Motherboard Power Requirements

[Arifi Koseoglu]

Dear Paul, hello again!

First of all, many many thanks again, as usual

As you say, the shunts seem to be more exepensive probably less practical,
but it turns out that my f-i-l has a clamp-on Ammeter, and will measure
the flow once he brings it along.

The 2.5" drive suggestion is also a very good idea - but this would ramp
up the cost - a Seagate 80G 5400RPM sells for 140 USD here - probably
would rather go directly to shuttle.

As soon as I can find a better 1U PSU, I am going to drop this one, but
yesterday night I gave the system a "blind" try.

Configuration:
1. Crappy PSU ( +3.3@7A +5@13A +12@6A +5VSB@1A)
2. MSI KM4AM-V with FSB set to 166 MHz
3. Sempron 3000+ running at rated speed.
4. Seagate 80G Sata drive ([email protected] [email protected])
5. LG 48X CDROM Player (+5V@1A [email protected])
6. 1.44 Floppy (?@? ?@?) (needed to load SATA drivers during OS
installation - and later for memory diagnostics)

The MoBo does not have any voltage or fsb control in the BIOS, and
provides jumpers to set the FSB to 100, 133, 166 or 200 only.

The system booted, was able to boot from the CD to install OS (W2KSRV).
Once the
OS was installed, I removed the CDROM.

The PSU was getting pretty warm during OS installation.

The BIOS reports the voltage values as follows (the least significant
digit changes in time):

VCORE = 1.53
+3.3V = 3.10
+5V = 5.18
+12V = 11.61
-12V = -12.28
-5V = -5.09
5VSP = 4.93

MSI's monitoring software reports +12V as 11.46 in Windows, other '+'
values are reported almost the same.

W2K SP4 installed w/o problems, and the PC idled w/o problems for > 1hr.

Then I ran Microsoft Memory Diagnostics (a scaled-down Memtest86). And, I
got several errors on my Kingmax 512MB DDR400 memory sticks. I have no
prior test results for them, except for the fact that they constituted
half of the 2Gigs I have on my own desktop which never crashed since its
first day (for over 4 months now - nock on the wood). I will have to
install a floppy drive and do the same test on my PC too (no CDROM since
the Intel MoBo will not allow more than 4 ATA devices - and I have 4
disks).

Is there a possibility that the PSU bottleneck is causing the memory
problems?

Today I will "push" the system by compressing large files. That should get
the HDD and CPU loaded at the same time.

Will post again with observations later.

Best,
-arifi

Hello again. Today I performed some compression on the system, here are the
proceedings:

Configuration like stated in the quoted lines above, except for no cdrom and
no floppy.

1. I have compressed directories of sizes from 2.3 GB to 16 GB using Winzip
9.0 SR1, using the portable maximum compression method.

2. Compressing 2.3 GB took approx. 10 minutes of continuous >98% load on the
CPU, and compressing 16GB took 55 mins of similar load. Continuos disk
access in these periods, too.

3.RAM usage did not exceed 140MB out of available 1GB.

4. During these operations the CPU temperature did not exceed 72 degrees
Celsius.

5. The voltage on the 5V rail was OK, but the +3.3V occasionally dropped to
3.08 and the low on the +12V line was 11.28V. These lines roughly averaged
3.15V and 11.35V, respectively.

6. No lock-ups were experienced, HOWEVER, the compressed file did not
succesfully de-compress in the 16GB case. I ran 13 compression sets in
total. Besides the 16GB, two 4.5GB sets also experienced de-compression
problems (Bad CRC).

7. Interesting (for me, at least), but probably irrelevant observation: In
one of the sets I tried simultaneous compression of three directories of
approx. 4.5GB in size. During this operation the CPU load averaged only 35%,
but the disk light was almost constanly on. I assume the HDD become the
bottleneck here and the CPU mostly had to wait for the HDD to finish I/O.
(One of the three outputs of this set did not de-compress due to CRC errors)

The system seems holding up, but what is the reason for the CRC errors in
large compression sets? A sign of insufficient power, bad RAM chips (Kingmax
PC3200), or just expected behavior for sucn large sets?

Tonight I will leave Prime95 running...

Best,
-arifi

 

[Arifi Koseoglu]

Tonight I will leave Prime95 running...

Best,
-arifi

Well, it crapped out at the first test....

 

[Paul]

Well, it crapped out at the first test....

I guess the +12V was finally too low for the PSU to
stay running. The +3.3V also seems to be getting
loaded pretty well too. Underclocking the processor
might help with the +12V, but I don't know how you can
help the +3.3V . Still, you got much further than I
expected. I didn't think you'd get an OS installed on
there, due to the hard drive and CDROM loading.

Maybe the errors on the disk are due to the voltage
dropping at some point during the compression process.
You might want to dig up whatever equivalent MSI has
to Asus Probe, and use the MSI utility to record
the measured voltages over time. By leaving that
utility running during your compression run,
you may actually get to see the voltage drop out
at some point.

And I wouldn't leave Prime95 running with the PSU
close to its limits. You never know whether the clever
people who made it, designed it to handle overloads
gracefully. You wouldn't want to come to work the
next morning to a cloud of smoke

Paul

 

[Paul]

Well, it crapped out at the first test....

If Shuttle components are available in your area, I noticed this
power supply PC40 from Shuttle. It is 250W and the same form factor
as your current supply. I cannot find out how many connectors and
what they do, so maybe your Shuttle dealer can help you there.
Should give you a little more power to work with, without buying
a complete Shuttle system.

http://www.newegg.com/app/viewproductdesc.asp?description=17-157-005&DEPA=0

3.23" x 1.7" x 7.48"
3.3@18A 5.0@19A 12@16A [email protected] [email protected]

Paul

 

[aberger]

Hello Arnie,

I will end up underclocking the CPU probably, if I cannot get a proper PSU
and problems arise, but performance is also a factor - not a happy
alternative.

