after CPU VRM burned, CPU still good?

[Stefan Payne]

I've got a Gigabyte X79-UD5 with burned VRMs, looks like this.

How probable is it that the CPU survived (or not)??

[tom66]

Re: after CPU VRM burned, CPU still good?

If on the high side, marginal. CPU would have been subject to +12V for at least a millisecond or so which would probably destroy it. If on the low side it might survive.

[Stefan Payne]

Re: after CPU VRM burned, CPU still good?

Well, it looks like that was the high side...

The connector on the upper right corner is the EPS12V CPU connector, followed by a coil.
So that burned one looks like it was on the high side...

[kaboom]

Re: after CPU VRM burned, CPU still good?

Those CPU PWM ICs turn the low side on if/when the output exceeds the VID value.

Unless the low side completely opened up/burned clear and was unable to shunt the incoming supply, the CPU may actually be good.

If on the high side, marginal. CPU would have been subject to +12V for at least a millisecond or so which would probably destroy it. If on the low side it might survive.

What error amp, especially one for a CPU regulator, is so slow and useless?

It works much faster than that... Besides, the output caps have to follow the increasing voltage, so the change is not instantaneous. It might seem like that to us, but there's plenty of time for the error amp to back down the duty cycle, then ultimately keep the low side devices on.

(The PWM IC drives the MOSFETs via dedicated hi/low gate drivers, but the effect is the same.)

[tom66]

Re: after CPU VRM burned, CPU still good?

If the top FET is shorted, there is nothing the error loop can do. You will basically form a resistive divider with about 10mohm on each ladder. Output voltage will be briefly determined by the ratio of resistances. Most controller ICs however will rely on the body diode to commute an over-voltage event. Very few of them switch on the bottom FET during transients because it makes the control loop trickier to design.

A big "it depends", really....

[kaboom]

Re: after CPU VRM burned, CPU still good?

Output Crowbar
As part of the protection for the load and output components of
the supply, the PWM outputs will be driven low (turning on the
low side MOSFETs) when the output voltage exceeds the upper
Power Good threshold. This crowbar action will stop once the
output voltage has fallen below the release threshold of approxi-
mately 450 mV.
Turning on the low side MOSFETs pulls down the output as the
reverse current builds up in the inductors. If the output overvolt-
age is due to a short of the high side MOSFET, this action will
current limit the input supply or blow its fuse, protecting the
microprocessor from destruction.

...

[tom66]

Re: after CPU VRM burned, CPU still good?

As I said, this is a best case scenario. You've still got a large amount of bulk capacitance (PSU, probably at least 2200uF, plus motherboard and others) that will dump a lot of energy into the processor through the upper FET. And, you're assuming that the IRF devices (controller included) implement this function. As I said many controllers don't implement it because it complicates the control loop during transient response.

[Stefan Payne]

Re: after CPU VRM burned, CPU still good?

Well, contacted Intel and they say, they're able to exchange my CPU. So it doesn't matter if the CPU is still good or not...

[momaka]

Re: after CPU VRM burned, CPU still good?

Those CPU PWM ICs turn the low side on if/when the output exceeds the VID value.

Unless the low side completely opened up/burned clear and was unable to shunt the incoming supply, the CPU may actually be good.

+1

I have an Intel DG33 that shorted it's upper MOSFET twice on me (after working for a few minutes each time). Both cases resulted in PSU buzz (i.e. short-circuit protection kicking in). After repairing those, I had a lower MOSFET short, and again the result was the same - PSU short-circuit protection kicking in. I though, man, that CPU sure must be toast now. So I decided to play with the board a little. I swapped the inductors with ones from a completely different motherboard (4-turn larger core in place of 3-turn small core), changed all high-side MOSFETs with different (but matching) ones. When I tried to turn on the motherboard, I got a nasty squeal. Then, I decided, screw it, let's see if this possibly fried CPU can damage another motherboard. So I took the CPU and tried it in another motherboard I didn't care much about. But what do you know? The CPU (Core 2 Duo E2160) was perfectly fine and functional, even after all of that BS.
As for my Intel mobo, I can't really say the same . All I know is, all gate drivers are fine. It just eats MOSFETs or squals or shorts out the PSU .

