looks like I left a
lot of performance on the table for several years by leaving
everything at low bus speeds. That would have helped awhile ago when
I was approaching the limits of usefulness for day-to-day tasks.

Thanks again for the overclocking tips flipper. Things are running a
lot better.

gene
February 16, 2005 4:10:46 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Tue, 15 Feb 2005 19:59:14 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hello Flipper (and anyone else watching this thread),
>
>Ok, I've got my 500 MHz K6-2 cpu.
>
>I did some benchmarks before the upgrade, with FSB=100, PCI=33, and
>AGP=66.
>
>Then I overclocked to PCI=124, PCI=41, AGP=83. The multiplier was set
>to 3 for a slight overclock of my 350MHz processor to 372 Mhz.
>
>I did see some major improvements in the memory bandwidth benchmark
>that Sandra does (about a 25% improvement). That was pretty amazing.
>I think you aluded to this flipper, due to the lack of onboard L2
>cache on the K6-2?
>
>There was about a 17% improvement in the File System benchmark from 14
>to 16 MB/s. Probably a rounding error in there, as it just uses
>integers.
>
>I found the system was unstable at this overclock for some reason. I
>restored it to the way I had it. Could be as simple as needing to up
>the voltage on the cpu for the modest 350 to 372 MHz. However, the
>video card had some kind of hiccup at the AGP speed of 83. It wanted
>a reboot, and was showing some color distortion on the desktop. It
>had happened a couple times before with no overclock, so it could be a
>coincidence, or a function of the cpu sputtering. I suspect the cpu
>because another application crashed where I've had no problem before.
>
>Then, I got my new cpu and installed it, again with the FSB et al
>overclocked to 124/83/41. The CPU is running at 4x multiplier = 496
>MHz. Ran the benchmarks again. The CPU-dependent tests... Drhystone
>and Whetstone turned up about 40% better than the starting tests.
>The multimedia tests came up 42% better, which is largely due to the
>faster CPU clock speed.
>
>Interestingly, there wasn't much improvement on the memory bandwidth
>that could be attributed to the faster CPU. Most gains came from
>overclocking the PCI Bus and the FSB speeds.
>
>The time from power on to the windows desktop improved from 1:14
>originally to 0:58.
>
>Copying a 961 MB file from my other computer over the network improved
>from 4:34 to 3:59; most of this improvement was from the faster bus
>speeds again.
>
>Copying the file on the same computer improved from 3:23 to 2:45,
>again due to faster bus speeds.
>
>Converting 10% of that 961MB wav to mp3 using cdex improved from a
>time of 3:17 to 2:17. This was a mix of faster bus speeds and faster
>overall cpu speed. Slightly more of this improvement was from faster
>CPU.
>
>Copying the file from the computer to the other computer over the
>network (reverse trip) went from 3:34 to 3:15. Almost the entire
>improvement was from faster bus speed.
>
>I haven't overclocked the CPU because I don't have any thermal paste
>on the heat sink, just the wax pad. Actually, the old heatsink didn't
>even have a wax pad on it, so I guess this heatsink is a little
>better.
>
>I wondering if I should try overclocking the CPU, or leave it at 496.
>I think I'll leave it there for several days to make sure it's running
>stable at these elevated bus speeds. Then, I may just get some
>thermal paste and knock up the cpu speed a notch... to 558MHz.
>
>I'm just wondering if I need the paste and if I need to turn up the
>voltage to 2.3 from 2.2 spec. If I decide to bump it up, I can look
>into it a little further.
>
>So... those are the results for what it's worth. Looks like I left a
>lot of performance on the table for several years by leaving
>everything at low bus speeds. That would have helped awhile ago when
>I was approaching the limits of usefulness for day-to-day tasks.
>
>Thanks again for the overclocking tips flipper. Things are running a
>lot better.

Your analysis is about right. I'd point out that memory bandwidth is
basically the FSB, as you deduced, which makes sense because that's
the bottleneck to memory and the memory runs at that speed as well.

The problems with the 350 overclock were probably due to the processor
as not all K6-2's are created equal and 350 was the 'top end' for the
earlier versions. It might have settled out better with more core
voltage but that's kind of moot now with the 500 on hand.

The lack of a thermal pad didn't help either. Heat is the enemy of
faster speeds so you sometimes need better cooling to get the
overclock.

Good tests though and you got a good look at how the different things
are affected.

>gene
February 16, 2005 4:03:24 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Wed, 16 Feb 2005 01:10:46 -0600, flipper <flipper@fish.net> wrote:

Hi flipper,

>Your analysis is about right. I'd point out that memory bandwidth is
>basically the FSB, as you deduced, which makes sense because that's
>the bottleneck to memory and the memory runs at that speed as well.
>
>The problems with the 350 overclock were probably due to the processor
>as not all K6-2's are created equal and 350 was the 'top end' for the
>earlier versions. It might have settled out better with more core
>voltage but that's kind of moot now with the 500 on hand.

I see. 350MHz was a stretch for the processor. You also mention the
heat pad, which makes sense. One other thing might be the fact that
the old CPU was routinely running at maximum capacity. The little
graph that task manager gives was showing CPU usage frequently at
100%. Radmin seems to take a lot of the processor's capacity.

The new processor is usually running below 70%. If I had thermal
paste, I'd use it for sure. I don't have any, and I'm wondering where
I can find it for next to nothing. Maybe I'll go dumpster diving
behind the computer store. :-)

Not sure overclocking the cpu would really yield much benefit for what
I'm using it for anyway. Seems most of the benefit has been achieved
already by the FSB/PCI/AGP overclock.

The system has been running about 20 hours now, and it's stable so
far. I'm not stressing it, just running it the way I normally would.
Seems like a legitimate way to test it.

Anyway, thanks for all the help.

gene
February 16, 2005 9:41:59 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Wed, 16 Feb 2005 13:03:24 -0600, gene <noemailplease@nowhere.com>
wrote:

>On Wed, 16 Feb 2005 01:10:46 -0600, flipper <flipper@fish.net> wrote:
>
>Hi flipper,
>
>>Your analysis is about right. I'd point out that memory bandwidth is
>>basically the FSB, as you deduced, which makes sense because that's
>>the bottleneck to memory and the memory runs at that speed as well.
>>
>>The problems with the 350 overclock were probably due to the processor
>>as not all K6-2's are created equal and 350 was the 'top end' for the
>>earlier versions. It might have settled out better with more core
>>voltage but that's kind of moot now with the 500 on hand.
>
>I see. 350MHz was a stretch for the processor.

Well, it depends on when the processor was made. It's been a while so
I don't remember off the top of my head but there was a code you could
look at to see if it was the newer version. Was all the overclocking
rage at the time.. "look for the gold letters 'whatever', etc." (I had
one but un judiciously burned it up during an experimental overclock
from 350 to 550 with either 2.8 or 2.9 Vcore. But it was stable right
up to when it died. hehe)

Most K6-2s weren't all that great for overclocking, though, and
getting 450 from a 400 was 'good'.

