[weblacky]

Hi Jwhat,
The term "VRM" is kind of general and maybe I misused it in this case but what I mean is where ELF specifically highlights on SGUG on his mainboard (not IP board): https://forums.sgi.sh/index.php?threads/...rails.140/

There seem to be, in fact, two BUCK converter circuits (which do technically meet the criteria of VRM) on the actual mainboard. Elf has claimed this IS the source of the AUX 1.8v ENV reading. Assuming it's true, this is where I've stared at (picture-wise). While you cannot be 100% certain the ideal layout shown for these LDO controllers in their datasheet is what's actually used in the real tezro circuit...it will be similar.

The diodes are very easy to recognize (package format) and given the converter works on HIGH/LOW mosfets layout...there should be "pairs" of diodes since the layout mimic each other (just with polarity reversed between HIGH/LOW), where each diode from a pair are at each of the MOSFETS. In my case I think it's something like the signaling clean-up diode along with the aged output cap here (page 1): https://www.analog.com/media/en/technica...3728fg.pdf

So either the Zeners in M1 & M2 area (which MIGHT actually be built-in into body of a DUAl MOSFET package and not directly accessible...then I'd just change the chip) or the D3 and D4 diodes on the TG1 & TG2 along with the C_OUT cap.

That's my starting point when I get there for me as my voltage droops into problem..not cuts off or goes to zero instantly. So I'm staring at this picture a lot: https://forums.sgi.sh/index.php?threads/...18&slide=2

No other work at this time, only good thing is a have TWO Tezros and only one has this issue...so I can compare readings between them so hopefully it won't allude me when I get there....

if you want to take really nice close up pics of the 1.8 VRM region on your equipment and post them for us, that would be helpful to see more.

Thanks.

[jwhat]

Hi Weblacky,

I think all these “bits” are visible in photos on other O350 env location thread: https://forums.irixnet.org/thread-3424.html

The elf picture is not good enough quality to make out detail, but seems to be in equivalent area to my close up of DS1780s and Dallas RTC/NVRAM.

Can you see what you need on these pics ?

Cheers from Oz,

jwhat/John

[weblacky]

Hi Jwhat,
Yes, your previous picture is very good and can be used as a starting point.  What I’ve done is quickly highlight the power diodes that are obvious in these two voltage converters (I cannot tell you which is which).  Now these aren’t necessarily the only diodes is here.  This form factor (package) should be known as SMB, though it might be SMA. These are the same type of diodes you see in many SGI VRM circuits (like that Fuel PIMM I worked on).  Now there may be additional diodes on SOT packaging (three-legged smaller packages). But those will not be in a main voltage generating path (only a control or signal path) due to the size needed to withstand the voltage generating traces of the circuit (larger body required).  

You cannot test these in-circuit!  You must desolder each and let them cool and test outside of circuit.  But you can get them easily with either hot tweezers or two soldering irons at once and pluck them off.  

There is a fair (not definitive) chance you may find one or more either shorted or leaky (if you have equipment to show semiconductor leakage).  

What I found in that PIMM VRM was the lack of one in working order can cause excessively high voltage or low voltage depending on which MOSFET the diode was attached to.  

Since these are easily removed and tested, it’s low-hanging fruit to check!  Now also when you remove the diodes, leave them all out and check for shorts on the MOSFETS as well, with the diodes out you may get real readings of them!  With diodes in, all looks shorted and weird to me. 

I forget if your symptom is like mine (slopes into low voltage, versus instantly low voltage).  But instant is more more likely diode.  

To be clear, these kinds of diodes have seemed to change a little over the years and so if you find one bad…you need to swap both in the pair (or just do all of them).  Because you won’t find a matching diode.  The ones on that Fuel PIMM were like 600mV forward voltage and that isn’t available anymore (like it’s not a thing anymore to be offered) so I went with 700mV diodes with the same everything else. They worked fine, but that triggering level MUST BE IDENTICAL between pairs and also all DIODES in the common circuit path, so they fire at the same time.  

If you mix and match it’s like having you car fire some spark plugs much sooner than others which is basically misfire.  The diode inherent body design triggers when the voltage builds across that small area of the circuit, if they don’t fire nearly simultaneously you’ll end up with big problems, likely additions damage due to improper ability to handle load with only partial/initial firing diodes.  

They are super cheap parts, so it’s easier to just refresh them and try it.  I’d also do the output cap but, it’s less of an issue for testing and you could ignore it for purposes of a yes/no pass on the diode work.  

Also while these larger diodes are out it would be a great idea to find the test points in the LTC LDO chip for the legs of pins TG1 & TG2 because that’s the two signals for firing the MOSFET gates (control). See if the diodes you removed are connected to them and function like D3 & D4 in the data sheet diagram (purely by position and direct trace connection, I mean).  

That will tell you if that also affects the “cleanliness” of the MOSFET control signal(s).  Which is what you’re after! If you hit diodes of another form factor that would be good to know to find more diodes. 

None of this is a guarantee but based on my prior experience SGI use this same style of converter all over the place. And you can see how some of the diode‘s share the same trace as the drain on the MOSFET body or part of the gate. For the drain, it often has to do with the cleanliness of trying to eliminate any kind of fly back or back spike. On the gate control it has to do with the cleanliness of the on off signal.  

If you have the ability to easily pluck the diodes out it’s actually very quick work. The biggest issue is simply recording the orientations and the writing on the diode’s because their orientations may be opposite each other in each pair.  Then with the diodes out check a few other things now that you’ve disconnected a lot of reverse grounding and multimeter checks should be a lot more clean. Then just order yourself a bunch of the diode‘s and install as you saw them before. Again this is the reason I got real hot tweezers to do this work easily. But installation is much easier than removal.  Simply use a small amount of flux and do not put any solder on the pads after you fully clean them and put the new diode down and simply tack one and at a time until the diode is fixed in place then go back over each side to make it look pretty. Pretty easy for SMD work. 

I won’t be getting to this for a while because I need to finish off my power supplies. Without good Tezro power I’m not going to touch the board because I think the majority of these problems are attributed to the failing power supply to begin with.

In response to the question in your head about whether the voltage is actually going low or is this an issue with damaged DS1780 ICs, Elf claimed he manually attached a probe to the VRM and eventually saw the bad readings himself.  I believe the monitoring chips are OK at this time because they showed no other symptoms that I recall. I’m not against replacing them but at this point I have not seen a monitoring chip that wasn’t either totally shorted or had a permanent offset due to a change in internal resistance from damage. That is to say I’ve never seen one have a correct value and then just skew into incorrect versus the value was always offset to begin with even if it moves up and down in value. 

While I don’t know if the same thread on SGUG, Elf also talked about the fact that when he used a thermal imager on the circuit he saw the mosfet for the low side being “stuck on”.  

I responded with the observation that a MOSFET doesn’t stick unless it’s shorted and it definitely wasn’t shorted so it was being commanded to stay on. Whether this command is from the actual LDO or from a bad diode in the signal path is my theory. But at least it was a stated fact that one of the MOSFETs was being run in the on position more than it should and because he found it to be the low position that would definitely describe a low-voltage problem. That is the rail was being pulled to ground too much and therefore the final output voltage was itself low.  

Thanks

[jwhat]

Hi Weblacky,

thanks and here is more zoomed in view of marked areas, for your future reference:



This is from one of my smashed in shipping boards.

Cheers from Oz,

jwhat/John