Archived from groups: alt.comp.hardware.homebuilt (More info?)

I am shortly going to wipe my hard drive and reinstall my OS(es). This
seems like a good time to make any changes to system set up without risk of
instability compromising my data - since I will have backed it all up and
can reinstall the OS(es) again if they should become corrupted.

I shall detail my machine specs at the bottom of the post.

So, I am wondering about the following:

PCI Latency - in the past I have used as high as 160 and as low as 32. This
board allows me 32, 64 and 128. I'd be very interested in what factors have
caused experienced builders to choose one setting over another as I am
looking to set it to the optimal for my system, obviously and any factors
you have found which impact on it or are impacted by it are of interest.

Memory timing - It is my understanding that CAS latency, RAS to CAS delay
and RAS precharge should all be as low as can be achieved stably. RAS
activation however I am less sure of.. obviously too short is not good, but
is it optimal to set it as long as possible?

AGP - has anyone noticed any real performance gains from an out of spec AGP
bus? I have frequently had mine out of spec, sometimes as high as 80 Mhz
but I am not entirely convinced one way or the other and would be interested
in experiences and theory from others.

I have always aimed to have my FSB as high as possible even at the cost of a
slightly lower effective CPU speed, rather than have a higher multiplier at
a lower bus speed. Do you agree this is correct trade-off when
overclocking? If not, why? Indeed, if so, why? LOL

OK, system specs:-

AMD Athlon XP 3000+ cpu
Gigabyte GA-7NNXP m/b
nVidia FX 5600 gfx
2 x 512Mb DDR-400 branded mem.
Maxtor 160Gb SATA-150 hdd
Pioneer 105 DVD-ROM
MSI 52x CD-RW

FSB is 202MHz, CPU has 11.5 multiplier (2323MHz)
FX 5600 is stock nVidia part and timings on 82MHz bus
(Core 325, Memory 400)
Memory is obviously "DDR-404" as it were, timing:
CAS 2, RAS to CAS 2, RAS precharge 2,
RAS activation 8, i.e. 2-2-2-8
SATA Controller is Silicon Image 3112, drive is first channel.
Pioneer is RPC-1 (BIOS 'fix'), MSI is stock.

Well, that is a pretty full description of my system.. I hope I have been
clear enough in what I am asking, I look forward to your input ;¬)

John

P.S. I am not a novice I just would like the benefit of your experience to
add to my own, especially on the areas where I have no clear 'optimum'

Archived from groups: alt.comp.hardware.homebuilt (More info?)

On Sat, 2 Oct 2004 08:27:16 +0000 (UTC), "Gaidheal"
<some.one@some.isp.net> wrote:

>I am shortly going to wipe my hard drive and reinstall my OS(es). This
>seems like a good time to make any changes to system set up without risk of
>instability compromising my data - since I will have backed it all up and
>can reinstall the OS(es) again if they should become corrupted.
>
>I shall detail my machine specs at the bottom of the post.
>
>So, I am wondering about the following:
>
>PCI Latency - in the past I have used as high as 160 and as low as 32. This
>board allows me 32, 64 and 128. I'd be very interested in what factors have
>caused experienced builders to choose one setting over another as I am
>looking to set it to the optimal for my system, obviously and any factors
>you have found which impact on it or are impacted by it are of interest.
>

Generally it's best left at default values, probably 32 for
nForce2, unless you happen to have a PCI device that's not
getting enough time, then raise latency. Since you have the
Silicon Image controller you might want to bench that drive
while simultaneously doing something like playing an MP3
from a CD, then try a latency adjustmet and re-benchmark in
same scenario. The goal not being to expect utmost
performance figure, but rather how the latency change
effects it... or chose your own more likely event using PCI
bus while there'd be drive access. For example, copying
from a DVD to destination across a LAN, using the network
interface controlled by the chip on your board (Intel?), not
the southbridge-intergral nForce controlled adapter.

Or another more common way of looking at the latency setting
(the generic response): if it ain't broke, don't fix it,
rather leaving it at default value.

Another test might be copying from the SATA drive across the
lan with the Gb adapter and checking network throughput,
since it's likely to be effeted enough by PCI thoughput that
if PCI latency changes make a difference, it'd show up
then... but it may require using jumbo frames to keep CPU
utilization low enough, and of course a machine on the other
end capable of receiving at >= same rate.

>Memory timing - It is my understanding that CAS latency, RAS to CAS delay
>and RAS precharge should all be as low as can be achieved stably. RAS
>activation however I am less sure of.. obviously too short is not good, but
>is it optimal to set it as long as possible?