> You also mention the
>heat pad, which makes sense. One other thing might be the fact that
>the old CPU was routinely running at maximum capacity. The little
>graph that task manager gives was showing CPU usage frequently at
>100%. Radmin seems to take a lot of the processor's capacity.

Seems odd.

An overclock should work at full load anyway or else, what's the
point of 'more speed' if you can't use the speed?

>
>The new processor is usually running below 70%. If I had thermal
>paste, I'd use it for sure. I don't have any, and I'm wondering where
>I can find it for next to nothing. Maybe I'll go dumpster diving
>behind the computer store. :-)

Radio Shack white thermal compound.

One of the mistakes people make is figuring the processor is cool if
the heatsink is cool. It might be but it can also be that the heat
simply isn't getting to the heatsink. It'll be 'cool' then too but the
processor isn't.

The heatsink must be warmer than the surrounding air for it to
dissipate heat so it generally feels 'warm' when working well.


>Not sure overclocking the cpu would really yield much benefit for what
>I'm using it for anyway. Seems most of the benefit has been achieved
>already by the FSB/PCI/AGP overclock.

Normally I'd agree, unless the person is doing something CPU intensive
like video editing and then it translates to faster process times. But
I'm not quite sure what to think with the loads you say you're getting
from Radmin.

>
>The system has been running about 20 hours now, and it's stable so
>far. I'm not stressing it, just running it the way I normally would.
>Seems like a legitimate way to test it.

That's a reasonable first test since if it won't do 'normal' then it's
obviously in trouble. But you should also stress test it

>
>Anyway, thanks for all the help.
>
>gene
February 17, 2005 12:37:53 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hello flipper,
>
>Well, it depends on when the processor was made. It's been a while so
>I don't remember off the top of my head but there was a code you could
>look at to see if it was the newer version. Was all the overclocking
>rage at the time.. "look for the gold letters 'whatever', etc." (I had
>one but un judiciously burned it up during an experimental overclock
>from 350 to 550 with either 2.8 or 2.9 Vcore. But it was stable right
>up to when it died. hehe)

Well... you sure were getting good bang for your buck right until the
meltdown. :-) Makes me think of all of those sites that show photos
of people's nifty cooling systems. I guess if I were an expert metal
worker. I'd set up some kind of liquid cooling system just for fun.

One of the more interesting things I saw was a guy who immersed his
entire pc in oil (mineral oil?)

I was thinking that an old drinking water cooler would do a credible
job if one put some kind of pump in the reservoir to pump water to a
heatsink on the cpu.

Buying a commercial cpu water cooler would be less interesting, but
more reliable.

>
>Most K6-2s weren't all that great for overclocking, though, and
>getting 450 from a 400 was 'good'.
>
>> You also mention the
>>heat pad, which makes sense. One other thing might be the fact that
>>the old CPU was routinely running at maximum capacity. The little
>>graph that task manager gives was showing CPU usage frequently at
>>100%. Radmin seems to take a lot of the processor's capacity.
>
>Seems odd.
>
>An overclock should work at full load anyway or else, what's the
>point of 'more speed' if you can't use the speed?

Ok, I'm probably off-base on that maximum cpu capacity argument then.
I'd had my heatsink off once some time ago... I must have noticed that
there was no pad on it, but I think that was one of the few heatsinks
I'd actually held in my hand.

Oh wait, now I remember. I replaced the heatsink and the fan. They
came in one package. Startech socket 7 heatsink and fan.

I got my new fan (along with heatsink and this 500 MHz CPU) from ebay.
I did some research on some of the other fans on ebay and some of them
weren't rated for 500 MHz. They must have been for older cpus for the
socket 7/370 format.

Perhaps even the Startech one I purchased about a year ago was
borderline. The lack of a wax pad on the heatsink is suspect.
Actually, the bottom of the heatsink isn't even that smooth. it's got
little striations on it. Heat is probably the main stumbling blocks I
had for overclocking the 350MHz. That, along with your comments about
the 350MHz, and the K6-2 in general, not being great for overclocking.



>
>>
>>The new processor is usually running below 70%. If I had thermal
>>paste, I'd use it for sure. I don't have any, and I'm wondering where
>>I can find it for next to nothing. Maybe I'll go dumpster diving
>>behind the computer store. :-)
>
>Radio Shack white thermal compound.

I'll have to look into that. There is a radio shack in town here.
The wax pad on the heatsink is scratched up pretty badly. Not sure
what they did to it... the whole assembly is a "working pull".

>
>One of the mistakes people make is figuring the processor is cool if
>the heatsink is cool. It might be but it can also be that the heat
>simply isn't getting to the heatsink. It'll be 'cool' then too but the
>processor isn't.
>
>The heatsink must be warmer than the surrounding air for it to
>dissipate heat so it generally feels 'warm' when working well.

That's a good tip. Thanks for that.


>
>
>>Not sure overclocking the cpu would really yield much benefit for what
>>I'm using it for anyway. Seems most of the benefit has been achieved
>>already by the FSB/PCI/AGP overclock.
>
>Normally I'd agree, unless the person is doing something CPU intensive
>like video editing and then it translates to faster process times. But
>I'm not quite sure what to think with the loads you say you're getting
>from Radmin.

Radmin does seem like a real cpu hog for some reason. There's a
process called r_server.exe that routinely takes 35% of the 500MHz cpu
while I'm accessing the computer remotely. That can peak up to mid
40s. And this is *with* the FSB/PCI/AGP overclock of 124/41/83. I
can't recall the cpu usage with the 350MHz, but it must have been
higher. The funny thing is that I would often just leave the machine
connected to radmin, not thinking that it was taking up so much system
resources. Now I just use it to peek in once in awhile and promptly
switch it off.

The other process I'm routinely running is the btdownloadgui, which
may take about 20% of the cpu capacity on average.

>
>>
>>The system has been running about 20 hours now, and it's stable so
>>far. I'm not stressing it, just running it the way I normally would.
>>Seems like a legitimate way to test it.
>
>That's a reasonable first test since if it won't do 'normal' then it's
>obviously in trouble. But you should also stress test it

Is this what's referred to as burn-in testing? Any particular
recommendations on a program to try this with?

gene
February 18, 2005 2:00:03 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Thu, 17 Feb 2005 09:37:53 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hello flipper,
>>
>>Well, it depends on when the processor was made. It's been a while so
>>I don't remember off the top of my head but there was a code you could
>>look at to see if it was the newer version. Was all the overclocking
>>rage at the time.. "look for the gold letters 'whatever', etc." (I had
>>one but un judiciously burned it up during an experimental overclock
>>from 350 to 550 with either 2.8 or 2.9 Vcore. But it was stable right
>>up to when it died. hehe)
>
>Well... you sure were getting good bang for your buck right until the
>meltdown. :-) Makes me think of all of those sites that show photos
>of people's nifty cooling systems. I guess if I were an expert metal
>worker. I'd set up some kind of liquid cooling system just for fun.

