Is this voltage oscillation normal for an ATX power supply for a PC?

[PirateYarr]

Hi everyone!

I had an issue with my home server, where it randomly rebooted once every other week or so with no error messages being written in the syslog. After consulting some people, the conclusion was that there is probably some hardware fault with the server. So I stress tested it, ran memory tests, Prime95 to torture the CPU etc. and it passed without a flaw. Yet, the random reboot happened after about 8 days after the stress testing was completed.

So I connected my oscilloscope to the 12 V rail of the Corsair CX550F power supply (on the connector that goes from the PSU to the CPU power pins, 8 pins in total), and found something shocking (no electrical pun intended).
Over one night, I had about 10 occurrences of the following phenomenon, seen in the attached screenshots. The voltage goes unstable and oscillates back into its nominal voltage. It reaches as high as 16.88 V and as low as 7.28 V. The whole process takes a few microseconds.

For information, the server is not power hungry at all. It has an ASUS ROG Strix B660-I motherboard, Intel Pentium G7400 CPU, Corsair Vengeance 32GB (2x16GB) DDR5 5200MHz CL40 RAM, 2x Gigabyte SSD 256 GB M.2 2280 PCIe 3.0 x4 (NVMe) and 4x Western Digital RED 4 TB HDDs. The system consumes about 32 W at idle, and around 60 W when all drives are spinning and the CPU is performing some tasks.

I got in touch with Corsair and they replaced the PSU with an RM750x. Now that I have scoped its 12 V, I see a similar phenomenon occurring multiple times per day, but the voltage doesn't swing as wildly. On the RM750x, it swings up to 12.98 V, then down to 11.00 and it stabilizes back to nominal voltage in about 2 microseconds. The RM750x seems to keep the voltage swing confined to a smaller interval and recovers faster.

Which picture shows which PSU:
SDS00007.png = CX550F PSU
SDS00012.png = RM750x PSU

But the main question is: are these transients normal for computer PSUs? If so, what causes them?

Thanks!

[janoc]

That looks like an inductive ringing due to some load being connected or disconnected. E.g. a drive motor starting/stopping (hint, you have 4 spinning drives there ...) or a fan starting up/shutting down.

I.e. nothing to do with your PSU. That the power supplies have slightly different regulation characteristics and response is normal and to be expected, esp. with the CX line which is the Corsair's low end stuff. It is also very unlikely that the supply will be able to filter out such spikes, esp. if you are measuring at the end of the relatively long cables with their parasitic inductance which prevents the supply from being able to react this fast anyway.

Normally such short spike is not going to do anything because within a microsecond the voltage was back in spec. Anything connected to the 12V rail will have large capacitor banks that will "eat" any such short disturbance.

Your machine is most likely crashing because of something else and not this - e.g. bad RAM or some motherboard defect. Or bad/buggy software/OS.

[PirateYarr]

Thanks for your reply!

These transients or ringings happen randomly without drives starting as far as I can recall. The drives in the server only spin up when they are accessed, and I wasn't accessing them while testing over night. All system fans run continuously - there is no starting/stopping going on there. As far as measuring goes, I measured at the PSU side, as closely as I could to the PSU.

I thought it was software first as well, since that usually is the case, but the server is running Linux (Unraid OS) and there is nothing in the syslog about anything crashing, which points to a hardware fault.

The RAM was tested both with Memtest and Prime95 Large FFT, which uses up the entire RAM capacity. No faults.

The only thing I can think of right now would be if the Intel Pentium G7400 CPU was causing these because of C states or some type of frequency boosting, but if I remember correctly, I disabled all that stuff in the BIOS to make the CPU run as constantly as possible, i.e., no sudden changes in frequency or power consumption.

As you can see with the CX550F, the voltage drops to about 7 V. If the CPU happened to require a lot of power at that exact moment, I am guessing the low voltage could cause it to halt and reboot or get stuck. 9 times out of 10, the system restarts randomly, but it can also stay powered on with no response, i.e., it locks up.

I guess time will tell, since the server is now running and I am monitoring it.