This is not my only experience with a chip that survived a shorted upper MOSFET. I also have an AGP GeForce 7600 GS that went through a bad MOSFET. Still works with my replacement 85 Amp MOSFET (vs. 13A for the original one ).

In other news, I have a Shuttle PC with (likely) bad NB. But that one was due to faulty OST caps. NB had only two caps on its VRM low side (large OST RLX and small OST RLG). Both bulged and failed, thus causing the upper MOSFET in the buck converter to die. With no (or very little) output capacitance from those dead caps, NOW you have recipe for disaster.

It works much faster than that... Besides, the output caps have to follow the increasing voltage, so the change is not instantaneous. It might seem like that to us, but there's plenty of time for the error amp to back down the duty cycle, then ultimately keep the low side devices on.

Yup.
I tried messing with some dead Xbox 360 boards (badly warped due to fly-by "repair" shops working on them). It doesn't matter if you short 12V to GPU V_core or V_core to ground - both will cause an over-current condition and trip the PSU.

So most likely the OP's CPU is fine. But I guess that is a moot point now.

In any case, CPUs are some tough cookies.

[Stefan Payne]

Re: after CPU VRM burned, CPU still good?

Yeah, got my replacement CPU from intel, was replaced and took about a week (and no cost for me).

The LGA2011 sys runs fine now with no real OC (well, I upped the turbo multi a little )

[kc8adu]

Re: after CPU VRM burned, CPU still good?

xbox 360 power supplies have a touchy overcurrent protection.
the input cap charging on my hf radio will trip it if you plug it in live.
but it does not seem to mind 25a peaks demanded on voice peaks.
theres a 3300uf cap on the dc input on this yaesu ft-847.
and i too have seen vrm's totally burned up and the cpu survived.
its a matter of luck.

I tried messing with some dead Xbox 360 boards (badly warped due to fly-by "repair" shops working on them). It doesn't matter if you short 12V to GPU V_core or V_core to ground - both will cause an over-current condition and trip the PSU.

[kaboom]

Re: after CPU VRM burned, CPU still good?

xbox 360 power supplies have a touchy overcurrent protection.
the input cap charging on my hf radio will trip it if you plug it in live. (1)
but it does not seem to mind 25a peaks demanded on voice peaks.
theres a 3300uf cap on the dc input on this yaesu ft-847.
and i too have seen vrm's totally burned up and the cpu survived.
its a matter of luck. (2)

1- Most hard drives are the same way. Hotplug anything in parallel with one, and it will reset.

2- Maybe they just adhere the old power supply adage of "protect the expen$ive load, no matter what."

I have an Intel DG33 that shorted it's upper MOSFET twice on me (after working for a few minutes each time). Both cases resulted in PSU buzz (i.e. short-circuit protection kicking in). After repairing those, I had a lower MOSFET short, and again the result was the same - PSU short-circuit protection kicking in. I though, man, that CPU sure must be toast now. So I decided to play with the board a little. I swapped the inductors with ones from a completely different motherboard (4-turn larger core in place of 3-turn small core), changed all high-side MOSFETs with different (but matching) ones. When I tried to turn on the motherboard, I got a nasty squeal. Then, I decided, screw it, let's see if this possibly fried CPU can damage another motherboard. So I took the CPU and tried it in another motherboard I didn't care much about. But what do you know? The CPU (Core 2 Duo E2160) was perfectly fine and functional, even after all of that BS.
As for my Intel mobo, I can't really say the same . All I know is, all gate drivers are fine. It just eats MOSFETs or squals or shorts out the PSU .

Some of those substitute devices may not be optimal. While there are dedicated gate drivers, you've got to watch gate charge, gate capacitance, and feedthrough capacitance. If any trash couples via D-G capacitance, the drive waveform could get distorted, depending on the layout and capabilities of the driver IC(s).

What were the existing ones, and what did you use to replace them with?

Do you think one of the low side MOSFETs was leaky, and taking out the high side ones? Just because I'm curious, was it also squealing w/o the CPU?

This is not my only experience with a chip that survived a shorted upper MOSFET. I also have an AGP GeForce 7600 GS that went through a bad MOSFET. Still works with my replacement 85 Amp MOSFET (vs. 13A for the original one ).