Water cooling has it's own problems, not the least of which is cost.


>
>One of the more interesting things I saw was a guy who immersed his
>entire pc in oil (mineral oil?)

Might have been mineral oil but there is one, the name of which I
don't remember off the top of my head. specifically for that sort of
thing.

>
>I was thinking that an old drinking water cooler would do a credible
>job if one put some kind of pump in the reservoir to pump water to a
>heatsink on the cpu.

Possibly but 'coolers' don't handle as much heat flow as people seem
to think.

>
>Buying a commercial cpu water cooler would be less interesting, but
>more reliable.
>
>>
>>Most K6-2s weren't all that great for overclocking, though, and
>>getting 450 from a 400 was 'good'.
>>
>>> You also mention the
>>>heat pad, which makes sense. One other thing might be the fact that
>>>the old CPU was routinely running at maximum capacity. The little
>>>graph that task manager gives was showing CPU usage frequently at
>>>100%. Radmin seems to take a lot of the processor's capacity.
>>
>>Seems odd.
>>
>>An overclock should work at full load anyway or else, what's the
>>point of 'more speed' if you can't use the speed?
>
>Ok, I'm probably off-base on that maximum cpu capacity argument then.
>I'd had my heatsink off once some time ago... I must have noticed that
>there was no pad on it, but I think that was one of the few heatsinks
>I'd actually held in my hand.
>
>Oh wait, now I remember. I replaced the heatsink and the fan. They
>came in one package. Startech socket 7 heatsink and fan.
>
>I got my new fan (along with heatsink and this 500 MHz CPU) from ebay.
>I did some research on some of the other fans on ebay and some of them
>weren't rated for 500 MHz. They must have been for older cpus for the
>socket 7/370 format.

Heatsink 'speed ratings' are close to useless. Not entirely, but
close. For example, how 'hot' does the processor get, and in what
ambient temperature, with that 'speed rating'?

Generally it means, or is the best it could mean, that the processor
should not exceed maximum operating temp in the manufacturer's maximum
ambient temp requirement. But just keeping the processor under 65-70C
case temp isn't good for overclocking nor is it particularly great for
it's lifespan.



>Perhaps even the Startech one I purchased about a year ago was
>borderline. The lack of a wax pad on the heatsink is suspect.
>Actually, the bottom of the heatsink isn't even that smooth. it's got
>little striations on it.

Sound like a typical 'generic' extruded heatsink and the 'pad' is
supposed to take care of the poor finish.

K6-2 500 puts out about 30 watts, under load, which isn't a terrible
lot by today's standards but it isn't trivial either.

> Heat is probably the main stumbling blocks I
>had for overclocking the 350MHz. That, along with your comments about
>the 350MHz, and the K6-2 in general, not being great for overclocking.
>

Ah, ah, ah. I remember now. The 'good' one was the "CTX" core K6-2.


>>>
>>>The new processor is usually running below 70%. If I had thermal
>>>paste, I'd use it for sure. I don't have any, and I'm wondering where
>>>I can find it for next to nothing. Maybe I'll go dumpster diving
>>>behind the computer store. :-)
>>
>>Radio Shack white thermal compound.
>
>I'll have to look into that. There is a radio shack in town here.
>The wax pad on the heatsink is scratched up pretty badly. Not sure
>what they did to it... the whole assembly is a "working pull".

Well, 'working pull' pretty much explains it. It was installed and
then removed. The typical 'pad' is a one use thing.



>>One of the mistakes people make is figuring the processor is cool if
>>the heatsink is cool. It might be but it can also be that the heat
>>simply isn't getting to the heatsink. It'll be 'cool' then too but the
>>processor isn't.
>>
>>The heatsink must be warmer than the surrounding air for it to
>>dissipate heat so it generally feels 'warm' when working well.
>
>That's a good tip. Thanks for that.
>
>
>>
>>
>>>Not sure overclocking the cpu would really yield much benefit for what
>>>I'm using it for anyway. Seems most of the benefit has been achieved
>>>already by the FSB/PCI/AGP overclock.
>>
>>Normally I'd agree, unless the person is doing something CPU intensive
>>like video editing and then it translates to faster process times. But
>>I'm not quite sure what to think with the loads you say you're getting
>>from Radmin.
>
>Radmin does seem like a real cpu hog for some reason. There's a
>process called r_server.exe that routinely takes 35% of the 500MHz cpu
>while I'm accessing the computer remotely. That can peak up to mid
>40s. And this is *with* the FSB/PCI/AGP overclock of 124/41/83. I
>can't recall the cpu usage with the 350MHz, but it must have been
>higher. The funny thing is that I would often just leave the machine
>connected to radmin, not thinking that it was taking up so much system
>resources. Now I just use it to peek in once in awhile and promptly
>switch it off.
>
>The other process I'm routinely running is the btdownloadgui, which
>may take about 20% of the cpu capacity on average.
>
>>
>>>
>>>The system has been running about 20 hours now, and it's stable so
>>>far. I'm not stressing it, just running it the way I normally would.
>>>Seems like a legitimate way to test it.
>>
>>That's a reasonable first test since if it won't do 'normal' then it's
>>obviously in trouble. But you should also stress test it
>
>Is this what's referred to as burn-in testing? Any particular
>recommendations on a program to try this with?

It's common for people to call it "burn in" but that's an incorrect
usage of a standard industry term. "Burn in" is the process of
stressing devices (often at elevated temperature, hence the name) to
induce infant mortality failures rather that let the weak devices get
into the field where they would reduce reliability and increase
warranty costs. Of course, the ideal case would be that none of them
fail but, the point is, you're looking for device failures, not that
they 'work right'.

What you're really doing is qualification testing: testing that it
performs properly under 'specified' operating conditions. E.g that it
operates properly at maximum load in the expected environment.

Technically you should do that at the temperature extremes you expect
it to operate in or else you might find it works fine now but has
problems if the room is a bit warmer than usual, and it's not unusual
to hear casual overclockers start complaining about 'problems' when
summer rolls around.

*What* you use to do that testing is a matter of judgment and how
much assurance you want that it'll operate properly.

Prime95 is a common test as it heavily exercises the CPU. CPUBurn is
another one in that category. 3dmark is often used because that
exercises the AGP port, display card, and memory, along with the CPU
(although not as heavily as CPUburn or Prime95). Memtest is another
one, used to test, tada, the memory. Another popular test is to
zip/unzip some very large files as that tests the CPU and,
specifically, the hard drive (and PCI bus) since the file has to be
read and written to the hard drive.

