Clock problem

[Heffgé]

Sometime My PC doesn't keep date and time. Not when there's no main on the
power supply, not when it's running, only when it's just off for a few
minutes.

The battery have been switched, checked, etc., the problem seems to be
elsewhere.

The tests I ran show that battery feeds the CMOS RAM only when the power
supply is unplugged. When it's plugged it's seem the battery is relayed by
the standby voltage as there's no more current drained from it. When the
computer is on, I dont know, I couln't test further.

I also used an oscilloscope to check the crystal frequency. It's the same in
the 3 states (about 34 kHz measured but that scope is old and not that much
accurate).

I ran a small program in command mode which claims to access the RTC.
Windows time and RTC time match even after having running quite a long time.

I know CMOS RAM and RTC are both located in southbridge and backed up by the
battery. But I don't know how it should be feeded when the computer is
running or in standby mode. How is it supposed to work ?

The battery is OK, the oscillator works fine in any case but the registers
in the CMOS RAM are corrupted while the computer if off but still plugged.

What's wrong ?. The southbridge (VT8233) itself ?

Any help would be appreciated

Heffgé

PS : Be indulgent for my english

 

[Paul]

Sometime My PC doesn't keep date and time. Not when there's no main on the
power supply, not when it's running, only when it's just off for a few
minutes.

The battery have been switched, checked, etc., the problem seems to be
elsewhere.

The tests I ran show that battery feeds the CMOS RAM only when the power
supply is unplugged. When it's plugged it's seem the battery is relayed by
the standby voltage as there's no more current drained from it. When the
computer is on, I dont know, I couln't test further.

I also used an oscilloscope to check the crystal frequency. It's the same in
the 3 states (about 34 kHz measured but that scope is old and not that much
accurate).

I ran a small program in command mode which claims to access the RTC.
Windows time and RTC time match even after having running quite a long time.

I know CMOS RAM and RTC are both located in southbridge and backed up by the
battery. But I don't know how it should be feeded when the computer is
running or in standby mode. How is it supposed to work ?

The battery is OK, the oscillator works fine in any case but the registers
in the CMOS RAM are corrupted while the computer if off but still plugged.

What's wrong ?. The southbridge (VT8233) itself ?

Any help would be appreciated

Heffgé

PS : Be indulgent for my english

This is a reference schematic for a motherboard. (440BX with dual
processor schematic)

http://www.intel.com/design/chipsets/designex/BXDPDG10.PDF

If you look on page 32, in the lower left hand corner, there is
a zener regulator, which takes +5VSB (standby supply) and
regulates it down to "3VSB". That is a lesser standby voltage,
used for when the computer sleeps. The MMBZ5226BL zener used, is
a 3.3V zener, in a 225mW SOT-23 package. So "3VSB" might well
be close to 3.3V.

If you look on page 18, in the lower left hand corner, there is
shown the circuit creating the RTC_BAT voltage. There are
two Schottky diodes. They "diode-OR", picking one voltage
supply or the other. "BT1" is the CR2032 motherboard battery.
The 1K ohm resistor is present, because when the motherboard
battery is being used, the 82371EB is in a low power state, and
only draws maybe 10 microamps or so. When the 82371EB is running
in its normal state, then the "3VSB" voltage is available, and
larger transient currents may flow occasionally. But there is
no current limiting resistor in that path.

Right off hand, I cannot think of a reason why the circuit is
failing. Because it would fail to keep time sitting in the
BIOS screen, if the "3VSB" supply was completely dead.

You will notice on page 18 as well, the crystal used for the
RTC, is 32.768KHz, the same frequency as is used in a
digital watch. By dividing by 32768 (2**15), a once-per-second
pulse is created for time keeping. A simple ripple divider in
the RTC, does the division by 2**15.

HTH,
Paul

 

[RJK]

Cooh ! ...that's going back a few years - 440BX chipset !

