Athlon 64 - seek advice on higher FSB problem

G

Guest

Guest
Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

Hi People,

I am overclocking my new system:

MSI Neo2 Platinum
Athlon 64 3000+ Winchester
2 x Corsair 512MB XMS CL2 PC400
Arctic Freezer 64

For my first experiments I set:

CPU mult = 9x
DDR maxclock = 133MHz
PCI/AGP = 66MHz
HT mult = 3x

With this I started looking at different configurations of Vcore
and FSB (hence CPU Freq.). I am pretty much at the beginning
but I plotted an x-y chart of the results so far. Using survival of
the most intensive Prime 95 test for 12 iterations as a criteria, the
red points represent unstable configurations and the blue points
stable configurations:

http://home.arcor.de/david_ffm/oc_res1.jpg

Now it looks to me as if there is a curve close to linear separating
the stable from the unstable area, but that it has a lid on it somewhere
near 2500MHz. Up to this point I didn't need to increase Vcore much,
but above this point even larger increases in Vcore didn't seem to help
stability at all. This suggested to me that something other than the CPU
was causing problems above FSB 275 MHz, so to test this hypothesis
I took a configuration that was just unstable (FSB 285 MHz) and
reduced the CPU multiplier to 6x, so that it was well below spec. This
was also unstable which seems to confirm my idea. But what else could be
causing problems? RAM and HT should be well within spec in this config.
Could it be the board itself?

Now something else weird: the config as above with
CPU mult = 6x and FSB = 275 MHz (CPU 1650 MHz)
is unstable (P95 error after a few minutes at most).

But the IDENTICAL config with CPU mult = 9x (CPU 2475 MHz)
is stable! At least it has survived over 4 hours of P95 torture!

How is that possible? Is the CPU multiplier of 9 particularly
favourable for some reason?

I'm not hell-bent on getting the very last MHz out of this CPU,
but I feel that since I got to 2475 MHz with little effort, moderate
Vcore increase, and reasonable temperatures, it would be a pity
to stop there because of what seems to be a problem with some
other unidentified component.

It seems to me that a higher CPU multiplier would be a good way
to go - is there any (not too risky) way of hacking the lock on that?

I'd be most grateful for any thoughts on these issues or advice,
especially since this is my first go at overclocking!

TIA
David
 
G

Guest

Guest
Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

also try alt.comp.hardware.amd.x86-64


--
Ed Light

Smiley :-/
MS Smiley :-\

Send spam to the FTC at
uce@ftc.gov
Thanks, robots.
 

chip

Distinguished
Nov 16, 2001
513
0
18,980
Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

"David Johnstone" <david_ffm@mail.com> wrote in message
news:crc5mf$mbr$05$1@news.t-online.com...
> Hi People,
>
> I am overclocking my new system:
>
> MSI Neo2 Platinum
> Athlon 64 3000+ Winchester
> 2 x Corsair 512MB XMS CL2 PC400
> Arctic Freezer 64
>
> For my first experiments I set:
>
> CPU mult = 9x
> DDR maxclock = 133MHz
> PCI/AGP = 66MHz
> HT mult = 3x
>
> With this I started looking at different configurations of Vcore
> and FSB (hence CPU Freq.). I am pretty much at the beginning
> but I plotted an x-y chart of the results so far. Using survival of
> the most intensive Prime 95 test for 12 iterations as a criteria, the
> red points represent unstable configurations and the blue points
> stable configurations:

The DDR Maxclock setting is a little unusual. Of course its *supposed* to
work OK asynchronously, but trying to run your FSB at such high speeds and
your memory at such low speeds may be stressing something somewhere. Did
you try any overclocks with the memory and FSB at 1:1? Or is your ram not
up to it?

> http://home.arcor.de/david_ffm/oc_res1.jpg
>
> Now it looks to me as if there is a curve close to linear separating
> the stable from the unstable area, but that it has a lid on it somewhere
> near 2500MHz.

Seems reasonable. That's pretty much how you expect modern CPU's to behave.
To get significantly beyond a given CPU's "natural" maximum, you often have
to increase the voltage *significantly* and introduce exotic cooling to keep
the temps down. Spectacular overclocks you see (3GHz+) are often with 2v or
more on the CPU.