[PirateYarr]

And to be clear - the focus of this thread is not to help me find the root cause of the random restarts (although that would be nice if someone just happened to have the answer), but rather to assess whether these transients are to be expected in the microsecond range in a PSU.

[jonpaul]

Just see this thread. Have designed, and mfg SMPS since 1980s.

The scope shots may be picking yp transients and ringing unrelated to the ATX PSU.

The fast rise, and high ring freq may be artifacts of the measurement and probe technique.

A photo and sketch of the EXACT probe, ground return, pace the voltages are measured is needed to commnet.

1/ See the fine AN-90B     HP DC Power Supply Handbook
 for technique of PARD measurement on SMPS.
https://archive.org/details/DC_Power_Supply_Handbook_Agilent_Technologies_Application_Note_90B

2/ Our technique for DC PS noise

Find output bus electro cap,  connect DIRECT to the cap + and return.
NO X10 or other probe.
Use a BNC to RG174 coax, make a DIRECT connect from end of cable to DC bus at the cap. Keep both center and shield of cable SHORT to the cap. (~ 1 cm)
Set scope on AC, be sure bus V will not damage scope input.

The transients shown may be EMI or power mains originated and you are picking up the stray noise unrelated to the PSU.

Jon

[DimitriP]

When all other tests pass, power supply is the culprit.
(Unless you have in intermitently shorting reset button switch!)

[PirateYarr]

Thanks for your detailed reply jonpaul! I will definately check your mentioned resources out when I have the time. It's not easy to find time for all this tinkering with a toddler around

I think you and janoc were right - this voltage transient may not be the culprit after all. 4 hours ago, the server restarted again. I will have to dig deeper. Thanks for your replies!

[RoGeorge]

That voltage variation is way off, not normal.  Most probably the power supply is also the cause for the random resets and unexpected computer crash you are seeing.

I would replace the power supply with a new one.

[wraper]

What you measure probably has nothing to do with PSU. Ground alligator clip wire length alone is enough to catch all sorts of garbage, not to say long cable from PSU. Unless you attach some small capacitor like 10uF to measurement point and make direct connection to GND of the probe without wire loop to remove interference, this proves nothing and is totally worthless. As of memory tests, IME they can find instability or RAM failure, but in no way they can provide proof that RAM is actually stable. I had too many cases when test can run OK for days but computer crashes once in a week or two until you change RAM or lower clocks. BTW most of the consumer RAM is factory overclocked, often even JEDEC timings in SPD, not to say XMP.

[PirateYarr]

What you measure probably has nothing to do with PSU. Ground alligator clip wire length alone is enough to catch all sorts of garbage, not to say long cable from PSU. Unless you attach some small capacitor like 10uF to measurement point and make direct connection to GND of the probe without wire loop to remove interference, this proves nothing and is totally worthless. As of memory tests, IME they can find instability or RAM failure, but in no way they can provide proof that RAM is actually stable. I had too many cases when test can run OK for days but computer crashes once in a week or two until you change RAM or lower clocks. BTW most of the consumer RAM is factory overclocked, often even JEDEC timings in SPD, not to say XMP.

Thanks for slaughtering my measurement method! No but really, I am fully aware of my limited experience with the oscilloscope and measuring "esoteric" signals, so this kind of feedback is very useful to me, so thanks for that.

Yes, I did try to find the "really default for real" timings for the RAM but I keep coming up with this (from the corsair website):

| Performance Profile       | XMP 3.0                           |
| SPD Latency                 | 40-40-40-77                    |
| SPD Speed                   | 4800MHz                         |
| SPD Voltage                 | 1.1V                                |

And I did try running them at those specs earlier, but I will have to try again soon. I feel like I'm gonna have to start a new round of thorough methodic testing to solve this damn annoying issue.

Anyway, thanks so much for the replies and the insight to all who replied so far!

[PirateYarr]

Set scope on AC, be sure bus V will not damage scope input.