I do not know much (actually anything) about solid state disks - but would
assume they would cost much more than standard ones. How are
capacity/performance characteristics ?

Thanks,
-arifi

They're certainly faster than a standard disk, so that could make up
for the speed issue. The cost obviously depends upon the capacity that
you need. However, the more I think about your application the more I
think that a good laptop will be a better solution.

With USB 2.0, you can hang as much stuff as you like on it and you can
always add an outboard monitor. I'm not sure what you get in terms of
flexibility of a small footprint desktop versus a laptop, other than
cost. However, I'm seeing really good laptops in the $1000 range right
now and you aren't going to save much more money. Also, doing a demo at
a customer's site (something I'm painfully experienced with) you would
really be better off not having to do a set-up. I worked for HP for 15
years and did a lot of demos myself. I think a laptop is the way to go.

arnie

 

[Arifi Koseoglu]

They're certainly faster than a standard disk, so that could make up
for the speed issue. The cost obviously depends upon the capacity that
you need. However, the more I think about your application the more I
think that a good laptop will be a better solution.

With USB 2.0, you can hang as much stuff as you like on it and you can
always add an outboard monitor. I'm not sure what you get in terms of
flexibility of a small footprint desktop versus a laptop, other than
cost. However, I'm seeing really good laptops in the $1000 range right
now and you aren't going to save much more money. Also, doing a demo at
a customer's site (something I'm painfully experienced with) you would
really be better off not having to do a set-up. I worked for HP for 15
years and did a lot of demos myself. I think a laptop is the way to go.

arnie

Dear Arnie,

Yes, a laptop is always a viable alternative. But, the ones that can be had
for $1000 (actually, we already have one), cannot provide the required
performance all the time, especially whenever we switch to a "Virtual
Server" environment (using VMWare) they come to their knees just too easily.

There would be stronger desktop replacement alternatives, of course, let's
see...

Cheers
-arifi

 

[Arifi Koseoglu]

If Shuttle components are available in your area, I noticed this
power supply PC40 from Shuttle. It is 250W and the same form factor
as your current supply. I cannot find out how many connectors and
what they do, so maybe your Shuttle dealer can help you there.
Should give you a little more power to work with, without buying
a complete Shuttle system.

http://www.newegg.com/app/viewproductdesc.asp?description=17-157-005&DEPA=0

3.23" x 1.7" x 7.48"
3.3@18A 5.0@19A 12@16A [email protected] [email protected]

Paul

Dear Paul,

I will check with the shuttle suppliers immediately on availibility of this
PSU.

Every single posting of you teaches me something new. Very much appreciated.
Thank you very much.

Last update and last question on the project: Without changes in the
configuration, I attempted another run of Prime95. To my surprise, this time
it ran for almost 4 hours before Prime95 complained about a rounding error
due to a hardware problem.

Should such inconsistent behavior (in the first run of Prime95 it had given
an error in the first minute - and now it runs for 4 hours) be blamed more
on the PSU, hence freeing me from the worries on the RAM modules ??

Thank you for everything,
-arifi

 

[Paul]

Dear Paul,

I will check with the shuttle suppliers immediately on availibility of this
PSU.

Every single posting of you teaches me something new. Very much appreciated.
Thank you very much.

Last update and last question on the project: Without changes in the
configuration, I attempted another run of Prime95. To my surprise, this time
it ran for almost 4 hours before Prime95 complained about a rounding error
due to a hardware problem.

Should such inconsistent behavior (in the first run of Prime95 it had given
an error in the first minute - and now it runs for 4 hours) be blamed more
on the PSU, hence freeing me from the worries on the RAM modules ??

Thank you for everything,
-arifi

That is possible. Perhaps you could watch with MBM5, and see how the
power supply voltages are running. The KM4M-V isn't listed, but
the KM4M is, so there is a small chance MBM5 might work. (The
author of Motherboard monitor has stopped supporting it, and
that is why the level of support for newer boards is unknown.)

http://mbm.livewiredev.com/comp/microstar.html
http://mbm.livewiredev.com/download.html

My experience with Prime95 and bad RAM, was Prime95 would consistently
find the bad RAM location. So with the same initial conditions, it
would error out after 30 minutes. Using a Knoppix boot disk and
a Linux version of Prime95, I was even able to break the memory
up into chunks and isolate the bad location even more. (I did
this by running four copies of Prime95, each with their own
folder, so the settings could be adjusted for each.) But that
is a case of bad RAM.

If it is giving more variable results, then it could be a processor
or Northbridge induced error. Or the power supply that powers
both of them.

Power consumption of Prime95 varies with FFT size. If you select
small FFT in the options, the Prime95 computation stays mainly
in processor cache, which makes the processor heat up a bit more.
That will likely be the best setting for detecting a power supply
problem. A variable FFT or large FFT setting, will test system
memory more. If you find Prime95 errors more quickly with small
FFT, that could be power related.

Paul

 

[Arifi Koseoglu]

Dear Paul,

I have tested the system running from two PSU's (one powering only the CPU
(the 12V 4Pin connector only - and other the rest ot the system - I used the
"crappy" PSU to power the CPu because the other was even crappier )

The system ran the torture test (blend and small FFT) without any problems
for one 4 and two 8 hour periods, without any problems. After each of the
sessions I tried the same test with the crappy PSU alone, and got errors
sooner or later.

I believe this is enough proof for the (already-diagnosed) PSU-based
problem. I believe the RAM chips etc are OK. We are ordering the Shuttle PSU
you pointed us to.

Many, many thanks again.

Cheers
-arifi