BTW, you don't mention what CPU temps your are getting when running Prime95.
You really want low 50's - or even lower - for decent overclocking and
stability. The CPU is OK for up to 70C (as reported by the themal diode),
but you really don't want the temps going that high if you are trying to
squeeze performance out of it. Apart from anything else, I wouldn't trust
the motherboard to accurately report the temps anyway, so the lower you can
get them the better.

> Up to this point I didn't need to increase Vcore much,
> but above this point even larger increases in Vcore didn't seem to help
> stability at all. This suggested to me that something other than the CPU
> was causing problems above FSB 275 MHz, so to test this hypothesis
> I took a configuration that was just unstable (FSB 285 MHz) and
> reduced the CPU multiplier to 6x, so that it was well below spec. This
> was also unstable which seems to confirm my idea. But what else could be
> causing problems? RAM and HT should be well within spec in this config.
> Could it be the board itself?

I doubt it. More likely your CPU, or the FSB/Memory speed combination not
working too happily together.

> Now something else weird: the config as above with
> CPU mult = 6x and FSB = 275 MHz (CPU 1650 MHz)
> is unstable (P95 error after a few minutes at most).
>
> But the IDENTICAL config with CPU mult = 9x (CPU 2475 MHz)
> is stable! At least it has survived over 4 hours of P95 torture!
>
> How is that possible? Is the CPU multiplier of 9 particularly
> favourable for some reason?

I don't know *why* this happens, but its quite common for certain CPU's to
like some multipliers rather than others. I had an old Athlon XP mobile
that would not do 2500MHz under any circumstances, unless the multiplier was
12x. And then it would do around 2550 at "only" 1.8v.

> I'm not hell-bent on getting the very last MHz out of this CPU,
> but I feel that since I got to 2475 MHz with little effort, moderate
> Vcore increase, and reasonable temperatures, it would be a pity
> to stop there because of what seems to be a problem with some
> other unidentified component.

I suspect you have simply found the sensible limits of your CPU. Like I
said above, I am sure it could go higher if you start to get fanatical about
it. But if you are not "hell-bent" on overclocking then I reckon 2500MHz is
about your limit.
>
> It seems to me that a higher CPU multiplier would be a good way
> to go - is there any (not too risky) way of hacking the lock on that?

Nope. None whatsoever.

I suspect you could go a little higher with your CPU and FSB at 1:1. (i.e.
set Max Memclock to "Auto"). But perhaps you need better ram for that. And
I suspect another 1GB of decent TCCD memory would cost you more than a new
(faster) CPU.

> I'd be most grateful for any thoughts on these issues or advice,
> especially since this is my first go at overclocking!
>
> TIA
> David

Cheers

Chip
 
G

Guest

Guest
Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

Hi Chip,
thanks very much for the informed and informative reply!

> > MSI Neo2 Platinum
> > Athlon 64 3000+ Winchester
> > 2 x Corsair 512MB XMS CL2 PC400
> > Arctic Freezer 64
> >
> > For my first experiments I set:
> >
> > CPU mult = 9x
> > DDR maxclock = 133MHz
> > PCI/AGP = 66MHz
> > HT mult = 3x
> >
> > With this I started looking at different configurations of Vcore
> > and FSB (hence CPU Freq.). I am pretty much at the beginning
> > but I plotted an x-y chart of the results so far. Using survival of
> > the most intensive Prime 95 test for 12 iterations as a criteria, the
> > red points represent unstable configurations and the blue points
> > stable configurations:
>
> The DDR Maxclock setting is a little unusual. Of course its *supposed* to
> work OK asynchronously, but trying to run your FSB at such high speeds and
> your memory at such low speeds may be stressing something somewhere. Did
> you try any overclocks with the memory and FSB at 1:1? Or is your ram not
> up to it?


The term "DDR Maxclock" in the bios is somewhat misleading.
What the bios actually seems to do is to work out the divider it
would need to use on the cpu frequency to keep the ram at or
below this DDR Maxclock, but assuming that the CPU was
running at its nominal 1800MHz. For example, running at FSB
270MHz CPU mult 9x (2430MHz) and DDR Maxclock 133 MHz
the system uses a CPU/DDR ratio of 14. This would give a ram
freq. of 128,57MHz if the CPU were running at 1800MHz, but since
it is running at 2430 MHz the true ram freq is 173,57MHz.