Benchmarking programs are useful to see if various error correction
mechanisms are simply covering up 'problems', The benchmark will come
in unusually low as the error recovery mechanism slows things down.
I've seen, for example, some overclocked processors benchmark
surprisingly low because they were getting cache errors. The cache ECC
caught them but that meant the entire cache was being routinely
flushed and refilled, slowing the processor down.

Benchmark programs also stress things but the problem there is they do
not run long enough for a proper stress test. Especially for CPU
testing as it can take an hour, or more, for temperatures to stabilize
throughout the system.

IMO, if it passes CPUburn (and runs 'normal' things properly) then
odds are fairly good it's stable, although I had one system that
passed CPUburn but failed to run 3dmark all the way through. However,
that was a particularly unusual system in that I was overclocking an
Intel Tualatin Celeron on a motherboard that is not supposed to
support tualatins at all, but was modified to do so, and it turned out
that the modification was insufficient and needed a slight change.

Of course, if you were overclocking the display card then you'd
certainly want to include 3dmark as well.

IMO, memtest is most useful when you suspect a memory problem, but not
really as a 'stress' test.

So, in addition to verifying that everything seems to work 'normally',
I'd run CPUburn for an hour, or so, monitoring temps, run 3dmark, and
then Zip/Unzip (or Rar, or Ace, etc.) a few huge files. That isn't a
guarantee nothing else can go wrong but it covers the most likely
candidates.


>
>gene
February 18, 2005 1:19:42 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hiya flipper,

On Thu, 17 Feb 2005 23:00:03 -0600, flipper <flipper@fish.net> wrote:

>Sound like a typical 'generic' extruded heatsink and the 'pad' is
>supposed to take care of the poor finish.

Actually that one didn't even come with a pad. I didn't closely
examine the finish on the heatsink I just put in, but the pad has
surface scratches. Like you say, one use only. I guess I should
replace that pad... but I probably won't if the cpu doesn't give me
problems.

>
>Ah, ah, ah. I remember now. The 'good' one was the "CTX" core K6-2.

My 350 was an "AFR". Manufactured 16th week, 1999.


>>Is this what's referred to as burn-in testing? Any particular
>>recommendations on a program to try this with?
>
>It's common for people to call it "burn in" but that's an incorrect
>usage of a standard industry term. "Burn in" is the process of
>stressing devices (often at elevated temperature, hence the name) to
>induce infant mortality failures rather that let the weak devices get
>into the field where they would reduce reliability and increase
>warranty costs. Of course, the ideal case would be that none of them
>fail but, the point is, you're looking for device failures, not that
>they 'work right'.

I see. The manufacturer burns in to ensure quality control on
released cpus. I wonder how much it costs to manufacture a cpu,
disregarding all R&D costs. I showed my wife the 350 MHz cpu I pulled
out of the machine, and she was amazed how small it was. I guess I'd
never thought of it, but yes, for a device with over a million
circuits, it is indeed small.



>
>What you're really doing is qualification testing: testing that it
>performs properly under 'specified' operating conditions. E.g that it
>operates properly at maximum load in the expected environment.
>
>Technically you should do that at the temperature extremes you expect
>it to operate in or else you might find it works fine now but has
>problems if the room is a bit warmer than usual, and it's not unusual
>to hear casual overclockers start complaining about 'problems' when
>summer rolls around.
>
>*What* you use to do that testing is a matter of judgment and how
>much assurance you want that it'll operate properly.
>
>Prime95 is a common test as it heavily exercises the CPU. CPUBurn is
>another one in that category. 3dmark is often used because that
>exercises the AGP port, display card, and memory, along with the CPU
>(although not as heavily as CPUburn or Prime95). Memtest is another
>one, used to test, tada, the memory. Another popular test is to
>zip/unzip some very large files as that tests the CPU and,
>specifically, the hard drive (and PCI bus) since the file has to be
>read and written to the hard drive.
>
>Benchmarking programs are useful to see if various error correction
>mechanisms are simply covering up 'problems', The benchmark will come
>in unusually low as the error recovery mechanism slows things down.
>I've seen, for example, some overclocked processors benchmark
>surprisingly low because they were getting cache errors. The cache ECC
>caught them but that meant the entire cache was being routinely
>flushed and refilled, slowing the processor down.
>
>Benchmark programs also stress things but the problem there is they do
>not run long enough for a proper stress test. Especially for CPU
>testing as it can take an hour, or more, for temperatures to stabilize
>throughout the system.
>
>IMO, if it passes CPUburn (and runs 'normal' things properly) then
>odds are fairly good it's stable, although I had one system that
>passed CPUburn but failed to run 3dmark all the way through. However,
>that was a particularly unusual system in that I was overclocking an
>Intel Tualatin Celeron on a motherboard that is not supposed to
>support tualatins at all, but was modified to do so, and it turned out
>that the modification was insufficient and needed a slight change.
>
>Of course, if you were overclocking the display card then you'd
>certainly want to include 3dmark as well.
>
>IMO, memtest is most useful when you suspect a memory problem, but not
>really as a 'stress' test.
>
>So, in addition to verifying that everything seems to work 'normally',
>I'd run CPUburn for an hour, or so, monitoring temps, run 3dmark, and
>then Zip/Unzip (or Rar, or Ace, etc.) a few huge files. That isn't a
>guarantee nothing else can go wrong but it covers the most likely
>candidates.

Thanks for taking the time and keystrokes for this lengthy
explanation. I did have the pleasure of using memtest86 for finding a
flawed memory stick in my 2500+ Athlon. I had a real problem with
crashes and my troubleshooting finally diagnosed the bad RAM. It was
frustrating because I would get a crash anytime I was running a virus
scan, so I could never tell for sure if a virus was causing my
problem.

I think I eventually came across a website that suggested testing the
RAM. I guess that's a different topic altogether.

I'll download CPUburn and see how it does. I guess if it passes, I
can up the multiplier. Of course, I'd feel more confident to do that
with a new wax pad and some thermal paste. I think I've used 3dmark
before. Didn't seem too time intensive, so I can run that one too.

You say to monitor the temperatures. Unfortunately, I don't believe
that my motherboard has that feature. Shouldn't be a real problem
though. I can just watch to see that it survives the test. Zipping
and unzipping, like you say, sounds like a good test too. I'll give
that a try.

Thank you,

gene
February 19, 2005 3:52:15 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Fri, 18 Feb 2005 10:19:42 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hiya flipper,
>
>On Thu, 17 Feb 2005 23:00:03 -0600, flipper <flipper@fish.net> wrote:
>
>>Sound like a typical 'generic' extruded heatsink and the 'pad' is
>>supposed to take care of the poor finish.
>
>Actually that one didn't even come with a pad. I didn't closely
>examine the finish on the heatsink I just put in, but the pad has
>surface scratches. Like you say, one use only. I guess I should
>replace that pad... but I probably won't if the cpu doesn't give me
>problems.
>
>>
>>Ah, ah, ah. I remember now. The 'good' one was the "CTX" core K6-2.
>
>My 350 was an "AFR". Manufactured 16th week, 1999.

