View Single Post
  #4  
Old February 3rd 05, 02:50 AM
Michael Brown
external usenet poster
 
Posts: n/a
Default

James Bond wrote:
I purchased this mainboard soon after it was released and fitted it
with Corsair XMS2-5400 DDR2 RAM, P4 540 3.2E GHz CPU (D0 stepping),
Abit RX600 XT-PCIE graphics card, and Creative Labs Sound Blaster
Audigy2 ZS Platinum Pro PCI card. I have 4 x 250GB SATA hard disks
in two RAID arrays (using the onboard RAID).

I then added watercooling using Asetek blocks on the CPU, VGA, and NB.

The best semi-stable overclock I can manage is about 240/3840 MHz.
This is done with the following settings:
VCo 1.6125V, DDR2V: 1.90V, NBV: 1.70V
CPU temp at idle: 45C, CPU temp at load: 56C, Ambient temp: 25C

On very hot days the CPU full load temp approaches 60C, whereafter the
machine normally crashes.

[...]

I suspect that the overclocked PCIe bus might be at the root of this
problem.


If it was the PCIe bus, I suspect you'd be seeing it long before 240MHz.
Most PCI start to malfunction around 37-38MHz (corresponding to a FSB of
~225MHz), but your system is (almost ) stable at 40MHz. Perhaps PCIe
devices are more tolerant to an overclocked bus, I don't really know.

What you need to to is take the standard overclocking approach and isolate
each part by itself.
1) Get rid of the SB card.
2) If you have a PCI graphics card that you're familiar with (in terms of
what it will take), use that instead of the X600
3) Remove everything except video card, RAM, and CPU.
4) Find the limit of you RAM. Keep the FSB and memory speed in sync (or even
run the memory above the FSB if possible) and ramp up the speed until
memtest86 gives errors (or the system fails to boot ).
5) Find the limit of your CPU. Ideally, use a bootable Linux CD image along
with Prime for stability testing (there's several out there, but don't know
the names off the top of my head). Alternatively, use a hard disk you're
familiar with, or one of your SATA drives if you don't have anything
valuable on there yet. Do not use RAID. I usually let Prime run for about 5
minutes during the "first stage", increasing the FSB if it can last 5
minutes. When it starts to error, keep backing down until it can do a full
24 hour run. You may also want to drop the RAM speed a notch to be
absolutely sure that there's no problem there.

If you do this, you've largely isolated the two main components (CPU and
RAM). Pick a speed that both should work fine at, and slowly add components.
Run a test that focusses on that particular component (for example 3DMark to
make sure your graphics card is happy). You should then find where your
stability problems are, or alternatively end up with a system that is
stable, and pretty much sitting at the maximum speed possible.

If it's your CPU which is limiting you (which I suspect, given that you
mention that temperature has an effect on stability) then run one of the
throttle-monitoring programs. If your CPU is throttling at high loads, then
you're just going to have to back down the voltage/frequency or cool it
better.

Another thing to try is to take the side off your case (if it's in a case)
and blow air in using a desk fan. One "problem" with watercooling is that it
changes the airflow across the motherboard from what has been designed. The
airflow from the NB fan (for example) might also be cooling a couple
MOSFETs. With the fan removed, these components heat up more under load and
can cause stability problems. Creating a lot of airflow across the board
will show if this is potentially an issue (though won't resolve it entirely,
as it will also cool down things like the CPU, etc).

Overclocking is a continual process. You don't just get your system and
immediately go to the highest possible speed. You find the limits of each
component, and continually explore different configurations, different
cooling methods. It can take months of weekend fiddling until you get close
to the maximum of your system. I was still getting improvements out my main
system a year after I installed most of the bits, using pretty much the same
cooling components as I started with. Also, every system is different. You
need to find out the limits orf your particular system by isolating
components and testing them. Overclocking everything at once it not the way
to go, as you end up in the situation you are now, unsure of where or what
the problem is.

[...]
--
Michael Brown
www.emboss.co.nz : OOS/RSI software and more
Add michael@ to emboss.co.nz ---+--- My inbox is always open