Repair PSU of IBM X3650 M3 server: Item# FS7023-030G

[Bicurico]

Hi,

My latest addiction is setting up a small "server farm": I bought three IBM X3650 M3 servers. This is a temporary break from TEA (Test Equipment Addiction)... 

These servers are pretty cool and can be bought very cheaply. The interesting thing is that they can host one or two Intel Xeon CPU's, the most powerfull ones are X5650, X5660, X5670, X5680 and X5690. These are hexacore processors with Hyperthreading, so you can end up with 24 cores on one server! This is a lot of calculation power, if you master paralelization.

Anyway, one of my servers has an issue: when the server is in standby (turned off), the two redundant PSU's will turn their internal fans ON. This should not happen and on my other servers the fans are OFF, when the servers are on standby.

Naturally I swapped the PSU's and yes, the issues is with the PSU's and not some obscure BIOS/IMM setting.

So I opened one PSU and found the mandatory thermistor (?) attached to the heatsink.

My guess is that the thermistor is broken, which is why the fans are switched permanently on.

Now the question is: how to I know which replacement part I should use?

If I change it with just any thermistor, its values will be different and as such the behaviour of the fans, right?

I use the servers at home and they need to be shut down at night, due to the noise and power consumption. Having the fans of the PSU's turn ON when the server is OFF is not acceptable...

Any ideas?

(Yes, I already bought two replacement PSU's cheaply, but I would like to repair these two, anyway.)

Regards,
Vitor

[DMM123]

You know, I've asked about a thermistor in another thread.  And I got the same "no response" too.  I guess the question is either too simple or too hard ......

[Bicurico]

I think there is much more to PSU repairs than what one could assume.

I never got any answer and I ended up buying two PSU's at eBay, which fixed the issue. Fortunately, in this case (X3650M3) there are plenty of second hand PSU's on the market at reasonable prices.

But I am currently sitting on 3 HP PSU's from 2 otherwise working spectrum analyzers (HP 8594E). They are broken and I have spent lots and lots of hours studying the schematics and trying to replace and/or testing all the usual failing parts in a switching PSU. With no success.

And here there are thermistors, too. And I have no idea how to replace them or what the sensitivity/precision is to be expected. I measured the sensor on the cooler of the MOSFET on all three of them (with PSU disconected). All three measure different resistance. No idea if this is within tolerance or not.

What I learned is this. PSU repair is easy when you have to replace:

- Exploded diodes (replace all of them)
- Exploded/dried/leaking electrolytic capacitor (replace the ones that look good, too)
- When diodes or switching pulse controlers go bad, change the MOSFET's, too
- When the fuse is gone, normally the diodes are as well
- The optocoupler can go bad and should be tested
- Watch for any burned resistor or trace and replace/fix accordingly

If these steps don't fix the PSU, then repair will be in a totally different league. You will need to have schematics or be able to make them yourself. You need to know what each component does and what equivalent replacement there is. And you need to know how to carry out measurements while having the PSU connected to live power, which is very dangerous (up to 1000V may be present).

Conclusion and coming back to your question: you have a bad thermistor. OK. But if you don't know the specs, nobody will be able to help you in replacing it, as each thermistor has its own curve (Temperature vs Ohm) and the remaining circuit is tuned to that curve...

I think this is  the reason there is no help on how to replace it...

Regards,
Vitor

[Dacke]

What I learned is this. PSU repair is easy when you have to replace:

- Exploded diodes (replace all of them)
- Exploded/dried/leaking electrolytic capacitor (replace the ones that look good, too)
- When diodes or switching pulse controlers go bad, change the MOSFET's, too
- When the fuse is gone, normally the diodes are as well
- The optocoupler can go bad and should be tested
- Watch for any burned resistor or trace and replace/fix accordingly

If these steps don't fix the PSU, then repair will be in a totally different league.

Another really common issue with SMPSs would be cracked solder joints,  which can be quite difficult to see.  I've repaired quite a few at this point,  and this is recurring problem,  especially with components that are mounted on heat sinks and perpendicular daughter boards with control circuitry.

[Bicurico]

You are right - I forgot to mention that failure. Actually I had this once.

But the three HP PSU's: I resoldered every single point with no luck at all on any of the three...

I guess PSU's can be a nightmare when resistors/capacitors start drifting away from the nominal value they should hold. Not like one resistor is off by 50% - imagine 5 resistors, each 10% off... The sum makes some IC decide to shut down the PSU for no aparent reason. Measuring the components will not help as you won't be able to distinguish if a value is reasonably OK or not.

Thermistors seem, in my point of view, especially unpredictable. Imagine one measures 10 Ohm, while the same on a different PSU measures 14 Ohm. What does that mean?

1) Nothing - normal fluctuation
2) One is broken - which one?

And looking at the thermistor, how should you know what a compatible replacement to use?

My decision right now, concerning second hand test equipment is:

1) Never buy a vintage test equipment with broken PSU
2) Never buy a vintage test equipment with working PSU, for which no schematics are available and no replacement PSU's are being sold
3) Consider points 1 and 2 for newer test equipment, unless the price is so low, that you could sell individual parts on ebay and have a return of investment

Regards,
Vitor

[Dacke]

I don't think I've come across a faulty heatsink attached thermistor before in the supplies I've repaired,  but maybe I just haven't seen it yet.  If I was in a situation where I suspected the thermistor and it had no markings on it and I had no known good one to compare it against,  I would look first to see if it is connected to a control IC that's monitoring it.  If it is,  I would look up the datasheet for that IC in the hopes to find out what it expects to see there with regards to the thermistors beta value (a given resistance at a given temperature).  I would then have something to go on.

[james_s]

Server PSUs are extremely dense and virtually undocumented making repair a big pain. Add to that the abundance of good used ones and it's not too surprising there are not a lot of repair resources.

[DMM123]

I never got any answer and I ended up buying two PSU's at eBay, which fixed the issue. Fortunately, in this case (X3650M3) there are plenty of second hand PSU's on the market at reasonable prices.

Conclusion and coming back to your question: you have a bad thermistor. OK. But if you don't know the specs, nobody will be able to help you in replacing it, as each thermistor has its own curve (Temperature vs Ohm) and the remaining circuit is tuned to that curve...

I think this is  the reason there is no help on how to replace it...

Regards,
Vitor

I agree.  I was hoping someone had replaced the thermistor and found a substitute.  I contacted Fluke and they say they have no idea, which seems a little strange since they appear to still be building this BP120 battery packs.  Here's what they said:

"Your question has come back to my team. 

I’m not sure who suggested the product manager, but there are two problems: there is no product manager for the Fluke 123 since it was retired in 2016, and product manager is a marketing role rather than a technical one.

I looked at several versions of the Service Manual and there is no additional information on the battery pack. The battery pack itself (the BP120MH) is a replaceable item and is not considered repairable."

Regards,

Steve *****
Technical Support Engineer
Fluke Industrial "

So, if the manufacturer can not or will not supply the info, it's not available without buying a Fluke battery pack at over a hundred dollars or more.

I've taken the old thermistor and recorded over 40 readings at different temperatures after the meter and the thermistor stabilize.  I have also found out that thermistors degrade over time and lose tolerance and so this old thermistor is not a good choice to model a new one on.

I'm going to contact different thermistor manufacturers and get their help on this.  Time to design this from the ground up.

Honestly, it would be cheaper (timewise) to buy the Fluke battery, but now I have been "challenged" - so I'm rising to the occasion!