Can't tell from "AFR." That's the suffix designating package A (they
all are), F Vcore range (2.1 to 2.3, in that case), and R temperature
range (0-70C, in that case).

'CTX" is the core 'code name', so to speak, and isn't written on it.
You can tell if it has, in gold lettering, "26050 " along the edge on
the ceramic chip carrier. It also would have a CPUID=0580h.

The previous core 'code name' was "Chomper," and that isn't written
anywhere either.

Btw, it's with the CTX core that the x2 multiplier became x6. With
Chomper cores x2 is x2.


>>>Is this what's referred to as burn-in testing? Any particular
>>>recommendations on a program to try this with?
>>
>>It's common for people to call it "burn in" but that's an incorrect
>>usage of a standard industry term. "Burn in" is the process of
>>stressing devices (often at elevated temperature, hence the name) to
>>induce infant mortality failures rather that let the weak devices get
>>into the field where they would reduce reliability and increase
>>warranty costs. Of course, the ideal case would be that none of them
>>fail but, the point is, you're looking for device failures, not that
>>they 'work right'.
>
>I see. The manufacturer burns in to ensure quality control on
>released cpus.

Close enough but, as is said in Q.A. circles, you can't 'test in'
quality it's built in. Testing is when you find out if it's there.

> I wonder how much it costs to manufacture a cpu,
>disregarding all R&D costs. I showed my wife the 350 MHz cpu I pulled
>out of the machine, and she was amazed how small it was. I guess I'd
>never thought of it, but yes, for a device with over a million
>circuits, it is indeed small.

A million circuits is peanuts these days. Even a K6-2 has 9.3 million
transistors in it and the corresponding Celeron (PPGA) has 19 million.

Athlon XP, Barton core, is 53 million and a Northwood P4 is 55
million.

For those of us who remember discrete transistors it's staggering.
Imagine, a transistor with a failure rate of one per million years
sounds rather decent but that would mean a K6-2 failure every 5 weeks,
or so!

Needless to say, reliability has improved along with integrated
circuits.


>>What you're really doing is qualification testing: testing that it
>>performs properly under 'specified' operating conditions. E.g that it
>>operates properly at maximum load in the expected environment.
>>
>>Technically you should do that at the temperature extremes you expect
>>it to operate in or else you might find it works fine now but has
>>problems if the room is a bit warmer than usual, and it's not unusual
>>to hear casual overclockers start complaining about 'problems' when
>>summer rolls around.
>>
>>*What* you use to do that testing is a matter of judgment and how
>>much assurance you want that it'll operate properly.
>>
>>Prime95 is a common test as it heavily exercises the CPU. CPUBurn is
>>another one in that category. 3dmark is often used because that
>>exercises the AGP port, display card, and memory, along with the CPU
>>(although not as heavily as CPUburn or Prime95). Memtest is another
>>one, used to test, tada, the memory. Another popular test is to
>>zip/unzip some very large files as that tests the CPU and,
>>specifically, the hard drive (and PCI bus) since the file has to be
>>read and written to the hard drive.
>>
>>Benchmarking programs are useful to see if various error correction
>>mechanisms are simply covering up 'problems', The benchmark will come
>>in unusually low as the error recovery mechanism slows things down.
>>I've seen, for example, some overclocked processors benchmark
>>surprisingly low because they were getting cache errors. The cache ECC
>>caught them but that meant the entire cache was being routinely
>>flushed and refilled, slowing the processor down.
>>
>>Benchmark programs also stress things but the problem there is they do
>>not run long enough for a proper stress test. Especially for CPU
>>testing as it can take an hour, or more, for temperatures to stabilize
>>throughout the system.
>>
>>IMO, if it passes CPUburn (and runs 'normal' things properly) then
>>odds are fairly good it's stable, although I had one system that
>>passed CPUburn but failed to run 3dmark all the way through. However,
>>that was a particularly unusual system in that I was overclocking an
>>Intel Tualatin Celeron on a motherboard that is not supposed to
>>support tualatins at all, but was modified to do so, and it turned out
>>that the modification was insufficient and needed a slight change.
>>
>>Of course, if you were overclocking the display card then you'd
>>certainly want to include 3dmark as well.
>>
>>IMO, memtest is most useful when you suspect a memory problem, but not
>>really as a 'stress' test.
>>
>>So, in addition to verifying that everything seems to work 'normally',
>>I'd run CPUburn for an hour, or so, monitoring temps, run 3dmark, and
>>then Zip/Unzip (or Rar, or Ace, etc.) a few huge files. That isn't a
>>guarantee nothing else can go wrong but it covers the most likely
>>candidates.
>
>Thanks for taking the time and keystrokes for this lengthy
>explanation. I did have the pleasure of using memtest86 for finding a
>flawed memory stick in my 2500+ Athlon. I had a real problem with
>crashes and my troubleshooting finally diagnosed the bad RAM. It was
>frustrating because I would get a crash anytime I was running a virus
>scan, so I could never tell for sure if a virus was causing my
>problem.
>
>I think I eventually came across a website that suggested testing the
>RAM. I guess that's a different topic altogether.
>
>I'll download CPUburn and see how it does. I guess if it passes, I
>can up the multiplier. Of course, I'd feel more confident to do that
>with a new wax pad and some thermal paste. I think I've used 3dmark
>before. Didn't seem too time intensive, so I can run that one too.
>
>You say to monitor the temperatures. Unfortunately, I don't believe
>that my motherboard has that feature. Shouldn't be a real problem
>though. I can just watch to see that it survives the test.

If you watch it closely, yes. But a system can appear 'locked up' and
still be consuming power and a processor can consume power when it's
'crazy' .

That's how my experiment got away from me. I didn't notice it was
locked up, as it was an extended test, and it cooked itself to death
even though, for all appearances, it was 'doing nothing'.

Point is, you can't count on it going back to 'cool' just because the
system 'locks up'. It isn't 'fail safe'.

> Zipping
>and unzipping, like you say, sounds like a good test too. I'll give
>that a try.
>
>Thank you,
>
>gene
February 22, 2005 3:19:23 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hi flipper,

I think I'm about done with my overclocking now. I don't think I'll
bother pushing the multiplier any more than the 496MHz it's running
at. I'm sure I could easily push it to 558MHz, but my cpu cooling is
marginal, so I think I'll leave it at that.

I haven't done any of the extra CPUburn tests, but I guess I should.
Wouldn't be too difficult.

>Can't tell from "AFR." That's the suffix designating package A (they
>all are), F Vcore range (2.1 to 2.3, in that case), and R temperature
>range (0-70C, in that case).
>
>'CTX" is the core 'code name', so to speak, and isn't written on it.
>You can tell if it has, in gold lettering, "26050 " along the edge on
>the ceramic chip carrier. It also would have a CPUID=0580h.