....try disconnecting mains power, take out the CR2032 button battery, and
leave it out for a week, (for capacitors to discharge) - or if you can't
wait that long - whip out the board, (if you can't access the underside in
your particular case), and manually short / discharge all capacitors around
the CMOS. ...then refit board and battery, ...bios will have gone to
defaults so you'll have to redo all your bios settings of course.

....if that was the board that also occasionally gave the "Error acessing the
A20 Line" ...it was so long ago, ...I know it was a 440 something or other
!!

regards, Richard

 

[Heffgé]

I'm afraid it's not a problem of clearing CMOS or things like that. It has
already be done.

My board is a MS-6382 (MSI OEM for Medion) with a Via KT266A chipset where
southbridge is VT8233. While it's a bit more recent than a 440BX power
managment might be similar.

Diodes to pick up one ot the other from the 2 power sources is just what I
guessed. Thanks for your detailed explanations.

But even with a third source, the 5V VCC, that would explain everything is
OK while computer is running, I can't imagine what happens when it's turned
off. It seems the southbridge keeps on being feeded just time a capacitor is
discharged. If I switch off main a few seconds later clock is OK. If I wait
for a few minutes clock is late, from minutes to days. If I wait for hours
clock is reset to 01.01.2000.

What does that mean ?

By the way, I checked current from the battery : 20 µA when power supply is
unplugged, none in the other cases. It seems a lot ?

Heffgé

 

[Paul]

I'm afraid it's not a problem of clearing CMOS or things like that. It has
already be done.

My board is a MS-6382 (MSI OEM for Medion) with a Via KT266A chipset where
southbridge is VT8233. While it's a bit more recent than a 440BX power
managment might be similar.

Diodes to pick up one ot the other from the 2 power sources is just what I
guessed. Thanks for your detailed explanations.

But even with a third source, the 5V VCC, that would explain everything is
OK while computer is running, I can't imagine what happens when it's turned
off. It seems the southbridge keeps on being feeded just time a capacitor is
discharged. If I switch off main a few seconds later clock is OK. If I wait
for a few minutes clock is late, from minutes to days. If I wait for hours
clock is reset to 01.01.2000.

What does that mean ?

By the way, I checked current from the battery : 20 µA when power supply is
unplugged, none in the other cases. It seems a lot ?

Heffgé

The reason I presented the 440BX schematic, is because other boards continue
to use the same concept for powering. The diode used is typically a
dual diode in a SOT-23 package (has three legs). Sometimes that dual diode
gets burned, when a user does a "clear CMOS" with the power supply still
turned on. On three different motherboard models here, the SOT-23 has
"K45" stamped on the top, and the component is near the CR2032 battery.
(You need a magnifying glass to read the letters.)

There are now some chipsets, where the "clear CMOS" function is handled
by a separate pin on the Southbridge, which avoids the burned diode problem.

With the power running, you should take a look at the "3VSB" voltage value.
On a modern board, they may not use a Zener to make 3VSB, but just a small three
terminal regulator. I cannot honestly say I've identified where 3VSB comes
from, on my motherboards here.

When the power supply is switched on at the back of the computer,
then +5VSB is available. You can check with a multimeter and see if the
voltage is stable, both when the computer is running, or if the computer
is sleeping. +5VSB should never "wink out" or disappear on you, and should be
present for as long as the switch is ON, on the back of the computer.

Your 20 microamp measurement might not be that far off from the truth.
Since you aren't likely to be able to find a datasheet for a VIA chip,
the value might be more or less than competitive chips. To run a
digital watch, takes about 2 microamps. So the additional current
is leaking through the standby well (i.e. all the CMOS circuits powered
by VCC_RTC). It should not take a lot of current to run a ripple
divider with a 32KHz clock source. Maybe the additional leakage is
due to the 256 bytes of CMOS memory in the Southbridge.

The VCC_RTC voltage should never drop below 2.0 volts. That is a
typical value for the ability to retain the 256 byte CMOS contents.
When you allocate 0.4V for the Schottky voltage drop on either path
of the dual diode, that means the battery or 3VSB, must provide more
than 2.4V, for the RTC and CMOS to function properly.

