XFX XTR 650W power on issue

[MichR]

Hello guys,
I've had this XFX power supply for a long time, and recently it started behaving like one of these Seasonic X units.
TL;DR: I have to flip the on/off switch multiple times in order to startup the power supply (once it starts, it works great). I need to do this when I shut down or put in standby the PC.
From the previously linked thread, I found this helpful advice:

im gonna take another wild guess at the problem of the seasonic x and say that it could be the overvoltage protection tripping and shutting down the psu even on the standby rail. that explains why it wont power on (totally no response as if the psu were dead). also explains why the kung fu trick with the psu switch works in turning the unit back on. flipping the psu switch is usually used for resetting the protection trips by the protection ic.

as for why its tripping, could be like what momaka said. malfunction in the apfc side causing the 5vsb rail to overshoot and causing the protection ic to trip the psu. so due to this strange malfunction in the 5vsb or apfc, the psu cant run with only the 5vsb line active for long. it must be quickly turned on into mobo-on mode when power is applied with all rails running. quickly turning the unit on when mains power is applied somehow masks the problem that is causing the psu to malfunction or trip in mobo-off mode.
[...]

Therefore I decided to take apart the unit and investigate on the issue.
You can find HQ pictures of the boards here.
A first visual inspection did not reveal any defective/bad components. I then tested the single components on the entire board looking for a short/bad behavior, with no success.

Finally, following ChaosLegionnaire's suggestion, I decided to plug it in, connect to an old motherboard and measure the signals with my oscilloscope.

Strangely enough, I can find the following pattern on every output (and also ground!, that's why it is weird) I place my probe to.
These "oscillations" have a duration of about 2us and each one is visible 7.6us after the previous one. The maximum measured voltage is about 0.6V.

Is all of this normal/has it something to do with the aforementioned issue?

[Per Hansson]

Re: XFX XTR 650W power on issue

How are you probing? If you are using the long "antenna ground lead" as I like to call it that's your issue...
Try to probe using the metal spring that comes with the probe (that you attach right at the end).
Watch the following video: https://www.youtube.com/watch?v=Edel3eduRj4

[MichR]

Re: XFX XTR 650W power on issue

Yes, I was using the ground lead provided with the probe. As suggested from the video, by using a much shorter and more direct ground path I was able to remove much of this unwanted e.m. "noise".
I also noticed that this "noise" is generated by the power supply. Indeed, when I shut it off the "noise" vanishes completely.

[Per Hansson]

Re: XFX XTR 650W power on issue

Yes, this is a very common and easy mistake to make.
The EM can also be created by things like dimmers for lights, cheap china LED lights, your dishwasher, etc etc etc etc ad infinitum

This video is also a good thing to see on this subject: https://www.youtube.com/watch?v=BFLZm4LbzQU&t=

[MichR]

Re: XFX XTR 650W power on issue

Thank you for linking that video, it was of a great help.

Coming back to the topic, I was finally able to see the waveforms of the outputs. I did not even bothered to measure the ripple, since it was very low (<50mV) and no overshoots could be seen on startup/shutdown of the power supply.
It must be said that the issue is not showing on my testing setup. If no one has any suggestion, I think I will reassemble the unit, (externally) connect it to my PC and monitor the outputs before and after the issue.

[MichR]

Re: XFX XTR 650W power on issue

It's been a week since I've reassembled the unit. I connected it to my PC, but kept it outside the chassis. During all this time it didn't give me any issues.
So today I decided to put it back inside the chassis and the described issue showed up at its first use. Strangely enough, when I removed it again and reconnected from the outside, the issue persisted. So one of these causes must have "triggered" this behavior:

1. Grounding: I thought the physical contact between the PC and PSU chassis could have been the cause, but they are always electrically connected through the ground cables. I don't think this is what caused the issue.
2. Temperature: outside, the PSU have always been cold because no warm air could go through its vent. This changed when I put it inside. However, the issue showed after around 5 minutes of idling, so maybe this was not the cause either.
3. Being upside-down: maybe, who knows? I am not able to completely "reset" its behavior. Now that the issue has been triggered, it shows up everytime I try to startup the PC.

Anyway, in order to understand what is going on, I hooked my oscilloscope to the PSU.

• 5VSB: always, always, always at fixed 5V, no matter what. OK.
• +3.3V, +5V, +12V and -12V: when I'm able to startup the unit, the voltages go to their nominal values. No overshoot, undershoot, etc.; they seem to work great.
• PWROK: at 0V when shut down and around 5V when the unit starts. OK
• PSON: this requires some more words. When the PSU is off, it's at 0V and it does not change no matter what I do. If I'm able to startup the unit, it is kept to 0V, but once I shut down the PC it goes to around 3.5V, which should be OK. It then stays at this fixed voltage until I press the power button on my PC. Then, it goes to 0V for around 7 seconds, comes back to 3.5V for another 5 seconds and finally it goes to 0V again, and there it stays. During this time, all the power rails as well as PWROK stay at 0V.

