[kaigan]

Thanks for all of that information! I do actually have a fiberglass pen, so I should be able to work on this process to get the capacitor more properly installed. As a note, with the exception of the original re-soldering of the capacitor, I've used brand new 330uF aluminum electrolytic capacitors in that spot. I ordered 20 of them from Mouser, so I still have some brand new replacements, if needed.

For the ICs, they look to be Fairchild 6612A and 6670A MOSFETs, though there is one other type I'll mention as well. Here's a sample of the direct text from a couple of them:

E8F5:
D0306
FDD
6612A

F1E6:
D0250
FDD
6670A

Both G1G7 and J0G5 are FR9014 MOSFETs. Otherwise, all of the other ones with hard-to read text are 6612A or 6670A parts. There's also the MIC 29302BU voltage regulator I mentioned. Hopefully that's helpful. If you need any further information, just let me know!

[weblacky]

OK,
Great!  So I was on the money on that region.  The FDD 6612A: FDD6612A Data sheet
 
These are "D-PAK" packages, the large metal tab soldered downwards is the DRAIN (output). So those those tabs (and PCB ground) is where you want to start to measure DC Volts with your multimeter when running.  A power management IC reads the voltage output and wiggles the MOSFET gates to adjust for current draw (as voltage dips on draw).  A very simple little power supply.  For every two (pair) mosfets (close to one another) one DRAIN is the voltage, the other is actually ground.  Just use trial and error to figure out which is which.  You could also (when the board is unpowered) use the multimeter on continuity test to go from start of a COIL to a drain tab.  That tab will be the voltage tab.  But if you get 0V for one of the two tabs in a pair while the circuit is operating, then you know that's the ground MOSFET and the other MOSFET drain should be where your desired voltage is produced.  No harm will occur guessing wrong.
 
The coils and caps smooth it all over, before that, it's a bunch of ON/OFF pulses (so very dirty DC), the coil and cap combo help produce a stable DC output.  So if you have an oscilloscope you can measure at the gates to prove they are being operating correctly (not needed, just educational) and then you measure the drain tabs to show the dirty power coming out.  Measuring AFTER the cap should be your best power…if it’s not…see what it looks like on the other side of the cap (after the coil).  If it’s pretty much the same…BAD CAP.  
 
For replacing ALL the caps in this section, (I never done this personally, so based on my reading), You want to remove the caps by preheating the board and use hot air to remove them.  Then, trying getting solder paste, place solder paste down on the pads with a new (fluxed legs) cap, then attempt to hit it with hot air to get it to seat.
 
Solder paste has a lower melting point.  If you don’t want to use solder paste then you can try the hobbyist technique: Don’t use solder paste and use barely any solder, color change only, re-tin the cap pads, use flux on both legs, place cap down on PCB, introduce soldering iron from extreme side, touch each leg with small amount of new solder.  Try to get leg contact and sneak the heat and solder (using liquid flux) under the cap in a very close and quick maneuver.
 
 
If possible do your measurements first.  You’ll want to start with the problem caps first, then in case you run of steam to do the job, you’ve done the “have to have” caps already.

I've attached a simply laptop-style MOSFET DC-DC diagram to hope you visualize (best diagram I could find, sorry).  If you look at the MOSFETs in the diagram they don't have tabs, just legs.  Know that all 4 legs on the drain-side (DC Volt output) are tied together as one wire.  So in your head, replace the four legs where DC volts comes out as those large metal tabs on the ICs.  Then the coil and cap comes after the MOSFET drain (output).

[kaigan]

It looks like I'm getting around 0.188v from the drains on those MOSFETs, and along a number of the capacitors in that area. I'll need to do further investigation.

As an update, though, I am really concerned about the power coming in on 3.3v line. Further measurement with my meter is showing that a number of the PSU pins are related to the 3.3v line, but I'm getting varying readings from 3.22v to 3.28v along differing pins. My Altix 350s, which use the same pinout, are showing up at around 3.31-3.32v across all of the pins in a regular fashion. I've tried multiple PSUs and both of the PSU connectors on the O350 and I'm getting the same result.

I'm going to try using the NOS PSU connector that came with a NOS PSU that I ordered recently. If that doesn't make a difference, I'll try swapping out the incoming AC unit that connects to the PSUs. If neither of those make a difference, back to board testing I'll go.

[weblacky]

Hmm,
OK it's possible those DC converters don't produce 3.3v. Without finding the IC controlling their gate pins, and reading the datasheet, we may not know what voltage they are targeting. But I'm not sure what runs with 0.188v...that doesn't sound right, what did the signal visually look like on the oscilloscope? Was it stable DC or was it very erratic and jumpy by several volts? A multimeter will "average" a very dirty DC signal. So what the multimeter says may not be really representative of what's going on.

Can you use an oscilloscope instead? Look at the quality if of the voltage output.

In terms of the power supply, I'm a unclear on your description. I believe you are saying you're measuring power (backprobing) PSU pins while it's plugged in mainboard and ON.