The Corsair XMS CL2 RAM is only *rated* for 200MHz but it
is also supposed to be especially suited to overclocking and is
even *guaranteed* to at least be indestructable, so I will try
running it with higher maxclock and see what happens. The only
options in the bios are 100,133,166 and 200. Maybe the CPU/RAM
ratio is important too.


> >
> > Now it looks to me as if there is a curve close to linear separating
> > the stable from the unstable area, but that it has a lid on it somewhere
> > near 2500MHz.
>
> Seems reasonable. That's pretty much how you expect modern CPU's to behave.
> To get significantly beyond a given CPU's "natural" maximum, you often have
> to increase the voltage *significantly* and introduce exotic cooling to keep
> the temps down. Spectacular overclocks you see (3GHz+) are often with 2v or
> more on the CPU.


I have a few more points now http://home.arcor.de/david_ffm/oc_res2.jpg
Interestingly at the default voltage of 1,4V I can get all the way to 2367MHz.
It looks like the curve is flat, then linear for a short bit with a gradient of
about 4,3MHz per mV, and then flat again up to 1,6V which is as far as
I have dared to go. So the most sensible operating regions would be at
the beginning of one of the two flat areas, either 1,40V 2340MHz or
1,5162V 2502MHz (perhaps a tad more).

>
> BTW, you don't mention what CPU temps your are getting when running Prime95.
> You really want low 50's - or even lower - for decent overclocking and
> stability. The CPU is OK for up to 70C (as reported by the themal diode),
> but you really don't want the temps going that high if you are trying to
> squeeze performance out of it. Apart from anything else, I wouldn't trust
> the motherboard to accurately report the temps anyway, so the lower you can
> get them the better.

That is a concern, certainly. I am reading temps of up to 54°C. The temperature
in my lounge is also cool now, 15°, so I am asking myself what will happen in
summer at 35° ambient. The arctic freezer 64 is supposed to achieve a thermal
gradient of 0,20°C/W but I guess it depends where you measure the delta-T.
I may consider water cooling - that seems like a moderate and reuseable
investment.

>
> > Up to this point I didn't need to increase Vcore much,
> > but above this point even larger increases in Vcore didn't seem to help
> > stability at all. This suggested to me that something other than the CPU
> > was causing problems above FSB 275 MHz, so to test this hypothesis
> > I took a configuration that was just unstable (FSB 285 MHz) and
> > reduced the CPU multiplier to 6x, so that it was well below spec. This
> > was also unstable which seems to confirm my idea. But what else could be
> > causing problems? RAM and HT should be well within spec in this config.
> > Could it be the board itself?
>
> I doubt it. More likely your CPU, or the FSB/Memory speed combination not
> working too happily together.

I will try increasing DDR Maxclock, even if it means overclocking the ram,
and see what happens.

> > Now something else weird: the config as above with
> > CPU mult = 6x and FSB = 275 MHz (CPU 1650 MHz)
> > is unstable (P95 error after a few minutes at most).
> >
> > But the IDENTICAL config with CPU mult = 9x (CPU 2475 MHz)
> > is stable! At least it has survived over 4 hours of P95 torture!
> >
> > How is that possible? Is the CPU multiplier of 9 particularly
> > favourable for some reason?
>
> I don't know *why* this happens, but its quite common for certain CPU's to
> like some multipliers rather than others. I had an old Athlon XP mobile
> that would not do 2500MHz under any circumstances, unless the multiplier was
> 12x. And then it would do around 2550 at "only" 1.8v.
>
> > I'm not hell-bent on getting the very last MHz out of this CPU,
> > but I feel that since I got to 2475 MHz with little effort, moderate
> > Vcore increase, and reasonable temperatures, it would be a pity
> > to stop there because of what seems to be a problem with some
> > other unidentified component.
>
> I suspect you have simply found the sensible limits of your CPU. Like I
> said above, I am sure it could go higher if you start to get fanatical about
> it. But if you are not "hell-bent" on overclocking then I reckon 2500MHz is
> about your limit.
> >
> > It seems to me that a higher CPU multiplier would be a good way
> > to go - is there any (not too risky) way of hacking the lock on that?
>
> Nope. None whatsoever.