Generally 9-11.

>
>AGP - has anyone noticed any real performance gains from an out of spec AGP
>bus? I have frequently had mine out of spec, sometimes as high as 80 Mhz
>but I am not entirely convinced one way or the other and would be interested
>in experiences and theory from others.

It made more difference with AGP 2X, these days it's best to
leave it at default and get entire system running stable
before changing it... generally it's better to leave it
locked @ 66.

>
>I have always aimed to have my FSB as high as possible even at the cost of a
>slightly lower effective CPU speed, rather than have a higher multiplier at
>a lower bus speed. Do you agree this is correct trade-off when
>overclocking? If not, why? Indeed, if so, why? LOL


Generally yes, except that sometimes you reach a ceiling
where memory will do certain timings if you back the FSB
down a few notches (with memory synch'd to FSB as it should
be on nForce, and Athlon chipsets in general). Of course it
also depends on most demanding use of system and how much of
a reduction in CPU clock speed.


>OK, system specs:-
>
>AMD Athlon XP 3000+ cpu
>Gigabyte GA-7NNXP m/b
>nVidia FX 5600 gfx
>2 x 512Mb DDR-400 branded mem.
>Maxtor 160Gb SATA-150 hdd
>Pioneer 105 DVD-ROM
>MSI 52x CD-RW
>
>FSB is 202MHz, CPU has 11.5 multiplier (2323MHz)
>FX 5600 is stock nVidia part and timings on 82MHz bus
>(Core 325, Memory 400)
>Memory is obviously "DDR-404" as it were, timing:
>CAS 2, RAS to CAS 2, RAS precharge 2,
>RAS activation 8, i.e. 2-2-2-8
>SATA Controller is Silicon Image 3112, drive is first channel.
>Pioneer is RPC-1 (BIOS 'fix'), MSI is stock.
>
>Well, that is a pretty full description of my system.. I hope I have been
>clear enough in what I am asking, I look forward to your input ;¬)
>
>John
>
>P.S. I am not a novice I just would like the benefit of your experience to
>add to my own, especially on the areas where I have no clear 'optimum'
>

Haven't tried overclocking an FX5600 but you might be able
to squeeze a bit more out of it, especially the GPU speed.
Depending on the revision of your m'board's northbridge it
might hit closer to 220MHz FSB, possibly requiring relaxing
memory timings. Overall though, the tweaking at this point
is an up-hill battle, it's at the point where it'll take a
lot more work (and stability testing) for only a little more
gain, might be best left as-is unless you feel the CPU has
another couple hundred MHz in it.

Test with http://www.memtest86.com for several hours after
any FSB or memory change, before booting OS.

Archived from groups: alt.comp.hardware.homebuilt (More info?)

"Or another more common way of looking at the latency setting (the generic
response): if it ain't broke, don't fix it,
rather leaving it at default value."

Sounds reasonable ;¬) But being me, I'll probably attempt to make the PCI
bus the limiting factor on some activity and then mess with latency to see
what happens.

"Generally 9-11."

Thanks!

"It made more difference with AGP 2X, these days it's best to leave it at
default and get entire system running stable before changing it... generally
it's better to leave it
locked @ 66"

This seems to tally with my own experience, so I shall probably do exactly
that.

"Generally yes, except that sometimes you reach a ceiling
where memory will do certain timings if you back the FSB
down a few notches (with memory synch'd to FSB as it should be on nForce,
and Athlon chipsets in general). Of course it also depends on most
demanding use of system and how much of a reduction in CPU clock speed."

Memory ceiling on this system is indeed 202MHz, I don't think the CPU has
more than about 50MHz left, hence it is set the way it is (202x11.5)

"Haven't tried overclocking an FX5600 but you might be able to squeeze a bit
more out of it, especially the GPU speed. Depending on the revision of your
m'board's northbridge it might hit closer to 220MHz FSB, possibly requiring
relaxing memory timings."

FX5600 will overclock, I think I decided to leave it stock until I'd messed
with everything - normally I rewrite BIOS with altered settings (yes, I know
that's risky! LOL Did say I was not a novice ;¬) ). My m/b is one of the
first nForce2 "all singing, all dancing, special edition" boards. Very nice
it is, even now. But I believe the fastest the N'bridge will cope with is
210 and other on-board controllers lose it at about 205, from experience.
202 seems to work nicely. Not happy with my PSU, mind you, especially as
others seem to have a low opinion of Q-Tec as well. Memory timings did need
relaxing, for those [above current] speeds and in fact I can't have
aggressive cpu-memory timings (not surprising!) with current speed.
However, I established that the memory was not error-free above 202.