Are you measuring at the PCB connector solder joints or via backprobing of the PSU connector (wire-side)? If you're measure from the PCB-side, please measure DC from the PSU connector side (rear of the plug connector where wires come back to PSU).

If you see a difference on the same wire (from the PSU side of the connector to the PCB connector tracks), then I'd recommend you reflow (or remove old solder and add new resolder, with new flux) that PSU PCB mainboard board connector pin manually with a soldering iron, as you may have a cracked or bad solder joint, causing resistance issues on a single pin. You're correct in that most PSUs have many wires/paths for a single voltage rail, they do this so the wires can be thinner and share the same output among multiple connectors. But because of this, any current draw or resistance change (or heating affect) will affect the voltage on ALL wires that share that DC voltage output.

So if one of your connector pins or soldering joints for a single wire causing issues, the whole 3.3v line will show issues.

I don't know if you can physically do this but, the best way to see if there is the connection problem is under load. To that end if you can measure voltage on BOTH sides of the PSU connector (while it's plugged into the PCB mainboard and ON) via a track and back-probing the connector at the same time. When you measure the same wire you should see 0v. But if you see something like 50mV (0.050v) or 100mV (0.100v) and all you've done to measure is stick the red test lead into the back of the PSU connector (while connected and ON) and held the black test lead to PCB track that comes right off that pin and you see tens of mV reading of difference or higher...then you know you have a connector/crimp/solder joint issue. If something within that 1" of space is dropping the voltage (wiggling it) then target that connector or solder joint.

If you measure from the back of the plugged-in PSU connector to the first output track for that same pin on the PCB and you get like .005v or something then that is NOT a connection issue, nor an issue with the connector's attachment to the PCB board. But age and wire torsion could have affected the connection, it's easy to measure under load and find out to make sure the voltage wiggle isn't a mechanical issue.

I recommend this over an Ohms test because we are looking for something so intermittent and sensitive, you may not see it with just an Ohm meter. But while handling real current, if you see that voltage drop (showing a voltage difference on the same wire just before and after the PSU connector). Then you know you have a connection issue.

Let us know what you find.

[hamei]

  A multimeter will "average" a very dirty DC signal.  So what the multimeter says may not be really representative of what's going on.

It's not as good as a 'scope but you can set a digital meter on an AC scale and measure the ripple on a DC line, too.

[kaigan]

No difference with the replaced hardware, so the issue is definitely related to the board itself.

I am backprobing the PSU, pins, yes. This is just with the system plugged in. I can't turn the system on without the L1 controller available. I've noticed that the green light on the PSU is blinking. According to the O350 manual, this indicates that, "The power supply is receiving AC power, but the main primary DC power has not yet activated." Given what the issue seems to be, I'd guess that the PSU isn't getting whatever signals it requires back to indicate that all is well.


I'll see what I can do about examining the connections from the other side of the PSU connector while it's plugged in. It's pretty difficult to tell where everything is going, but I'll see what shows up.

Also, I do have an oscilloscope. I'll go ahead and take a look at those MOSFETs again and see how things look there, too.

Thanks again!

[hamei]

Also, I do have an oscilloscope.

Good deal. For anyone who is going to mess around with electronics, a cheap scope is a really good tool. You see so much more of what is going on than with a meter. And it's no harder to use.  Highly recommended.

Edit : Okay, you inspired me. I guess that's the right word. I pulled out my Frankenfifty and put power to it. (Altix 350 case dremelled to accept O350 boards, WHY was SGI so stupid ? We're talking 1/4" here and there, would it be so technologically difficult to make the boards fit one universal case ? So that maybe we they wouldn't have to stock two entirely different cases ? No wonder those morons went broke. Twice.

Anyhow, you get nothing at all on the display ? I at least get "L1 booting" which sits there forever .... any suggestions, j-j ?

Yeah I know, attach a terminal. From an old Macbook ? (Thank god it's not a new one, those have no peripherals, not even a farging CD drive cuz, you know, THIN ! Who cares if it's useful or not, gotta be THIN !) Jesus. Could these people make life any more difficult if they tried ?

Computer companies suck.

[kaigan]

Anyhow, you get nothing at all on the display ?

The front display is completely blank. The system won't respond over serial or to pressing the power button. From what I've read, that's typically indicative of a bad Dallas chip, but I know for a fact that this chip is good (and I've tried a couple of good ones). The socket also seems to be okay as I do get continuity from each internal socket connector to the mainboard. I might probe it again just for absolute certainty, but it should be fine.

I'll have to take the scope to the board when I have time and see what I can figure out.

Computer companies suck.

In general, I have to agree.

[weblacky]

OK,
I’m going to perform a huge leap forward in theory after examining the board pictures and (trying) to thoroughly read the data sheet for the LTC1629CG-PG ICs.

Short answer: You have a short in your system, AFTER the regulation circuitry we’ve been looking at, that’s why your 3.3v lines wiggles at the PSU and why ALL MOSFET drains measure so low and all the same value. The regulators are fighting to control what’s happening and limit damage and current. Now the short could be in something you have to attach to the board to make it function (another card or PCB board). But something on the line regulated by the MOSFETs is shorted.