All right. I notice that both my old cpu and new cpu had the code
26351 on the carrier. Initially the pull I bought had a bunch of
remnant thermal paste on it, so seeing the same code on the carrier,
and not being able to make out the lettering on the actual chip (due
to the paste residue) made for some anxious moments: "Did I just get
taken, and buy another K6-2 350MHz?" I quickly rubbed off the paste,
and lo, there was the "500" that gave me relief.


>
>
>A million circuits is peanuts these days. Even a K6-2 has 9.3 million
>transistors in it and the corresponding Celeron (PPGA) has 19 million.
>
>Athlon XP, Barton core, is 53 million and a Northwood P4 is 55
>million.

53 million? Ay carumba. Amazing. I remember reading some time ago
that cpus were reaching the limits of physics with regards to how
densely the circuits were packed on the chip and how much heat and
cross-current they would generate (or some such thing). The article
said they were testing other materials for building chips. This was
back in the early Athlon days I think. Since then, chips have come
quite a long way, so I guess that article was a little off.

>
>For those of us who remember discrete transistors it's staggering.
>Imagine, a transistor with a failure rate of one per million years
>sounds rather decent but that would mean a K6-2 failure every 5 weeks,
>or so!

I never even considered that. One per million years. So, the
circuits on a cpu are much more resillient than that? Must be. I
don't hear a lot of stories about cpus wearing out under normal
operating conditions... only under overclocking. Of course, most
processors reach their limits of usefulness before they wear out.

>
>If you watch it closely, yes. But a system can appear 'locked up' and
>still be consuming power and a processor can consume power when it's
>'crazy' .
>
>That's how my experiment got away from me. I didn't notice it was
>locked up, as it was an extended test, and it cooked itself to death
>even though, for all appearances, it was 'doing nothing'.
>
>Point is, you can't count on it going back to 'cool' just because the
>system 'locks up'. It isn't 'fail safe'.

Good to know. Those freeze ups are kind of scary in that light. My
MSI board with the 2500+ on it has some monitoring equipment on it. I
eventually turned it off because it kept giving an alarm on my case
fan speed. That system is running completely at spec, so I wasn't
concerned about over-stressing the system, so giving processing power
to a monitoring system that I didn't think was all that reliable
wasn't that compelling. Perhaps it was giving a warning because my
case fan is of quite large diameter, and likely spins slower than a
smaller fan would.

Actually, I'm a little embarrased to say, I failed to set the clock
speed properly in my first 6 or 8 months of operation. It was way
underclocked, since it was assembled by the computer store, and either
they didn't set the multiplier, or I had to do so much flashing to get
the built-in network card to work that it reset the multiple to
factory default.

It's kind of humorous in hind sight how poorly the mainboard operated
fresh out of the box. They were shipping product with major operating
flaws and with no documentation on how to make the networking to work.
I can't imagine figuring it out on my own, without the help of the MSI
forum. That is, with help from other users, not from MSI themselves.

Of course, I was lucky that was a second computer, not my first ever,
or else I would have had no way of connecting to the internet to see
how to make it work. I think the problem had to do with a MAC
address. The user has to manually enter the MAC address into BIOS.
Usually, to find the MAC address, the user has to use a mirror to see
the little sticker on the mainboard hidden close to the edge of the
case.

It's a good think I like puzzles. A tinkerer will find joy in weird
situations like this where someone else will just find frustration.

Sorry for the tangent! :-)

gene
February 23, 2005 12:45:18 AM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

On Tue, 22 Feb 2005 12:19:23 -0600, gene <noemailplease@nowhere.com>
wrote:

>Hi flipper,
>
>I think I'm about done with my overclocking now. I don't think I'll
>bother pushing the multiplier any more than the 496MHz it's running
>at. I'm sure I could easily push it to 558MHz, but my cpu cooling is
>marginal, so I think I'll leave it at that.

What's your core voltage at? Because that is the biggest factor in
'extra heat' from overclocking. Power goes up with the square of
voltage but linearly with speed.

>
>I haven't done any of the extra CPUburn tests, but I guess I should.
>Wouldn't be too difficult.
>
>>Can't tell from "AFR." That's the suffix designating package A (they
>>all are), F Vcore range (2.1 to 2.3, in that case), and R temperature
>>range (0-70C, in that case).
>>
>>'CTX" is the core 'code name', so to speak, and isn't written on it.
>>You can tell if it has, in gold lettering, "26050 " along the edge on
>>the ceramic chip carrier. It also would have a CPUID=0580h.
>
>
>All right. I notice that both my old cpu and new cpu had the code
>26351 on the carrier. Initially the pull I bought had a bunch of
>remnant thermal paste on it, so seeing the same code on the carrier,
>and not being able to make out the lettering on the actual chip (due
>to the paste residue) made for some anxious moments: "Did I just get
>taken, and buy another K6-2 350MHz?" I quickly rubbed off the paste,
>and lo, there was the "500" that gave me relief.

Oh wow. Sorry for the heart attack and I know don't how I got the
numbers mixed up but 26050 is the 'old' K6 core. 26351 IS the CTX
core.

I must have mispasted from the wrong column in the data sheet.

>>A million circuits is peanuts these days. Even a K6-2 has 9.3 million
>>transistors in it and the corresponding Celeron (PPGA) has 19 million.
>>
>>Athlon XP, Barton core, is 53 million and a Northwood P4 is 55
>>million.
>
>53 million? Ay carumba. Amazing. I remember reading some time ago
>that cpus were reaching the limits of physics with regards to how
>densely the circuits were packed on the chip and how much heat and
>cross-current they would generate (or some such thing). The article
>said they were testing other materials for building chips. This was
>back in the early Athlon days I think. Since then, chips have come
>quite a long way, so I guess that article was a little off.

Yes, I've seen articles talking like that too. They're always in the
context of 'what we know' and what current techniques are, though, and
people have a tendency to come up with new creative ideas <g>.

If you go WAY back to the germanium transistor days it was known that
silicon transistors were theoretically possible but you just couldn't
make them because it was impossible to get silicon pure enough for the
physics to work. Well, that is, until Texas Instruments figured out
how to make silicon that pure. And to add even more humor to the
story, they used a process that was known to not be good enough, and
by several orders of magnitude. hehe Just that no one had though to
repeat the unsatisfactory process and, as it turned out, it improved
exponentially on each pass.

Make one wonder who had the balls to ask "what happens if we do this
known-not-to-work useless process twice?" And why anyone listened.


>>For those of us who remember discrete transistors it's staggering.
>>Imagine, a transistor with a failure rate of one per million years
>>sounds rather decent but that would mean a K6-2 failure every 5 weeks,
>>or so!
>
>I never even considered that. One per million years. So, the
>circuits on a cpu are much more resillient than that? Must be. I
>don't hear a lot of stories about cpus wearing out under normal
>operating conditions... only under overclocking. Of course, most
>processors reach their limits of usefulness before they wear out.