"Test with http://www.memtest86.com for several hours after any FSB or
memory change, before booting OS."

That is precisely what I did and always do, but thanks! Memtest is what
showed that even though it booted fine, the memory was not 100%. I am very
demanding and sometimes do precision maths of a scientific nature, so this
was not acceptable to me even though I would probably never notice in my
gaming.

By the way, you might want to try Memtest86+ from http://www.memtest.org/ it
is an updated version of the now (so I believe) unsupported Memtest86. For
stress tests I use "CPU Burn-in" from here
http://users.bigpond.net.au/cpuburn/

Thanks muchly, kony. As ever you are a great resource ;¬)

John

Archived from groups: alt.comp.hardware.homebuilt (More info?)

On Sun, 3 Oct 2004 01:33:19 +0000 (UTC), "Gaidheal"
<some.one@some.isp.net> wrote:


>FX5600 will overclock, I think I decided to leave it stock until I'd messed
>with everything - normally I rewrite BIOS with altered settings (yes, I know
>that's risky! LOL Did say I was not a novice ;¬) ).

Not having an FX5600 I can't be sure, but think there's a
possibility of flashing an FX5600 Ultra bios which might
relax the memory timings a bit, potentially squeezing a bit
more out of it. I know that is the case with the FX5900
cards, but they use BGA memory and have better results with
Hynix than Samsung chips. Generally it's safer to get the
Ultra bios, edit it to reduce the core & mem speeds in line
with what the card would already do, then see if "coolbits"
(or whatever you prefer) will allow futher o'c than with
prior bios, then re-edit the ultra bios with the new tested
stable settings before 2nd flash. Or, maybe that's all a
waste of time on an FX5600, don't even know if those have
the bios-programmable GPU voltage changing feature or not.


>My m/b is one of the
>first nForce2 "all singing, all dancing, special edition" boards. Very nice
>it is, even now. But I believe the fastest the N'bridge will cope with is
>210 and other on-board controllers lose it at about 205, from experience.
>202 seems to work nicely. Not happy with my PSU, mind you, especially as
>others seem to have a low opinion of Q-Tec as well. Memory timings did need
>relaxing, for those [above current] speeds and in fact I can't have
>aggressive cpu-memory timings (not surprising!) with current speed.
>However, I established that the memory was not error-free above 202.

Might be just as well though, too much testing and tweaking
and you hardly get to use it before it seems old again.

>
>"Test with http://www.memtest86.com for several hours after any FSB or
>memory change, before booting OS."
>
>That is precisely what I did and always do, but thanks! Memtest is what
>showed that even though it booted fine, the memory was not 100%. I am very
>demanding and sometimes do precision maths of a scientific nature, so this
>was not acceptable to me even though I would probably never notice in my
>gaming.
>
>By the way, you might want to try Memtest86+ from http://www.memtest.org/ it
>is an updated version of the now (so I believe) unsupported Memtest86. For
>stress tests I use "CPU Burn-in" from here
>http://users.bigpond.net.au/cpuburn/
>


I wasn't aware that Memtest86 was now unsupported, seems
like the last update was only 6 months ago. I have and have
used memtest86+ but found it buggy, crashing sometimes on
otherwise completely stable systems, so I only use it as a
last resort if non-+ memtest86 won't run.

Archived from groups: alt.comp.hardware.homebuilt (More info?)

On Sun, 03 Oct 2004 03:33:11 GMT, kony <spam@spam.com>
wrote:


>Not having an FX5600 I can't be sure, but think there's a
>possibility of flashing an FX5600 Ultra bios which might
>relax the memory timings a bit, potentially squeezing a bit
>more out of it.

BUT, there's strong chance the VIVO may not work anymore, if
it uses a different chip for that on the (different) card's
firmware. You might find reports from others about which
firmwares will preserve this feature, in web forums most
likley.

Archived from groups: alt.comp.hardware.homebuilt (More info?)

I usually just hack the BIOS and alter the timings directly, without
downloading another BIOS version, unless I see a later 'stock' BIOS, in
which case I install that first and then alter it.

John

P.S. I couldn't care less about the VIVO btw - always considered it an
additional liability. If I do any serious video work I get a dedicated
capture/import/export board.