Switching PSU noise on +12V for CPU

[eccerr0r]

I was trying to see what I can do with my sick PVR (think it's dead... I think the onboard chipset GPU bought the farm) and noticed my +12V that was going to my CPU was pretty weird:

I think I see the typical switching PSU triangular waveform that is pretty bad but probably acceptable, got about 100mv P-P of switching noise. But at the +12V connector I see really tiny droops of another 150mV when the onboard switching PSU turns on. Couple those together it's still not quite -10% (or even -5%) but it looks a little alarming.

Will need to grab a SS from my scope at some point... though I really need to get GPIB going so I can get a "real" screenshot instead of a picture of the CRT.

[eccerr0r]

Re: Switching PSU noise on +12V for CPU



Top (ch1): the waveform at the connector at the motherboard of the +12V (2x2 connector)

Bottom (ch2): waveform at one of the Molex 1x4 for disk drives.

[stj]

Re: Switching PSU noise on +12V for CPU

caps are shot, or they would have atleast removed the small peaks and drops at the top of the wave.

[eccerr0r]

Re: Switching PSU noise on +12V for CPU

I tried checking another machine that works fine though different, older machine (sick machine is core2duo with ~250W PSU, comparison machine is a P4 with 400W PSU). The P4's power waveform at the power plug also has spikes but the total noise (main switch in PSU plus switching on motherboard) is about HALF seen on the core2duo. Yeah, looks like caps are probably shot.

How clean should it be, maybe both these setups need new caps.

[stj]

Re: Switching PSU noise on +12V for CPU

how clean?
well to be blunt, you want a flatline.
but makers will compromise and just get it good enough to not crash the cpu or corrupt anything.

i still see old psu's with a 7amp output smoothed by 6-12 caps in parallel.
like most things, the newer they are, the more corners they cut!

[eccerr0r]

Re: Switching PSU noise on +12V for CPU

Well of course you want a flat line but some ripple is expected. All SMPS will get some ripple from the switching action, but wondering how much is OK. Yes, single digit millivolts of ripple would be great, but even resistance of the wires, traces, and connectors which also contribute to ESR make a difference...

[stj]

Re: Switching PSU noise on +12V for CPU

actually you can flatline a psu output.
what you cant do is factor-in variable loads like harddrives.

to flatline a switching psu you need a combination of electrolytics, ceramics and chokes.
most dont have enough ceramics, but the ceramics remove the very short switching spikes that are too fast for electrolytics to catch.

before ultra-low esr caps, a design trick we used was to put a ceramic smd directly between the pins of each electrolytic on the underside of the pcb.

the 2 types of cap complimented each other - the ceramic was deadly-fast but had no real ripple capacity, the electrolytics were slow but had bulk capacity for ripple absorbing.

[ChaosLegionnaire]

Re: Switching PSU noise on +12V for CPU

Well of course you want a flat line but some ripple is expected. All SMPS will get some ripple from the switching action, but wondering how much is OK.

if u're asking about the maximum ripple in the atx design guidelines, the maximum peak to peak ripple on the minor rails (3.3v, 5v) is 50mv while on the 12v is 120mv. but of course u want as low ripple as possible to keep your hard drives running for as long as possible without kicking the bucket and losing your valuable data. it is for this reason that i only buy ultra-low ripple japcap psus for my computers.

as for how much ripple a cpu can take before it starts behaving abnormally or producing errors, u have to pull the cpu datasheet from the manufacturer. for intel, u have to look under electrical guidelines in the datasheet for that cpu. that should help u in catching bad caps in the cpu vrm without having to pull the caps to test them with an esr meter.

also, do note that the amount of ripple u get is dependant on the power draw of whatever the smps is connected to. the higher the power draw, the more the peak to peak ripple there is. thats why in psu reviews, they load the psu to its max stated wattage and scope the output ripple. thats how they sexpose scandalous fake rated wattage psus.

[eccerr0r]

Re: Switching PSU noise on +12V for CPU

Well one thing is the ripple generated by the PSU, the other thing is ripple caused by load. The really thin downward spikes are load caused, how that should be accounted for I'm not sure.

I tried another PSU and it wouldn't boot. But this PSU was questionable to begin with (it has a 20-pin ATX where the m/b has a 24). Plus it was a swapout since it couldn't power a P4 anymore.

The ripple at the power connector really has no bearing on the ripple seen by the CPU - as the onboard switching regulator will attenuate the noise even more... The ripple on the 1x4 molex 12V (trace ch2) is just about 100mV so it's getting close to the limit.

The unfortunate problem is that the original PSU is a custom one as it's a low profile microatx case. Cap replacement is my only option though I'm not sure even if they are replaced, whether it will be clean...or is it the motherboard caps that need to be replaced (not sure if there are any significant 12V-capable caps on the motherboard... most seem to be too low for 12V...)

[kaboom]

Re: Switching PSU noise on +12V for CPU

Maybe the input caps on that +12 to vcore converter have high ESR, and when the highside devices come on, you get the suck outs.


If the power supply had bad caps, you'd see the same, or more, noise on that unloaded molex. Did this ever occur to you?