Yep. Now, when you get to a detailed analysis of where failures come
from it turns out that 'packaging' and 'mechanical things, like the
little wires connecting the thing to the outside world, are a major
source of failures and, of course, on an integrated circuit the
connections are internal so some of that improvement comes
'naturally'.

But a lot is purity of the materials (see silicon transistors) and how
the dope layers are deposited and so on.

It's not a trivial thing, though. People laugh at how weak old
computers were but even more of a problem was keeping them running. It
was a red letter, banner headlines, day when 'large scale' computers
(vacuum tube) reached the reliability point that you could repair them
faster than they broke down. LOL That meant you could run it during
the day, put an all night maintenance crew to repairing it, and then
run again the following day! My goodness, what will they think of
next?

That is one reason why no one in the 1950/60s was really all that
worried about megalomaniac computers taking over the world. Who the
hell would repair them?

My favorite, though, is the computer that was afraid of the dark:
turn the lights off and it would go nuts. Apparently the light
provided some 'extra' stimulation to either the little neon indicator
lights or the vacuum tubes and without it the circuits would misfire.


>>If you watch it closely, yes. But a system can appear 'locked up' and
>>still be consuming power and a processor can consume power when it's
>>'crazy' .
>>
>>That's how my experiment got away from me. I didn't notice it was
>>locked up, as it was an extended test, and it cooked itself to death
>>even though, for all appearances, it was 'doing nothing'.
>>
>>Point is, you can't count on it going back to 'cool' just because the
>>system 'locks up'. It isn't 'fail safe'.
>
>Good to know. Those freeze ups are kind of scary in that light.

Not really. It isn't that the lock up itself is a heat problem. It's
mistakenly thinking one can leave it powered up like that. And it's
not like you have to catch it in 5 seconds. Just don't leave it that
way, which no one looking at it would do anyway.

> My
>MSI board with the 2500+ on it has some monitoring equipment on it. I
>eventually turned it off because it kept giving an alarm on my case
>fan speed. That system is running completely at spec, so I wasn't
>concerned about over-stressing the system, so giving processing power
>to a monitoring system that I didn't think was all that reliable
>wasn't that compelling. Perhaps it was giving a warning because my
>case fan is of quite large diameter, and likely spins slower than a
>smaller fan would.

Very possible. But that should be an adjustable parameter too.


>Actually, I'm a little embarrased to say, I failed to set the clock
>speed properly in my first 6 or 8 months of operation. It was way
>underclocked, since it was assembled by the computer store, and either
>they didn't set the multiplier, or I had to do so much flashing to get
>the built-in network card to work that it reset the multiple to
>factory default.

Flash probably did it. Some will reset to default on a set number of
failed power up posts too.

>It's kind of humorous in hind sight how poorly the mainboard operated
>fresh out of the box. They were shipping product with major operating
>flaws and with no documentation on how to make the networking to work.
>I can't imagine figuring it out on my own, without the help of the MSI
>forum. That is, with help from other users, not from MSI themselves.

Well, the above story about early computers will put it in some
perspective then. <g>

>
>Of course, I was lucky that was a second computer, not my first ever,
>or else I would have had no way of connecting to the internet to see
>how to make it work. I think the problem had to do with a MAC
>address. The user has to manually enter the MAC address into BIOS.
>Usually, to find the MAC address, the user has to use a mirror to see
>the little sticker on the mainboard hidden close to the edge of the
>case.
>
>It's a good think I like puzzles. A tinkerer will find joy in weird
>situations like this where someone else will just find frustration.

Yes. Well, as they say, computers are not toasters ;)  There's no other
'appliance' that requires anywhere near the 'smarts' to operate.


>Sorry for the tangent! :-)
>
>gene
February 23, 2005 12:01:26 PM

Archived from groups: alt.comp.periphs.mainboard.elitegroup (More info?)

Hello Flipper,

Your last post had me chuckling. Lots of neat history there.


>What's your core voltage at? Because that is the biggest factor in
>'extra heat' from overclocking. Power goes up with the square of
>voltage but linearly with speed.

Core voltage is running at 2.2. That's the factory specification.
Everything is hunky dory as far as I can tell.

>>All right. I notice that both my old cpu and new cpu had the code
>>26351 on the carrier. Initially the pull I bought had a bunch of
>>remnant thermal paste on it, so seeing the same code on the carrier,
>>and not being able to make out the lettering on the actual chip (due
>>to the paste residue) made for some anxious moments: "Did I just get
>>taken, and buy another K6-2 350MHz?" I quickly rubbed off the paste,
>>and lo, there was the "500" that gave me relief.
>
>Oh wow. Sorry for the heart attack and I know don't how I got the
>numbers mixed up but 26050 is the 'old' K6 core. 26351 IS the CTX
>core.
>
>I must have mispasted from the wrong column in the data sheet.

Oh, that's all right. Actually, it wasn't your numbers that freaked
me out, this was before you ever mentioned the gold numbers. I just
thought those gold numbers could be used to figure out which CPU/speed
was on the carrier. I thought that each processor would have a unique
"gold" number. I see now that the cpu itself has the unique number
telling about it's recommended operating conditions and manufacture
date, in addition to the name and speed which is clearly marked.


>
>>>A million circuits is peanuts these days. Even a K6-2 has 9.3 million
>>>transistors in it and the corresponding Celeron (PPGA) has 19 million.
>>>
>>>Athlon XP, Barton core, is 53 million and a Northwood P4 is 55
>>>million.
>>
>>53 million? Ay carumba. Amazing. I remember reading some time ago
>>that cpus were reaching the limits of physics with regards to how
>>densely the circuits were packed on the chip and how much heat and
>>cross-current they would generate (or some such thing). The article
>>said they were testing other materials for building chips. This was
>>back in the early Athlon days I think. Since then, chips have come
>>quite a long way, so I guess that article was a little off.
>
>Yes, I've seen articles talking like that too. They're always in the
>context of 'what we know' and what current techniques are, though, and
>people have a tendency to come up with new creative ideas <g>.
>
>If you go WAY back to the germanium transistor days it was known that
>silicon transistors were theoretically possible but you just couldn't
>make them because it was impossible to get silicon pure enough for the
>physics to work. Well, that is, until Texas Instruments figured out
>how to make silicon that pure. And to add even more humor to the
>story, they used a process that was known to not be good enough, and
>by several orders of magnitude. hehe Just that no one had though to
>repeat the unsatisfactory process and, as it turned out, it improved
>exponentially on each pass.
>
>Make one wonder who had the balls to ask "what happens if we do this
>known-not-to-work useless process twice?" And why anyone listened.

It must have been someone who actually washed their own dishes. "You
know, when I was oatmeal off a bowl, sometimes there's some stuck on
after the first wash. Sometimes I have to wash it again."

