|
Top-level
 same symptoms as before. it just...stopped working, and when i push the power button, the fans bump a little bit but that is it. i checked the POL power supply that was causing problems last time, but the issue is that it is getting only 4.5V instead of 12V. 53 comments
huh, it's a bit simpler inside than what i expected. there's only these two circuit boards. a few large parts, but where is the controller? it's on the back of one of the boards! it's a M51995AFP chip. time to do some reading. and maybe Lambda has a service manual or schematic. maybe. well, i've reverse engineered the dang thing. there are a bunch of component values missing but you can check it out here: https://github.com/schlae/sws600-12  @tubetime I'd like the illogical decision to inflict Microsoft Windows on these boxes analyzed. 🤷 🤔 there's a soft short on the 12V line going into the motherboard. this was preventing the big power supply from going power good. @tubetime Hi, I've not heard the term soft short before. Is it a short with a bit of resistance? looks like one or more MOSFETS in the big multiphase core regulator have failed in a short circuit. this is a very bad sign. @tubetime I have, in the past, more than once, lazily used a 18650 cell to find the offending MOSFET, removed it, and kept on running the machine minus that phase. @GeorgeRudolf strangely enough, the motherboard seems to work without the MOSFET. i'll still replace it though. @tubetime I used to live somewhere I didn't have the luxury of ordering from Mouser or DigiKey, so if my scrap pile didn't have it, it'd have to make-do on n-1 phases. But yeah. Make sure there's no gate drivers that could've gotten mangled...eh I'm preaching to the choir here :) @tubetime yeah normally it's source to drain, but I've seen gate driver/controller Ic failures before so thought it worth mentioning anyway. bad FET. it looks fine but it's not. unfortunately this means that the CPU was exposed to 5V on a voltage rail meant for something much less. the output filter caps have a 4V rating. @tubetime https://www.eevblog.com/forum/testgear/agilentkeysight-infiniium-mso9000-series-motherboard-repairupgrade/ Do these logic analyzers also run PCI-E over a SATA cable to the acquisition board? @humulus regular PCI to an interface card, and the card has a ribbon cable going to the backplane. so I pulled the MOSFET and tried to power it up, and it works!? I'll probably still find a replacement MOSFET to avoid excess stress on the other phases. @tubetime did you need to swap the cpu, or did it survive the overvoltage? i'm really hoping that a simple FET replacement will get this thing going reliably. the motherboard is technically off the shelf (Adlink M-880-Nitro) but people want lots of money for them. @tubetime seems a wonder it didn't die with that mosfet but maybe CPUs have features against mosfet latch up or the pmic detects it and shuts off the MOSFETs arrived. who knows if this will work... well it boots but now the frame isn't detected. so there's something wrong with the PCI card, the cable, or the backplane itself. BTW, this is why the CPU didn't fry--the multiphase buck controller (the ADP3168) has crowbar protection in case of a shorted high-side MOSFET! so the sequence of events in this case went: it would be nice to have a second probe cable. this is the less common 60-pin cable, not the stubbier 50-pin one used on the 1650 series. they're super hard to find for some reason. thinking about the failure a bit more. here's the precise cause with a nice diagram. a high-side MOSFET failed in a short circuit, which would put 12V on the sensitive CPU core rail. however, the controller chip detected the overvoltage and turned on all the low side FETs (circled in green).  @tubetime Do you know whe there are only three high-side FETs but six on the low side?   @tubetime Have you considered upgrading the CPU? Looks like there are a few that are compatible with the socket and wouldn't increase the TDP.  |
Gregory's Server
umm, so it looks like this enormous 600 watt 12v power supply doesn't work.