This is good to know! I've often wondered about video cards implementing "protect the load no matter what." Of course, all cards don't use the same PWM controllers, but at least for you, all turned out OK.

[tom66]

Re: after CPU VRM burned, CPU still good?

I honestly doubt most manufacturers care to be honest - it's simply a case of getting lucky. Why would they care about saving the CPU, when the VRM burns, the average guy is just going to throw away their console. Sad, but that's the way it is. Certainly, this is how the company I work for does it. Our video decode CPUs run at ~0.85V, from a 12V VRM, one test involves simulating the upper FET shorting, the CPU dies every time, and that's fine, all we care about is that the FET doesn't melt or distort the plastic casing/cause a fire etc. (We're only talking about a 12V, 5A plug-top in most cases so it's not a major concern.)

[kaboom]

Re: after CPU VRM burned, CPU still good?

This is good to know! I've often wondered about video cards implementing "protect the load no matter what." Of course, all cards don't use the same PWM controllers, but at least for you, all turned out OK.

This only assumes the PWM uses synchronous rectification. If the upper MOSFET shorts, and the lowside is a plain rectifier diode, the load sees gross overvoltage.

[momaka]

Re: after CPU VRM burned, CPU still good?

Some of those substitute devices may not be optimal. While there are dedicated gate drivers, you've got to watch gate charge, gate capacitance, and feedthrough capacitance. If any trash couples via D-G capacitance, the drive waveform could get distorted, depending on the layout and capabilities of the driver IC(s).

What were the existing ones, and what did you use to replace them with?

Original MOSFETs: FDD8780 and FDD8796 for upper and lower MOSFETs respectively. VRM is 3-phase, with 1 upper and 2 lower MOSFETs per phase (so 3 and 6 total).

I used 85N02 from ON-semi (Xbox 360 pulls). Yes, the 85N02 have considerably higher capacitance, but I figured it shouldn't be a problem, since the drivers are ADP3120A - which are similar to what the Xbox 360 uses sometimes: ADP3118 (at least I don't see much difference between the two). And those 85N02, along with 60N02 are often used in Xbox 360 VRMs.
The PCB around the MOSFETs on that board did look a bit overheated/dark. The copper pads on which the MOSFETs were soldered on were quite tiny compared to what I have seen on other motherboards, so I though maybe that was the problem, and the 85N02 would do better with their lower Rds_on resistance compared to the originals.

Do you think one of the low side MOSFETs was leaky, and taking out the high side ones?

It is entirely possible, granted I had to change two upper MOSFETs before a lower one blew.

Just because I'm curious, was it also squealing w/o the CPU?

I can't tell you, because the board is designed so that it won't turn ON if there is no CPU in the socket. But now that I mention that, I didn't think of trying to short the PSU's PS_ON wire to ground to see if I can force-start it. So, I might just try that next (when I get home from college, that is... bleh).

But with the CPU in there and my replaced MOSFETs, *and* the coil for the VRM phase that had the shorted lower MOSFET removed, the CPU VRM voltage does come up to about 0.6-0.8V - albeit with a lot of squealing.

I honestly doubt most manufacturers care to be honest - it's simply a case of getting lucky. Why would they care about saving the CPU, when the VRM burns, the average guy is just going to throw away their console.

I suspect it's not the manufacturers of the board that care, but rather the manufacturers of the PWM controller and/or drivers. They probably try to add those fail-safe features for marketing purposes. Whether they are implemented improperly (or messed up unintentionally) in the design by the board manufacturer is another story.

[ben7]

Re: after CPU VRM burned, CPU still good?

As I said, this is a best case scenario. You've still got a large amount of bulk capacitance (PSU, probably at least 2200uF, plus motherboard and others) that will dump a lot of energy into the processor through the upper FET. And, you're assuming that the IRF devices (controller included) implement this function. As I said many controllers don't implement it because it complicates the control loop during transient response.

How about the capacitance on the output of the VRM? It takes time to charge that through a resistance and an inductance.

[Stefan Payne]

Re: after CPU VRM burned, CPU still good?

I removed the defective MOSFET and the Board kinda works again.

Do you think it's safe to use or not?