Pity. I read that some people remove the heatspreader - if you
do that there are no bridges under there like the XP?

> I suspect you could go a little higher with your CPU and FSB at 1:1. (i.e.
> set Max Memclock to "Auto"). But perhaps you need better ram for that. And
> I suspect another 1GB of decent TCCD memory would cost you more than a new
> (faster) CPU.

Thanks for the tips - I will see if I can make any progress without
higher voltages!

One other question - how reliable are the voltage measurements? Mine are all below
nominal, some well below. Perhaps I have PSU problems (Seasonic 350W).

David
 
G

Guest

Guest
Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

Not sure how much I can add from my own experiences, overclocking is more an
art than a science. You would think that increasing Vcore would always
increase stability, but it is not true. Temperatures are very important
including the ambient room temperature. I did not achieve stability with my
2600+ XP-M at 2.5 GHz until I added an extra 80 cm case fan. I would lose
my NIC card about once, every 48 h, until I dropped the Vdd down to 1.6 V.

Now it has been very stable and I usually keep the system idle time to a
minimum because I am always running calculations on it.

It might be worthwhile backing off on some of the voltages and see what
happens.

Happy OC'ing,

Pete

"David Johnstone" <david_ffm@mail.com> wrote in message
news:crf1u0$ik1$02$1@news.t-online.com...
> Hi Chip,
> thanks very much for the informed and informative reply!
>
>> > MSI Neo2 Platinum
>> > Athlon 64 3000+ Winchester
>> > 2 x Corsair 512MB XMS CL2 PC400
>> > Arctic Freezer 64
>> >
>> > For my first experiments I set:
>> >
>> > CPU mult = 9x
>> > DDR maxclock = 133MHz
>> > PCI/AGP = 66MHz
>> > HT mult = 3x
>> >
>> > With this I started looking at different configurations of Vcore
>> > and FSB (hence CPU Freq.). I am pretty much at the beginning
>> > but I plotted an x-y chart of the results so far. Using survival of
>> > the most intensive Prime 95 test for 12 iterations as a criteria, the
>> > red points represent unstable configurations and the blue points
>> > stable configurations:
>>
>> The DDR Maxclock setting is a little unusual. Of course its *supposed*
>> to
>> work OK asynchronously, but trying to run your FSB at such high speeds
>> and
>> your memory at such low speeds may be stressing something somewhere. Did
>> you try any overclocks with the memory and FSB at 1:1? Or is your ram
>> not
>> up to it?
>
>
> The term "DDR Maxclock" in the bios is somewhat misleading.
> What the bios actually seems to do is to work out the divider it
> would need to use on the cpu frequency to keep the ram at or
> below this DDR Maxclock, but assuming that the CPU was
> running at its nominal 1800MHz. For example, running at FSB
> 270MHz CPU mult 9x (2430MHz) and DDR Maxclock 133 MHz
> the system uses a CPU/DDR ratio of 14. This would give a ram
> freq. of 128,57MHz if the CPU were running at 1800MHz, but since
> it is running at 2430 MHz the true ram freq is 173,57MHz.
>
> The Corsair XMS CL2 RAM is only *rated* for 200MHz but it
> is also supposed to be especially suited to overclocking and is
> even *guaranteed* to at least be indestructable, so I will try
> running it with higher maxclock and see what happens. The only
> options in the bios are 100,133,166 and 200. Maybe the CPU/RAM
> ratio is important too.
>
>
>> >
>> > Now it looks to me as if there is a curve close to linear separating
>> > the stable from the unstable area, but that it has a lid on it
>> > somewhere
>> > near 2500MHz.
>>
>> Seems reasonable. That's pretty much how you expect modern CPU's to
>> behave.
>> To get significantly beyond a given CPU's "natural" maximum, you often
>> have
>> to increase the voltage *significantly* and introduce exotic cooling to
>> keep
>> the temps down. Spectacular overclocks you see (3GHz+) are often with 2v
>> or
>> more on the CPU.
>
>
> I have a few more points now http://home.arcor.de/david_ffm/oc_res2.jpg
> Interestingly at the default voltage of 1,4V I can get all the way to
> 2367MHz.
> It looks like the curve is flat, then linear for a short bit with a