(3.3V) (0.4 drop) (> 2.0V)
+5VSB --- 3VSB ----------------> dual_diode ------ VCC_RTC
regulator +-->
(3.0V) |
3.0V ----/\/\/\-----------+
CR2032 1K ~10uA
Battery ohm (a guess)

HTH,
Paul

 

[Heffgé]

A magnifying glass is not enough it needs also twenty years old eyes.

I tried to measure voltage on every 3 legs surface component hangind around
: nothing.

Something really puzzling : the variation of the clock is not as
mathematical as I previously said. Actually it happens sometime that the
clock is fast, from hours to days.

I guess I'll never know the origine of this problem. When I'm fed up with
this board I'll get rid of it.

Anyway, it has been an opportunity to learn interesting things.

Thanks a lot for your help.

FG

 

[Paul]

A magnifying glass is not enough it needs also twenty years old eyes.

I tried to measure voltage on every 3 legs surface component hangind around
: nothing.

Something really puzzling : the variation of the clock is not as
mathematical as I previously said. Actually it happens sometime that the
clock is fast, from hours to days.

I guess I'll never know the origine of this problem. When I'm fed up with
this board I'll get rid of it.

Anyway, it has been an opportunity to learn interesting things.

Thanks a lot for your help.

FG

Timekeeping in Windows is done by counting clock tick interrupts.
As I understand it, when Windows starts, the contents of the RTC
are copied. From that point, Windows counts seconds itself, by using
the known number of clock tick interrupts that happen per second.
Effectively, that means the crystal on the clock generator chip,
becomes the time keeping source, rather than the 32KHz watch crystal
next to the Southbridge.

It is possible for that interrupt based mechanism to malfunction.
For example, on an Nforce2 chipset, there is a problem when the
board isn't running at a canonical clock value (100, 133, 166, 200MHz).
Disabling APIC seems to fix it, implying there is some kind of
bug in the programmable interrupt controller. I'm not aware
of any other chipsets having that bug. The clock error that
can accumulate in Windows, on an Nforce2 board, can be huge
due to that bug. Not even Network Time Protocol is enough to
correct the error that results.

When your computer is switched on and off, and you're using your
oscilloscope, do you see anything funny happen to the VCC_RTC
voltage ? In that link to the schematic I posted, I see a 0.1uF
capacitor on the VCC_RTC, and presumably that should be enough to
filter out any transients.

Paul

 

[Heffgé]

I tried to disable IPCA in BIOS setup. Windows doen't start even in safe or
command mode.

The processor (Athlon) is overclocked (1550 MHz in state of the genuine
1330 ) but it doesn't seem to have any relation since I discovered the
problem before updating the BIOS in order FSB can be set.

I know Windows reads RTC just once. But it updates it when a change is done.

I picked up a program which displays CMOS memory content
(http://www.fabbridjm.net/). It seems RTC and Windows time match.

I got once more the oscilloscope from the basement and I tried to record
what I saw.

Measurements are done beetween ground and one of the crystal pin.

Unplugged : peak to peak voltage 0.7 V - offset 0.6 V

Plugged : peak to peak voltage 0.75 V - offset 0.7 V - transition rather
clean

Running : almost the same - it takes a few seconds to stabilize and there
are what might be very short breaks

Shutting down : nothing to notice

Unplugging : after time for capacitors to discharge offset and peak to peak
voltage fall down (a few tenths of mV) before it returns to normal state

With the other pin votages are higher (1,3 and 1,5 V). Offsets are about the
same and transitions too. But the signal shape is quite less cleaner that
the perfect sinusoid of the other case .

I don't think of any transient problem (may be I'm wrong) because there's at
least the POST delay beetween power supply is switched on and Windows looks
at the CMOS. And to see if time is OK I don't start Windows, I just enter
BIOS setup where that's RTC which is displayed.

Heffgé