From all of this, I do think the OP in the thread I linked in the first post was actually right. It's for sure a problem related to the monitor IC, so my next move (if anyone does not come out with a better suggestion) will be to measure the voltages on the monitor pins, in order to understand what causes that PSON behavior.

[MichR]

Re: XFX XTR 650W power on issue

So I carefully measured and monitored all the pins of the monitor IC, a PS223. Here are the results:

1. PGI: goes to around 2V when I power on the PSU using the flip switch trick. It's 0V if the unit is off or I press the power button on the PC.
2. GND: always 0V.
3. FPO: it's 0V at first. When I power on the PSU using the flip switch trick it stays at 0V. When it's off it goes to 5V.
4. PSON/: it's 0V at first, but when the PSU has been shut down, it goes to 3.5V with a little of overshoot (0.5V). Then if I press the power button on the PC it goes again to 0V and behaves as described in the previous post.
5. IS12A: 12V when on; 0V otherwise.
6. RI: it's 1.2V when on; goes to around 4V with some overshoot when it's off.
7. IS12B: 12V when on; 0V otherwise.
8. VS12B: 12V when on; 0V otherwise.
9. OTP: always 1.25V
10. IS5: 5V when on; 0V otherwise.
11. IS33: 3.3V when on; 0V otherwise.
12. VS12A: 12V when on; 0V otherwise.
13. VS33: 3.3V when on; 0V otherwise.
14. VS5: 5V when on; 0V otherwise.
15. VCC: 12V when on; 5V otherwise.
16. PGO: 5V when on; 0V otherwise.

Here come the important observations:

I noticed that when I press the power button on the PC, PSON goes low (of course), but PGI remains low. PSON remains low until I remove AC power from the PSU. If I now use the flip switch trick to power on the PSU, I can see PGI going high (2V) and the power supply finally starts.
I checked the monitor IC datasheet and found out that when PSON is low, it's all up to PGI to start up the power supply. So the problem definetely lies on PGI.

Following my hypothesis, I decided to redraw part of the schematic which controls the PGI line, which is freely available in my attachment.
After some measures I could get rid of many possible causes of the malfunction. Finally, I found out that when the issue shows up, the +12V line is not able to 'start'. It reaches approx. 1.5V and then it slowly falls back to 0. I will redraw part of +12V schematic and carefully check there for any issue.

[MichR]

Re: XFX XTR 650W power on issue

Ok, I redrew the schematics used by the PSU to provide +12V to the output.
The more complete schematic is available in the attachment.

After some measurements, I finally found out the cause of this problem. Apparently, when the issue shows up, the pin "VINS" of ICE2HS01G is at 1.21V instead of 1.48V when the PSU has turned on. According to the datasheet, this IC requires a VINS voltage greater than VINSON (1.25V), otherwise it turns off and so the power supply.

This voltage drop is due to the current generator inside the ICE chip that is connected on VINS pin in parallel to the resistive voltage divider (R501/R503):

And this is verified by applying the superposition principle: VINS = VINS(Ih=0) + VINS(Vbus=0) = Vbus*R503/(R501+R503) - Ih*(R501//R503) = 1.44 - 0.23 = 1.21 (using Vbus = 7.48, when issue shows up).
According to the datasheet and verified by my calculations, the purpose of this current generator is to add an hysteresis to the comparator.
The datasheet provides two formulae to calculate the resistance of RinsH (R501) and RinsL (R503) when the two voltage levels Vbus(low) and Vbus(high) are known. Inverting these relations I could obtain that
Vbus(low) = 6.476V
Vbus(high) = 7.676V

When the PSU doesn't start, Vbus = 7.48V, so the comparator output is not known, but since VINS = 1.21V, output must be 0: the ICE is off and so the PSU. This tells me that Vbus voltage must have been lower (for a short period of time) than Vbus(low).

Actually, in the previous calculations using the superposition principle, I assumed Vbus was connected to a voltage generator. This is not really my case, since the real voltage generator (on the primary cap) is connected to Vbus through Q302. If I assume Q302 as a short and calculate the voltage on Vbus with and without the current generator Ih, I get
Vbus(Ih = 0) = 7.706V
Vbus(Ih /= 0) = 7.515V
which may confirm that the current generator is decreasing Vbus voltage, causing the ICE chip to turn off.

