[weblacky]

Hello All,

I'm finally back with some news.
I received the chips and some others components.

This week, i replaced the G7G5 DS1780 chip which is in charge for measuring the PCI 5V.
And i have the regret to inform you that the issue remained.

So after an extensive reading of the LTC1629 datasheet and measuring some key tensions ( RUN/SS, EAIN, Ith, etc).
I decided to exchange this chip with the one in charge of the 3.3V.
And to my surprise the PCI 5V is now operational and so does the 3.3V next door.

Well at least the 3.3V was operational for a few minutes, and is now faulty with 0.98V.
So i guess that the LTC1629 is faulty and i'm gonna need to bite another bullet (16€/piece).

Thank you for reading me and I will keep you informed ASAP.

While this is a longshot before you go removing and replacing that chip again. SGI uses that chip on a lot of their late model boards. It's like a bit of a rubber-stamped design. The data sheet on that claims that the C-out capacitor for each voltage rail is designed to be low ESR and sometimes even ceramic (MLCC) on that controller.  For SGI this is SMD solid aluminum electrolytic or on some of them SMD polymer.

That was a very new thing for these converters back then. If you read troubleshooting documentation on these kind of converter controllers they claim that the converter process can experience odd issues on the C-out Cap that will cause a very odd wobble in error correction algorithm when the output capacitors ESR changes or is slightly too high that it basically makes the error correction nose dive after having an odd oscillation that goes wider and wider as the controller attempts to correct the voltage and this can result in very odd behavior where the voltage would drop especially if you saw it drop slowly but steadily over a period of let's say 30 seconds to several minutes. Because a lot of old electrolytic caps will change ESR suddenly as they heat up from usage this is why they work for the first couple seconds or minutes until their state basically changes from being heated up and now they've tried to reform from use but things obviously went wrong. So before you rip that new IC out I would urge you to look at your C-out capacitor for that 3.3 rail on that rail and check for what the ESR might be or just unusual capacitance or any other reading. Maybe even just replace it blind and see if suddenly you're back up and going. 

It's my belief that this phenomenon may be responsible for quite a few low-voltage VRM buses going bad and Tezro model machines as well as O350 1.8v rail.

It's easy to find out. Just try the replacement of the final output capacitor on that rail and let us know what happens. It can't make it any worse.

Also you could accidentally lose a diode in the simple configuration and have something similar happen. So I would also check that as well but the diodes may will measure really weird in-circuit because they're in reverse polarity for half the configuration and the other half they are designed to clean up the signal to the MOSFET gate so consider blind replacing every pair of SMB diode you encounter. Most of the old stuff is 600 mV forward bias you can't find anymore. I've replaced these with 700 mV forward bias with zero problem but make sure you replaced them in pairs because they have to fire in pairs so if they don't have the same forward bias they'd fire unevenly and do damage . But again a cheap part that's easy to check and easy to solder in and solder out of the layout

If either of these end up being your answer please do let us know as this is my working theory so I have a bit of a vested interest in the outcome.

[Shiunbird]

This is something I actually would love to ask a microprocessor designer...

I understand that a single core can run two threads - however, I also understand (hopefully rightly) that the efficiency with which a single core can run two threads will depend on what tasks these threads are running.

AFAIK some of the newest IBM power cpus can do 4 threads in a single core - I just have no idea how and what the tradeoffs are.

So... way beyond my knowledge.

[aSpartan111]

Hello All,

I'm a bit late to give you some news, but as we say, better late than never

As @weblacky suggested I first tried changing the output capacitors (for the 3.3V and Pci 5V).
During my readings i went through some posts stating that the Cout could be the source of a regulation issue.
I had ordered some Panasonic 6.3V 220µF caps with others components (DS1780, diodes, mosfets, etc) for this repair.
Unfortunately, it didn't work.

After that, i changed the LTC1629 responsible for the 3.3V.
(Reminder, I previously exchanged the IC for the PCI 5V with the one for the 3.3V next to it, moving with it the issue).
And luckily it solved the issue, now the 3.3V runs at a stable 3.320V.

This brick is now repaired, and i tested it over a week and a half.
The issue is definitely gone, (until next time, finger crossed).

I hope this thread will help someone else in the future,

Thank you all for your help and suggestions,
I wish you some good holidays,

[weblacky]

Thank you for coming back and updating the thread.

As some of us have indicated you're not unique in having this problem so it's interesting that replacing most of the functional component block worked.

Of course given the order of events we will never know exactly what needed to be replaced however I think it's a fair bet that a lot of those items probably needed to be replaced once the controller itself had been damaged.

Given your post said that initially the swapped controller functioned correctly for a moment until it nose dived again into failure. This could've easily indicated that yes it was damaged, but also that maybe it causes damage to one of the other components. Because again it was functioning after it hadn't functioned in its old footprint!

It's not out of the realm of possibility that a damaged controller also means that the items within its control may also have received some damage. I still believe replacing the cheaper stuff first was the better way to go but like you said it took replacing the controller and shotgunning a few other components.

For my own work in the future, could you please indicate which components you actually did solder on the board before you replaced the controller as the final step? You mentioned you purchased quite a few replacement components but that may not mean that you actually had to use them. Did you actually replace all the MOSFETs in that region or just the diodes?

Also in regards to the DS1780 I'm not 100% sure because most of the time SGI protected these with resistor networks, but individual sensor legs can be damaged and misreport voltage. I've seen this on two of my own repairs but I'm not hundred percent sure what causes the damage. It could be a voltage much higher than the anticipated sensing voltage. But I will agree they are extremely delicate and if you're going to replace a VRM section for repair if you don't get good reporting it is highly likely that the reporting mechanism was also damaged.

I tend to treat the DS1780 ICs more like canaries, in that yes they do warn of a voltage or environmental problem but they can die as they perform that warning.

[aSpartan111]

Hello,

Indeed, i did a group purchase of several components.
I purchased :
- DS1780E+
- 10BQ040 (Schottky diode, not sure it's the right reference...)
- Mosfet N IPD30N03
- Mosfet N IPD70N03
- Cap SVP 100µF 16V
- Cap SVP 220µF 6.3V
- Cap FK 680µF 16V
- LTC1629CG

And i only replaced :
- G7G5 DS1780E
- F2E2 220µF 6.3V (PCI 5V C out)
- H6E9 220µF 6.3V (3.3V C out)
- F7F7 LTC1629CG (3.3V Reg. IC)

And in retrospect i should have change the 2 CIn caps (F2F1 - H0E9).
I also could have change the diodes, if i was 100% sure it was the right replacement.