RetroSpector78 wrote on 2024-02-20, 11:17:

I first thought this created a dead short and fried the mostfet. But then I thought that this is the equivalent of having the mosfet turned on and should in fact not harm the mosfet itself.
So am I wrong in saying that this short was simulating the transistor being one for a prolonged period ? If my understanding of VRMs is correct this would have caused the voltage to rise at the end of the coil reaching the cpu and potentially frying that.

Depends if its the low or high side mosfet, but generally yes. You fried one of
-cpu
-both high and low side mosfets but only replaced one
-mosfet driver/VRM controller

Apologies for the stupid follow up questions. Not really familiar with all of this stuff.

So I guess the high side mosfet is the one hooked up to the 12V rail (drain) , switching on/off at a high pace to apply a voltage to the inductor, allowing the inductor to output a suitable voltage to the CPU.
That's the one that got shorted by the multi-meter.

But if a shorted drain -source means the mosfet is effectively switched on (a "normal" state for a mosfet) , how would that destroy it ?

Would the other NEC chip be the low side mosfet ? Or can there be multiple high side mosfets ? (howto check)

mosfet driver/VRM controller = the PWM module generated the PWM to the mosfet ?

RetroSpector78 wrote on 2024-02-20, 13:51:

But if a shorted drain -source means the mosfet is effectively switched on (a "normal" state for a mosfet) , how would that destroy it ?

A short like that would not destroy the MOSFET you bridged with the meter probe, but the other one.

First of all I don't like calling one of these a "low side" because I don't see it as such. But I guess the name stuck. In fact this transistor could be unpowered, as long as the internal protection diode is still OK the circuit would work, but with lower efficiency. After all this is a pretty standard step-down converter (also called buck converter).

If you short D-S of the low side transistor then the high side (the main switching transistor) will pretty much see a short on its output. This will usually kill it, but the question is how. If it fails open then just replacing it should bring the circuit back to life (and in general you don't expect anything else to be affected).

If it fails short, or if you shorted the high side yourself with the probe, then it will kill the low side one - when it turns on it will pull all the current from the PSU. Eventually the PSU will trip or some trace will burn. But before any of that happens the permanently turned on high side will provide non-limited current for the choke, which will eventually pass it to the load. Frying the load most likely.

You didn't specify what exactly happened to the desoldered transisor - but note that it can happen that it will fail short and then the current surge (since the resistance of the shorted junction is never zero) might blow it open. But I would expect the CPU to be damaged now. Also note that if you replaced the high side MOSFET but the low side failed open, along with the protection diode (say, the interal gold wires detached rather than D-S junction blowing up), and you didn't replace this one as well... you probably killed the CPU the moment you powered the mobo after the repair. It depends how good the VRM chip is, can it get down to 0% PWM or not.

All in all you need to desolder and test both transistors, preferably replace both since they can't be trusted now. Then try the mobo with current CPU. If the CPU is now dead it could be shorted - this will either blow the MOSFETs again or trip PSU or mobo (if the VRM is smart enough) into shutdown. If the CPU is dead but not shorted this should at least let you verify the new MOSFETs work and regulate the voltage properly. Either way do not put a new CPU in, unless you feel very confident the transistors and VRM chip are working properly.

Another thing that can get damaged by passing 12V other than CPU and chipset are all downstream capacitors.
I can only find revision 7 diagram. Its cost optimized, simplified Cpu VRM is on page 30.
First thing I would do is lift L23 L24, put CPU back in and check resistance on CPU_Vcore - this usually can tell you if CPU is fried.
Then I would check the other Vcore "phase", in general desolder all Q3 4 5 6 7 etc and check those individually

Thx everyone for the responses. much appreciated.

So if I understand correctly :

• You need a load to verify the VRM (without a load I cannot measure anything).
• If the load is a shorted CPU it might cause damage to other components.
• If the load is a working CPU the CPU might be damaged due to the VRM.

That sounds pretty risky. (if I replace the 2 mosfets perhaps other failed components can kill them again).

Is there a minimum ohm that a cpu should have between vcc / vss ? I think mine has about 10 ohm.
Is it possible to put another load between the VCC / VSS pins of the cpu to test the VRM ?
How does the VRM know what voltage it needs to supply to the CPU ? Is there some kind of handhake going on, or is there a voltage reference pin ?

rasz_pl wrote on 2024-02-20, 17:35:

Another thing that can get damaged by passing 12V other than CPU and chipset are all downstream capacitors. I can only find revi […]
Show full quote

Another thing that can get damaged by passing 12V other than CPU and chipset are all downstream capacitors.
I can only find revision 7 diagram. Its cost optimized, simplified Cpu VRM is on page 30.
First thing I would do is lift L23 L24, put CPU back in and check resistance on CPU_Vcore - this usually can tell you if CPU is fried.
Then I would check the other Vcore "phase", in general desolder all Q3 4 5 6 7 etc and check those individually

Thx I'll take a look at those diagrams and take your advise. Did not know these were available somewhere.

Always a shame to have a motherboard / cpu die on you. but it did came in a box full of other motherboards / cpu that were half-working and was about to be scrapped. Will hopefully try to learn something from it.

You could probably load the various outputs (phases) with something like 3.3 ohm 5W or 10W power resistors, these mobos didn't really "talk" to the CPU like the current ones do. A dummy load will work. I'd probably start with the suspect CPU though, unless the MOSFETs turn out to be either expensive or hard to get. If a second pair blows with this CPU after all (because it's shorted and PSU didn't trip in time, and the VRM chip is too dumb to notice) then at least you know the CPU is properly toast. Otherwise it'll either work or make a decent dummy load.

As for the voltage regulation, can't remember how exactly Athlons did it. Possibly there is a Vref coming from the CPU? Everything up to 386 was 5V, then some 486 were 3V3, then Pentiums (apart from 60/66) were also 3V3 but MMX ones required further reduction in core voltage. P2 and early P3 were 2.05V and so far it was all static, at worst set via jumpers. From there you had some form of CPU output to the mobo that set the required core voltage.

from p2 onward it was VID pins on cpu all the way to some late i3/5/7 chips
there wont be any voltage without cpu in socket

rasz_pl wrote on 2024-02-20, 23:43:

from p2 onward it was VID pins on cpu all the way to some late i3/5/7 chips
there wont be any voltage without cpu in socket

How does that work? I'm an amateur, but looking at the VRM 8.4 spec, it looks like if your VID pins are all pulled to ground, it should provide a 2.05V core. Am I thinking about this wrong?

all VID pins are pulled high with no cpu, all 1 is undefined state for VRM