I’ll explain better, but all the regulators and all the MOSFETs are actually one, big, regulator (see datasheet, page 26). They are all regulating the same line as multiple “phases”. Phases in this sense isn’t like AC phases, it’s momentary on/off power that is shared by all the regulators to share thermal load for high-current applications to produce the rapid on/off switching pattern, like a switching PSU (DC), to create the new DC signal, but using several parallel components as a collective group, to emit power quickly in-turn (like round-robin) fashion to share the thermal load using a very quick time-slices. Think GPU VRMs or CPU VRMs.

Data sheet is here: https://www.analog.com/media/en/technica.../1629f.pdf

Printed page 26, describes the multiple controller layout (albeit with 6 MOSFETs on each controller instead of the 4 you have at each controller in your picture). You have two controllers with 4 MOSFETs each, total of 8 MOSFETS, but I believe this is counted as 4 phases (works in pairs).



OK, my explanation of my theory:

Just like a bench power-supply in current limited-mode, (though the overcurrent documentation isn’t too clear on the datasheet). When you exceed the maximum current rating, the regulators will show a voltage drop, the more severe the draw the bigger the voltage drop, lower the voltage value. In your case, I don’t know what the regulator is set for, is could be set for 1.6v, 3.3v, or higher (datasheet is unclear how Vout is set and has multiple graphs with several different Vout voltages, so this is unclear, I don’t know what yours is set for.

Since all MOSFETs are regulating the same line they are all in overload.

Something appears to be overdrawing a DC current that is present on AC plug-in alone. This short is happening while just plugged into the wall, without trying to power-on the system. The short is in stuff that just runs all the time (like the L1 circuitry).

The 3.3v (or the like) voltage rail is being pulled down to 0.188v DC, this appears to be a “Stable” short. Easiest solution is to get a thermal imager (even a cheaper phone-attachment model like a thermal seek would do). Look at “hot stuff” while the system is just on standby power and you should see one or more locations that light up. Look for the brightest first but look at all areas that come up. Then inspect for heat…lots of heat means lots of current dissipation. Hopefully you locate your short and it’s not a custom IC or something.

OR you could so buy a current tracer probe for less than a thermal imager on the used market but it’s harder to use.

OR you can use the dangerous bench PSU method to find the short:
Using a bench power supply, you set the voltage to something very low (don’t want to over-volt anything) like 1v-1.5v to start at 2A power. You locate the output of the regulator circuit (which very close to the coil output (but there is a beware issue here, some have 0Ohm shunt resistors on them to help measure current draw, those can blow, you want to do this after all 0,00, or 000 labelled SMD resistors, or green resistor/fuses) and on an unpowered, unplugged, separated from the system board, you apply power to the voltage rail that’s in short with the bench PSU, and watch the bench PSU display to see if the voltage just drops to 0v while the current Maxs out.

Then you know you have a short and use either your fingers or several different chemicals (like freeze-spray, rubbing alcohol, zippo light fluid, etc) to try to detect very hot spots on the board as you apply power to the shorted rail. You can increase the Amperage, but you don’t want to actually fully burn out the component that is shorted or it will go fully open and then draw nothing and unless it explodes, you’ll then never find it.

So, the trick is to heat it up (without breaking it) and try to find the bad component either by thermal touch, thermal imaging, or solution reaction to heat.

Also be aware if you remove the board and get NO current draw reaction from the bench PSU test, nor any thermal imaging hot-spots, then the short isn’t on that board…but on something else that is plugged into that board! You have to follow the boards (doing the same thing or the like, to track down the physical area with the shorted component that’s drawing down the entire regulated DC rail on this path.


To easily test the idea of the short being in another system component, get a kill-a-watt or other AC draw measuring device for your AC wall socket. And you could start to try to, slowly, unplug AC, remove a board or component, and reattach to the wall AC socket, check kill-a-watt meter for the power draw. If I’m right this power is really being drawn, so the meter should show like a few amps or something? Keep removing pieces with system off and unplug PSU every time you make a change, and reattach AC power. Maybe you can locate a CPU card, or add-in card, or something where the meter shows a sudden drop in current after you remove a piece of the system that isn’t on this board.

But again, if the short is on this board, then it really will be present all the time, if it’s an attachment, then changing the attachment would remove the short from the system and it would function again.

Kill-a-watts are like $25 or something...cheap to try, cool to have.

Anyway, that’s my theory, see what you can come up with.

[kaigan]

Thank you for another extensive and informative post! As additional information, literally the only things plugged into the system for my testing are a single, known-good PSU, the 2U Interface board, and the front LCD screen. I suppose I can try unplugging the LCD as well and see if things come up over serial. It seems unlikely, but it's worth the test.

I'll see about getting some sort of thermal imager. Presumably it'll take a few days to get here, so I'll provide an update once I've had a chance to look the board over with one.

Again, I really appreciate the info!