What's more, if I calculate them when the power supply is off (voltage on primary cap = 303V) I get:
Vbus(Ih = 0) = 6.526V
Vbus(Ih /= 0) = 6.335V
which really approach Vbus(low) voltage, and may cause the IC to shut down.

To sum up, my theory is that when I turn off the power supply, the voltage on the primary cap (303V) causes the ICE chip to shut down and the current generator to be connected. The latter causes a decrease of Vbus voltage which cannot be enough to turn the ICE chip on (higher than Vbus(high)) even when the primary cap is at its maximum voltage of 357.8V.


Also, after some other measures I noticed voltage decreases on several points around VINS area. VCC is one of these and, as shown in the schematics, is generated by the standby IC, an ICE2QR4765. Again, I measured the voltage on its pins, but couldn't figure out if there's something wrong with that.


Do you think the VCC voltage has something to do with Vbus and VINS?
If not, I think I know how I could solve this, but I would like to know your opinion first.
Thanks

[momaka]

Re: XFX XTR 650W power on issue

Wow, nice work so far.

I think this might actually be over my head to give you any accurate advice (though I also have to say I only read the thread quickly and in not great enough detail due to limited time today). In any case, what if you provide Vcc and/or Vins with an external isolated power supply. I think this might be able to confirm/bust your theory quickly. (But again, just make sure you are using an *isolated* power supply, as any regular grounded one will give you plenty of fireworks when connected to the primary side. I'm sure you know that, but I just have to mentioned it in case anyone else reads these troubleshooting "tricks".)

[MichR]

Re: XFX XTR 650W power on issue

Thank you, momaka.

So I decided to test my 'solution', which basically decreases the 'on' threshold voltage Vth(on) to be a little lower than the minimum Vbus voltage (which I have when the current generator inside the ICE chip is connected).
Requesting Vbus[min]>Vth(on) gives a numerical solution of R501 <(about) 120k.
I therefore placed a 100k resistor in place of the original 120k of R501. This, assuming 357.8V on the primary cap, gives the following theoritical voltages:
Vbus[max] = 6.983V
Vbus[min] = 6.783V
Vth(off) = 5.605V
Vth(on) = 6.605V
And as you can see Vbus[min] > Vth(on).

I've tested the PSU throughout two full days so far, and it seems to run good. I haven't tried to force VINS with an external power supply, but that must have been the problem. I'm actually pretty sure it was because even connecting a voltmeter (which has 10MOhm of resistance) caused the ICE chip (and PSU) to shutdown (the voltmeter's resistance decreased the total resistance to ground, and so the voltage on VINS pin).

[Per Hansson]

Re: XFX XTR 650W power on issue

Excellent work! It's always nice to play with stuff like this and learn a thing or two in the process

[momaka]

Re: XFX XTR 650W power on issue

Glad to hear the issue has been solved.
Keep us posted if you find anything else. I think there may be a good few people out there who still PSUs based on this platform and may find this thread very useful, particularly if they have the same issue.

[andmost]

Re: XFX XTR 650W power on issue

Hello, I would need your help. My XFX XTR 750W doesn't turn on. I tested before to open it, but nothing. Now I'm verifying the board but I would need the complete schematics to find the problem on power supply. The big capacitor is ok, the bridge rectifier GBU10V08 is ok too. But I don't know where can be the problem without a schematics to follow. Someone has the complete schematics or service of this PSU or similar ? Thanks a lot for your help

[madcrow]

Re: XFX XTR 650W power on issue

Just in case anyone else runs into the same issue and stumbles on this thread:

TL/DR
The actual root cause is a failed solder joint at the boost (PFC) controller's 1206-size voltage feedback resistors. Just resolder R312, R323..25 and C317 on the bottom of the main PCB, as well as pin #4 of the PFC&LLC controller board (pic attached) and profit!

Background
My experience was 100% same as OP's. One night I powered off my PC, the next day it refused to turn back on... most of the time... but sometimes it did. When performing measurements, I also received the same voltage values as OP did.