> gradient of
> about 4,3MHz per mV, and then flat again up to 1,6V which is as far as
> I have dared to go. So the most sensible operating regions would be at
> the beginning of one of the two flat areas, either 1,40V 2340MHz or
> 1,5162V 2502MHz (perhaps a tad more).
>
>>
>> BTW, you don't mention what CPU temps your are getting when running
>> Prime95.
>> You really want low 50's - or even lower - for decent overclocking and
>> stability. The CPU is OK for up to 70C (as reported by the themal
>> diode),
>> but you really don't want the temps going that high if you are trying to
>> squeeze performance out of it. Apart from anything else, I wouldn't
>> trust
>> the motherboard to accurately report the temps anyway, so the lower you
>> can
>> get them the better.
>
> That is a concern, certainly. I am reading temps of up to 54°C. The
> temperature
> in my lounge is also cool now, 15°, so I am asking myself what will happen
> in
> summer at 35° ambient. The arctic freezer 64 is supposed to achieve a
> thermal
> gradient of 0,20°C/W but I guess it depends where you measure the delta-T.
> I may consider water cooling - that seems like a moderate and reuseable
> investment.
>
>>
>> > Up to this point I didn't need to increase Vcore much,
>> > but above this point even larger increases in Vcore didn't seem to help
>> > stability at all. This suggested to me that something other than the
>> > CPU
>> > was causing problems above FSB 275 MHz, so to test this hypothesis
>> > I took a configuration that was just unstable (FSB 285 MHz) and
>> > reduced the CPU multiplier to 6x, so that it was well below spec. This
>> > was also unstable which seems to confirm my idea. But what else could
>> > be
>> > causing problems? RAM and HT should be well within spec in this config.
>> > Could it be the board itself?
>>
>> I doubt it. More likely your CPU, or the FSB/Memory speed combination
>> not
>> working too happily together.
>
> I will try increasing DDR Maxclock, even if it means overclocking the ram,
> and see what happens.
>
>> > Now something else weird: the config as above with
>> > CPU mult = 6x and FSB = 275 MHz (CPU 1650 MHz)
>> > is unstable (P95 error after a few minutes at most).
>> >
>> > But the IDENTICAL config with CPU mult = 9x (CPU 2475 MHz)
>> > is stable! At least it has survived over 4 hours of P95 torture!
>> >
>> > How is that possible? Is the CPU multiplier of 9 particularly
>> > favourable for some reason?
>>
>> I don't know *why* this happens, but its quite common for certain CPU's
>> to
>> like some multipliers rather than others. I had an old Athlon XP mobile
>> that would not do 2500MHz under any circumstances, unless the multiplier
>> was
>> 12x. And then it would do around 2550 at "only" 1.8v.
>>
>> > I'm not hell-bent on getting the very last MHz out of this CPU,
>> > but I feel that since I got to 2475 MHz with little effort, moderate
>> > Vcore increase, and reasonable temperatures, it would be a pity
>> > to stop there because of what seems to be a problem with some
>> > other unidentified component.
>>
>> I suspect you have simply found the sensible limits of your CPU. Like I
>> said above, I am sure it could go higher if you start to get fanatical
>> about
>> it. But if you are not "hell-bent" on overclocking then I reckon 2500MHz
>> is
>> about your limit.
>> >
>> > It seems to me that a higher CPU multiplier would be a good way
>> > to go - is there any (not too risky) way of hacking the lock on that?
>>
>> Nope. None whatsoever.
>
> Pity. I read that some people remove the heatspreader - if you
> do that there are no bridges under there like the XP?
>
>> I suspect you could go a little higher with your CPU and FSB at 1:1.
>> (i.e.
>> set Max Memclock to "Auto"). But perhaps you need better ram for that.
>> And
>> I suspect another 1GB of decent TCCD memory would cost you more than a
>> new
>> (faster) CPU.
>
> Thanks for the tips - I will see if I can make any progress without
> higher voltages!
>
> One other question - how reliable are the voltage measurements? Mine are
> all below
> nominal, some well below. Perhaps I have PSU problems (Seasonic 350W).
>
> David
>
>