Hey jonpaul, just curious about this. I'm going to ask a question that will clearly show my level of knowledge with oscilloscopes and metrology So, the question in question: setting the scope to AC mode will reject all DC signals according to my scope's manual. But I am both interested in the overall DC level (like, I want to know if the voltage is actually 12 V or 11.5 or whatever), AND I am interested in picking up anomalies on the bus. How would I do that? Using several channels of the scope? Several scopes?

For this particular measurement, I just grabbed one of the scope's original probes, set it to 10X (not sure why I did this, but I noticed I could adjust the voltage offset in finer steps this way but I do know the 10X acts as a "divide by ten" so one can measure higher voltages), then I used a little thing that cuts into a cable and locks on to it, almost like a "tap" so that one can access the wires inside the cable with ease. I am aware of the brutality of this method, but I find time is a scarce resource these days now that I have a toddler at home and I simply can't spend hours reading up on the best way to measure signals and preparing things in detail for it. Right now in life, my mindset is more like that of a butcher's: hack, rip and slash until the job's done. It gets me results that are of so-so quality, of course

[Haenk]

As the MB is quite new, I'd rule out bad caps.
Just try a new PSU, that's almost always the solution when the system behaves weirdly.
Had that issue 2 weeks a ago - a new PSU right out of he box, system rebooted in BIOS menu - after BOS update, it made it to Windows installation, but kept crashing in idle or under full load. Eventually exchanging the PSU solved all issues. Unfortunately it was already out of warranty, so a nice Platinum PSU went right into trash

[wraper]

Just remembered about this video Dave made 14 years ago. Almost exact like OP captured.

[wraper]

Had that issue 2 weeks a ago - a new PSU right out of he box, system rebooted in BIOS menu - after BOS update, it made it to Windows installation, but kept crashing in idle or under full load. Eventually exchanging the PSU solved all issues. Unfortunately it was already out of warranty, so a nice Platinum PSU went right into trash

PSU you trashed probably was not even faulty. I've seen some poorly implemented motherboards which would behave weird with some perfectly working brand new power supplies. For example part of USB ports would not work and how many would change on each power up. Also seen some other weird issues I remember not that well. I guess it had something to do with motherboard not liking voltage rise time on power rails or their sequence.

[wraper]

And I did try running them at those specs earlier, but I will have to try again soon. I feel like I'm gonna have to start a new round of thorough methodic testing to solve this damn annoying issue

Try leaving just a single stick in second or 4th slot from CPU. Manually lower the frequency and leave it running like that. Resolving RAM related problems can be time consuming and excruciating. It would be the best if you had some different RAM sticks to try. I basically gave up on non ECC RAM because of this, got fed up with random crashes on multiple PCs I had over time happening every so often. And which were only possible to sort by firing parts cannon and in the end it nearly always was RAM (RAM tests said it's fine) which also often worked perfectly fine in another computer.

[jonpaul]

Pirate:

Scopes are not very accurate for precise levels like DC. See scope and PSU specs.

Use a DVM/VOM to check DC levels but 99% of ATX/PC/MAC computer PSU do not fail on voltage cal.

10X probes are bad for looking at SMPS noise/ripple etc.

See the HP Power Supply book referenced before


Jon

[inse]

I would suspect a massive ground loop picking up disturbances.
As jonpaul already mentioned, please show the setup of your measurement.
The low voltage GND of your power supply is grounded via the mains connector.
The ground clips of your scope are grounded the same way, so there is a imaginary loop created consisting of ground lines.
Let’s start from the ground clip of the scope- mains connector- wall outlet- house wiring- wall outlet for the power supply- mains cable- secondary ground- back to probe clip.
This loop will pick up disturbances created by switching events (fridge kicking in, switching of inductive loads like fluorescent lamps, halogen transformers etc).
Those signals will be coupled into the input amplifier of the scope and interfere with your measurements.
The spikes appear to be too sharp to really be present on the heavily buffered 12V rail of the PSU.
Solution for the ground loop problem is to interrupt it by using an isolation transformer (I won’t mention the other method) or to use a differential probe.