As far as people coming up with creative ideas to further technology,
that's really true as well. It has been amazing how much faster and
better computers have become even from 10 years ago.

Of course, Telstar's pong game was the cat's pajamas back in the 70s.
We also had an apple computer with these stupid "paddles"; black knobs
you turn with a little red button on them. That was the big advance
of the Apple II computer; from Pong to Breakout. ;-)

It's surprising how long PCs went without someone developing a decent
sound card.

>
>
>>>For those of us who remember discrete transistors it's staggering.
>>>Imagine, a transistor with a failure rate of one per million years
>>>sounds rather decent but that would mean a K6-2 failure every 5 weeks,
>>>or so!
>>
>>I never even considered that. One per million years. So, the
>>circuits on a cpu are much more resillient than that? Must be. I
>>don't hear a lot of stories about cpus wearing out under normal
>>operating conditions... only under overclocking. Of course, most
>>processors reach their limits of usefulness before they wear out.
>
>Yep. Now, when you get to a detailed analysis of where failures come
>from it turns out that 'packaging' and 'mechanical things, like the
>little wires connecting the thing to the outside world, are a major
>source of failures and, of course, on an integrated circuit the
>connections are internal so some of that improvement comes
>'naturally'.
>
>But a lot is purity of the materials (see silicon transistors) and how
>the dope layers are deposited and so on.
>
>It's not a trivial thing, though. People laugh at how weak old
>computers were but even more of a problem was keeping them running. It
>was a red letter, banner headlines, day when 'large scale' computers
>(vacuum tube) reached the reliability point that you could repair them
>faster than they broke down. LOL That meant you could run it during
>the day, put an all night maintenance crew to repairing it, and then
>run again the following day! My goodness, what will they think of
>next?

Wow. Amazing. Think how it would be if we had to repair our
computers every night. People get peeved when they have to reboot
after installing new software.

I've never seen a computer read those paper programming cards, but
when I think of those cards, I think of a Simpsons episode where Apu
is showing the Simpsons the boxes of cards he used to write a program
that got him his Computer Science degree at Cal Tech (Calcutta Tech).
Bart grabs a random card and asks "What does this one do?".
Crestfallen, Apu takes the entire box of cards and empties it into the
garbage, since the sequence of cards has been disturbed, ruining his
program.


>
>That is one reason why no one in the 1950/60s was really all that
>worried about megalomaniac computers taking over the world. Who the
>hell would repair them?
>
>My favorite, though, is the computer that was afraid of the dark:
>turn the lights off and it would go nuts. Apparently the light
>provided some 'extra' stimulation to either the little neon indicator
>lights or the vacuum tubes and without it the circuits would misfire.

That's a funny story too. <Chuckle>


>
>
>>>If you watch it closely, yes. But a system can appear 'locked up' and
>>>still be consuming power and a processor can consume power when it's
>>>'crazy' .
>>>
>>>That's how my experiment got away from me. I didn't notice it was
>>>locked up, as it was an extended test, and it cooked itself to death
>>>even though, for all appearances, it was 'doing nothing'.
>>>
>>>Point is, you can't count on it going back to 'cool' just because the
>>>system 'locks up'. It isn't 'fail safe'.
>>
>>Good to know. Those freeze ups are kind of scary in that light.
>
>Not really. It isn't that the lock up itself is a heat problem. It's
>mistakenly thinking one can leave it powered up like that. And it's
>not like you have to catch it in 5 seconds. Just don't leave it that
>way, which no one looking at it would do anyway.


Oh, all right. That's not so bad then. Just read a book within sight
of the monitor while it runs its tests. Or, like I was reading last
night, a computer supply catalogue. Reading that made me think
perhaps the rate of technological advance in chip speed is progressing
a little slower. Seems like Intel has been using the P4 platform a
long time, and my 1 1/2 year old Barton 2500+ is still running very
well. I suppose if I were a major gamer, I might be looking at
upgrading, but I'm still very happy with its performance.

I suppose there will come a day when computers are sufficient for most
user's needs and upgrading will slow. However, maybe this is one of
those predictions that will sound ridiculous in a few years, like
"Someday, computers will fill only a single room."
>
>> My
>>MSI board with the 2500+ on it has some monitoring equipment on it. I
>>eventually turned it off because it kept giving an alarm on my case
>>fan speed. That system is running completely at spec, so I wasn't
>>concerned about over-stressing the system, so giving processing power
>>to a monitoring system that I didn't think was all that reliable
>>wasn't that compelling. Perhaps it was giving a warning because my
>>case fan is of quite large diameter, and likely spins slower than a
>>smaller fan would.
>
>Very possible. But that should be an adjustable parameter too.

Yes, that might be the case. I suppose at the time, I had other more
pressing needs, or just wasn't that interested in the problem. The
constant alarm was kind of unnerving, so I turned it off. Ignorance
was bliss, and I suppose it still is.
>
>
>>Actually, I'm a little embarrased to say, I failed to set the clock
>>speed properly in my first 6 or 8 months of operation. It was way
>>underclocked, since it was assembled by the computer store, and either
>>they didn't set the multiplier, or I had to do so much flashing to get
>>the built-in network card to work that it reset the multiple to
>>factory default.
>
>Flash probably did it. Some will reset to default on a set number of
>failed power up posts too.

Ok, I'll have to watch that I check it after every flash. Actually,
II just confirmed it's running at the right speed, since I can't
remember if I flashed it since I set the speed properly.

>
>>It's kind of humorous in hind sight how poorly the mainboard operated
>>fresh out of the box. They were shipping product with major operating
>>flaws and with no documentation on how to make the networking to work.
>>I can't imagine figuring it out on my own, without the help of the MSI
>>forum. That is, with help from other users, not from MSI themselves.
>
>Well, the above story about early computers will put it in some
>perspective then. <g>

lol. Yeah, that's for certain. It's great that once a machine is
running, rarely do I have to open the side panel to see what's going
on in there. This is evidenced by the amount of dust that accumulates
in there.
>
>>
>>Of course, I was lucky that was a second computer, not my first ever,
>>or else I would have had no way of connecting to the internet to see
>>how to make it work. I think the problem had to do with a MAC
>>address. The user has to manually enter the MAC address into BIOS.
>>Usually, to find the MAC address, the user has to use a mirror to see
>>the little sticker on the mainboard hidden close to the edge of the
>>case.
>>
>>It's a good think I like puzzles. A tinkerer will find joy in weird
>>situations like this where someone else will just find frustration.
>
>Yes. Well, as they say, computers are not toasters ;)  There's no other
>'appliance' that requires anywhere near the 'smarts' to operate.

I'm glad that tinkering with computers doesn't require near the amount
of tools that a mechanic (they call them Automobile Technicians now
around here) needs to fix a vehicle.

gene
!