[kaboom]

Re: after CPU VRM burned, CPU still good?

I removed the defective MOSFET and the Board kinda works again.

Do you think it's safe to use or not?

That was the bad one on the high side, right? I'd x-ref a sub and replace it if that's the case. That regulator was compensated for its original complement of phases. Remove a leg and you'll suffer poor ripple and step-load response, as well as worse ripple performance overall with fewer phases. (see link)

It could be enough to make your CPU cranky- a bit like a marginal O'C.

I don't know what you mean by "the board kinda works." I presume this means POSTs, but not stable enough to finish booting. Or perhaps crashes once device drivers come up to speed, and start "pushing" the system.

See Fig.4 on page 4- for a given duty cycle, ripple is worse as n decreases, with higher phase count giving smaller ripple-peaks.
https://cdn.badcaps-static.com/pdfs/...398f5d4066.pdf

If you can at least boot to a Memtest CD, replace that one MOSFET! Don't condemn the CPU or MB 'til you "feed the beast" properly.

If you scope VCore with that MOSFET missing, be aware that it will look nasty until all phases are working correctly.

Good luck...

[Stefan Payne]

Re: after CPU VRM burned, CPU still good?

I don't know what you mean by "the board kinda works." I presume this means POSTs, but not stable enough to finish booting. Or perhaps crashes once device drivers come up to speed, and start "pushing" the system.

No, I just haven't tested much.
Mostly because I don't have another heatsink for a LGA2011 CPU...

I thought I should ask here before trying to do anything else..

As for the VRM thing:
I kinda ripped the traces off the PCB, so soldering a replacement on it could be a little bit challenging (but not impossible).

Seems like I shouldn't touch that board until I'm able to afford a scope and scope it, right?

As for the board:
it's this one and advertised to have 12 phases...

[kaboom]

Re: after CPU VRM burned, CPU still good?

No, I just haven't tested much.
Mostly because I don't have another heatsink for a LGA2011 CPU...

Got any spare/junk 478/775, or other heatsinks you could either cut down or hold in place? All you want to test is POSTing, and maybe a Memtest CD. If you could clamp a substitute heatsink down with zipties thru the holes near the socket, it might get you by for testing.

I thought I should ask here before trying to do anything else..

As for the VRM thing:
I kinda ripped the traces off the PCB, so soldering a replacement on it could be a little bit challenging (but not impossible).

Seems like I shouldn't touch that board until I'm able to afford a scope and scope it, right?

If there's any question about step-load overshoot, you might consider going into the BIOS setup and lowering VCore several steps below the default VID- again, just for testing and to check viability of the board/CPU. Since there were 12 phases, the output capacitance is rather small- no "huge" capacitors like in other boards. Do you know the TDP and Vcore of the highest-spec CPU that board could take? And if so, what's the TDP and VCore of your current CPU?

Ripple/overshoot may not be a problem if you're far enough away from the maximum demand of that VRM. You're "ripple bumpiness" with varying duty cycle will not be as pronounced as the diagram I referred to earlier; you've still got 11 phases. While there will be a suckout during the missing phase's period, if your CPU doesn't pull VCore too low during this time and the loop remains stable, you just might get away with it.

Remember, there were some weird ASROCK boards that dropped phases as CPU activity decreased. Those didn't work too well and could be unstable, since they were 4-phase designs that actually single-phased at times. As you can imagine (look at Fig.4 diagram for reference), they tended to be less stable at part-load, especially when those Evercons aged- more ripple for a given duty cycle, as N decreases. Supposedly, they were compensated so the loop didn't oscillate, but switching-freq ripple still made it to the CPU. Silly ASROCK!

As for the board:
it's this one and advertised to have 12 phases...

Again, such tiny caps for VCore...

Edit: Nevermind, looks like VCore caps are missing in those pix! Found a -UD3 board with the usual compliment of polies in VCore:
http://www.amazon.com/GIGABYTE-GA-X7.../dp/B0064Z6YD2

If your -UD5 has the same VRM as that -UD3, it'll probably be fine as long as your CPU is 11/12, maybe 10/12, the TDP of top-spec for that board. All those polies in parallel make a nice, low-z supply. Is it perfect? No, but it's what ya have to work with...