[stj]

Re: Switching PSU noise on +12V for CPU

i would say the deep drops are a lack of ripple-handling, something very few meters can check.
the small peaks may be an esr issue.

mobo caps are probably done though or the board wouldnt show load peaks that sharp

[kaboom]

Re: Switching PSU noise on +12V for CPU

i would say the deep drops are a lack of ripple-handling,

No, they don't appear at the cap in the supply, which the unloaded disk drive connector "jumper wire" allows the scope to see.

[stj]

Re: Switching PSU noise on +12V for CPU

if you see something on the board but not at the psu, either your wires arent up to the job or you have a bad joint or poor/burned crimp terminal

[kaboom]

Re: Switching PSU noise on +12V for CPU

if you see something on the board but not at the psu, either your wires arent up to the job or you have a bad joint or poor/burned crimp terminal

No, "if you see something on the board but not at the psu, the load is backfeeding ripple into the caps (and feedback ckt if PSU's caps are marginal) in the originating supply."

Remember, certain video cards were supposedly "incompatible" with certain power supplies, whereby a "mismatch" made those vulnerable SMPSes hunt around and shutdown when the VGA card started taking load and backfeeding its ripple into the PSU.

Regardless, allwire has resistance, and the caps at the supply, while they are shunting the suckouts to a degree (seen on unloaded molex plux), the caps on the board are supposed to provide current to "fill in" those suck outs.

Even more with the choke most mobos have between the incoming +12 and their bank of caps before the high side devices.

And since all wire has resistance, you'll still measure more ripple at a load such as a CPU buck converter, unless extreme filtering is provided at the board; again, it is not the purpose of the caps in the SMPS to "anchor" something at the down at the load.

The ripple current loop at the downstream converter is commutated/shunted by those caps on the board! If increasing wire size back to the source reduces ripple voltage present at the load, then the caps at the load are marginal. Even w/o burned terminals- all wire has resistance.

[eccerr0r]

Re: Switching PSU noise on +12V for CPU

Right, as said earlier, I was wondering if it was OK, and some of this could possibly be wire related. The probe point is on the upper side of the connector, so the board 2x2 connector resistance is not part of the circuit to the scope. I should try more machines or find another PSU to compare with. I guess it does make sense as the MB should have a buck supply that yanks on the rail for a fraction of a cycle, but wonder where the cap in question should be (mb, psu?) Or is it just the wire?

I don't recall this PSU as a dual rail 12V (which could also explain things, alas it's not) but what bothers me is that if these circuits are truly hooked up to the same tap, the only additional series resistance to the CPU power plug is those two 1ft pieces of wire to the m/b, else all outputs should look like the same especially since they're unloaded.

[EDIT]
OK, I see one of my dead Asus motherboards actually has a 16V cap on it, assuming that is the onboard 12V filter for the 2x2. Now I need to look for the same for the Foxconn which is the board in question. Still surprised it will show on the 2x2 top pin but not on the other unloaded taps of the PSU.

[EDIT 2]
You know what, fsck this. I'll stick in some caps on the top of the 2x2 connector and see if the behavior changes... ha...

[EDIT 3]
I stuck in a phony Nichicon 3300uF 16V cap in parallel with the 12V 2x2 (stuck it into the top of the connector and forced it in). Got the ripple down to ~100mV total. The funny thing is that this is a phony and while I was experimenting...

POP! Goes the Weasel!

This is the first time I saw a cap pop. I don't even think I connected it backwards, it's correct according to the sleeve and the wire lengths (shorter wire is usually GND). Alas it popped, steam wheezed out. Dead Fake Cap, nothing lost, experimental data gained.

[eccerr0r]

Re: Switching PSU noise on +12V for CPU

Weasel: FAKE FAKE FAKE

FAKE NICHICON
Before popping: capacitance: 3500uF
ESR = 0 (cannot measure, my ESR meter resolution doesn't go that low)
Leakage at _10_ volts: 75 microamps

After popping: capacitance: 2100uF
ESR = 0.9 ohms
Leakage at _10_ volts: 75 microamps

Testing leakage at 15 volts, my Sencore says No Can Do, Leaks Too Much.

[stj]

Re: Switching PSU noise on +12V for CPU

16v is too low for a 12v line - even if a few people will now start flaming me over saying it.
also, if that cap has been sitting for over a year it may have started to depolarize.

i had that the other day, dug out a working board that i stored a couple of years back,
powered it up - BANG - fucking teapo 6.3v cap on the 5v rail!!
replaced the bitch and a matching one with a pair of 16v rubycon's

[eccerr0r]

Re: Switching PSU noise on +12V for CPU

Anyone have leakage meters? The caps that I have, the "good" ones - even used ones from top end manufacturers, barely leak at all at their rated voltage. It's these shitty fakes that aren't close. I think these caps are really 10V and not 16V, and no wonder why they blew up at 12V (though I did not expect them to behave as they did... thought they would last at least for a little while...)

[stj]

Re: Switching PSU noise on +12V for CPU

leakage rises with age.

[eccerr0r]

Re: Switching PSU noise on +12V for CPU

Leakage also rises with how close it is to their rated voltage...