Analysis
OP did an excellent job reverse engineering the schematic of the unit, and tracing the cause of the malfunction back to the failure of the LLC controller's "mains voltage range" check.
However the failing self-check is not the root cause. By altering its configuration you are treating only the symptom, not the underlying problem, you are masking the issue instead of solving it. There is nothing wrong with how it is set up out-of-the-box: the hysteresis lower threshold is 300V, higher threshold is 365V by default. The problem is the bulk/bus voltage. (Just to be clear, I am referring to the output voltage of the PFC/boost converter stage - the one that you measure accross the terminals of the big el. cap. The voltage OP is referring to as Vbus, is not the actual bus voltage.)
Generically speaking, in APFC-equipped SMPS design, the bulk voltage is typically in the range of 380-400V DC. 356V is very suspiciously low. It is very unlikely to be the design voltage in any SMPS. In fact if you calculate the intended bulk voltage from the ref voltage and the resistive devider network on the Vsense pin, you'll find that the PFC in this model was intended to deliver ~392V DC.
OK... so we have a malfunctioning control loop. But what is the cause?
Why is Vbus so low? Is the controller struggling to increase it further but the high-voltage part of the circuit just can't deliver? Or is it not even trying as it thinks it is already at the specified level. Apparently the latter, as Vsense = Vref = 2.5V.
So we have a failed feedback. Something is off with the resistive divider. Or maybe Q301. So let's see: what may cause the voltage on Vsense to be higher than expected? Well, either the perceived resistance of R307 increased or the resistance of any of the others (R323..25) decreased. Let's measure them! Suprise: all values close enough to nominal. Hmm... WTF?
OK then... lets see how it behaves live, under load. Let's measure the voltages on each terminal of each resistor ->determine their individual voltage drops (the ratio of which is equivalent with the ratio of their resistance) -> determine the resistances which correspond to that ratio considering Vbus=356V and Vsense=2.5V. As it turns out it gives: R323=940k, R324=1085k, R325=920k. Hmm...interesting. R323 and 25 are way out of tolerance, while R324 is nominal-ish (1085k was the exact same value that the resistance measurement returned in the off-state).
Maybe I measured wrong? It's a possibility since the imput impedance of my DMM (10MOhm) is in the same magnitude as the total resistance of the divider. But fortunately, we are measuring a control loop here -> the voltage at the measured point will drop momentarily -> Vsense decreases well -> and the controller immediately compensates -> by boosting Vbus until Vsense reaches 2.5V again and thereby the point being measured reaches its original voltage. Confirmed by simulation.
So it's not the measurement. It's really the resistors. Or are they? Let's swap the good one with one of the faulty ones. Let's see if the voltage error follows! (Beware: all HV board areas are protected by a layer of conformal coating. You can remove it with some acetone. In addition, all components on the bottom side of the main PCB are glued to the board! Your best bet is to use a hot-air rework station and twist the components once the solder on their terminals melted - as the glue is unaffected by heat.) To be thorough, I desoldered and labelled all 3 resistors as well as C317, and in the meantime also reflowed the first few pins of the PFC/LLC controller board, just for good measure. OK... resistors swapped... power applyed... no smoke... so far so good. Let's take some measurements! Surprise #2: boost voltage is now 392V. Exactly as calculated. PSON signal HIGH -> wait a bit -> PSON signal LOW -> the PSU turns on instantly, Vbus=392V. Repeat x N -> all good.

Additional thoughts and observations
The LLC contoller's power is cut each time PSON goes inactive/high. Which also means that at each power-on attempt it starts with the hysteretic current generator activated. So in principle, every time to press the power button / set PSON=LOW, the LLC controller should be able to start up only when VINS>365V. But wait... If that's the case, how could it tun on on some lucky attempts at all, considering it was delivering Vbus=355-356V only?
The answer is: the startup transients. Each time you activate PSON and the PFC stage starts up, there is very short overshoot / transient overvoltage on the Vbus line. Which is completely normal. Whether the PSU manages to turn on or not only depends on whether this pulse is long enough and as high enough voltage (>365V) to trigger the hysteretic comparator on the VINS pin of the LLC IC. I've made lots of measurements using the peak detect feature of my DMM. Each time the PSU failed to start the peak of the pulse was below 365V and each time it started, it was above 380V. In my case "cold starting" (waiting for the big cap to loose charge before turn-on) tended to create larger pulses almost every time, while "hot starting" (i.e. only allowing marginal discharge of the bulk cap) resulted in smaller pulses (not enough to trigger VINS).

Conclusion
The PSU is working again. All pieces of the puzzle fell into their places. Except for one thing... I still cannot figure out how a solder joint can fail in such a way that the resistor it is applied to shows decreased resistances, and only in a live circuit. If you have any theories, please share, I am genuinely interested!

That's it. You can find my circuitjs models (of the feedback circuits) below, in case you want to experiment, confirm theories, etc. Good luck!

[madcrow]

Re: XFX XTR 650W power on issue

Update:
For the past 5 months, the PSU kept working fine. Until today. Same phenomenon.
Maybe some of the resistors (R323..325) are actually broken, and were temporarily fixed by the thermal shock from resoldering?
I replaced R323, 324, 325 and C317 this time. So far so good... PSU is starting up correctly again. Fingers crossed.