PSU Gurus: PSU Ripple?

[tj2]

Having one of the Tektronix scopes at work all to myself this week, I'm having an orgy of PSU investigation. I checked two new AT power supplies -- one from Zippy and the other from Astec -- and I checked four different ATX power supplies -- two new units from Delta, the Seasonic that I'm going to recap and Everell mod, and my recapped old PC Power and Cooling. I also tested four other AT style PSUs that I had recently recapped.

I tested the PSUs with small loads on the +5 and +12 V rails; I used an #1156 auto bulb on the +12 V rail and a #1493 bulb on the +5 V rail. The load on each rail is about 20 W or so. To make the dummy load I soldered the bulbs to a drive cable extender from Radio Shack. In addition, on the ATX PSUs, I used an Enermax PSU tester to turn the PSUs on and off and to provide a (tiny) load to all rails.

The results are surprising, at least to me! Only the Zippy is within 1% ripple on all rails. The recapped PCP&C is close to being within 1% on the positive rails and the -12 V rail, but is off on the -5 V rail. All the rest of the new units are WAY out of spec! For comparison, I've listed last the results on my recapped old AGI BPS-2504-4U 250 Watt AT PSU. It's numbers look better even than those of the new Zippy PSU!

Here are my questions for the PSU Gurus. Can these ripples be right? Are PSU makers getting sloppy or are their (new) caps nearly bad? Would any of the PSUs be dangerous to the longevity of my motherboards and hardware? I didn't use any significant load on the +3.3 V rail. Would that affect the ripple?

Perhaps I should order some new quality caps and recap to see if the ripple improves.

Voltages:
Nominal	Actual		Ripple (p-p)
-------------------------------------
Astec SA301-3400 300 Watt AT (new)

 +5 V	 5.19 V	174 mV
+12 V	 11.69 V	336 mV
 -5 V	 -5.03 V	400 mV
-12 V	-11.80 V	470 mV

Zippy/Emacs SP2-4300F 300 Watt AT (new)

 +5 V   5.03 V	30 mV
+12 V  11.79 V	62 mV
 -5 V  -4.84 V	45 mV
-12 V  -12.11 V	81 mV

PC Power & Cooling Turbo Cool 300ATX (recapped)

 3.3 V	 3.47 V	 43 mV	
 +5 V   5.10 V	 47 mV
+12 V  11.71 V	 96 mV
 5 Vsb  5.01 V	 50 mV off 
 -5 V  -5.09 V	 90 mV
-12 V  -11.85 V	132 mV

Seasonic SS-300FS 300 Watt ATX (new)

 3.3 V	 3.39 V	 88 mV	
 +5 V   5.08 V	145 mV
+12 V  11.90 V	178 mV
 5 Vsb  4.96 V	 30 mV off
 -5 V  -5.01 V	198 mV
-12 V  -12.17 V	186 mV

Delta DPS-250JB 250 Watt ATX (new)

 3.3 V	 3.40 V	170 mV	
 +5 V   5.18 V	160 mV
+12 V  11.75 V	172 mV
 5 Vsb  5.17 V	 30 mV off / 102 mV on
 -5 V  -5.01 V	340 mV
-12 V  -12.08 V	830 mV

Delta DPS-300KB-1 A 300 W ATX (new)

 3.3 V	 3.35 V	 87 mV	
 +5 V   5.09 V	 91 mV
+12 V  11.87 V	203 mV
 5 Vsb  5.03 V	 34 mV off
 -5 V  -5.01 V	536 mV
-12 V  -12.02 V	150 mV

AGI BPS-2504-4U 250 Watt AT (recapped)

 +5 V   5.14 V	25 mV
+12 V  11.96 V	30 mV
 -5 V  -5.27 V	25 mV
-12 V  -11.64 V	25 mV

[Toasty]

Re: PSU Gurus: PSU Ripple?

When measuring ripple, you need to have -most- of the connectors connected to something. The PSU test setup I made has the mobo connectors tied to a block where loads can be applied. I use headlamps and power resistors to get at least a 20-30% load on all the outputs, including 5vsb. Several hard drives are also used to load the -last- connector on the drive supply lines.

Some of the biggest problems are:

- The loose unconnected wiring acting as antennas and picking up induced signals/voltages from the switching transformers. Being at the same frequency, you can't tell whether it is truly ripple or an induced signal.

- PSU cover/shielding not in place when measuring.

- Scope in close proximity to PSU.

- Input cable to scope too close to PSU or excess cable length.

- Improper connections from scope to PSU. I use a double-shielded wire as input to the scope and connected to the output block above and the ground line is only 3/8" long. Every voltage output has a ground (-) terminal as its neighbor. The typical scope probe ground is several inches in length with an alligator clip on the end. It's an great antenna!

If you are getting any reading on the scope -before- you connect, it's an induced signal.

BTW - holding the probe near a trafo is a good way to see if its working. No signal means it's not.

Toast

[tj2]

Re: PSU Gurus: PSU Ripple?

Thanks, Toasty. I did some searching after posting and found the Intel power supply design guide (, which says that one should put a 10 uF electrolytic cap in parallel with a 0.1 uF ceramic disk cap across the connector to "simulate system loading." I'm going to see if that changes things significantly. I imagine that these caps soak up any induced currents, which should not have anything like the kind of power that a true signal coming through the filters would. Looks like I'm going to be making some more dummy loads!

[370forlife]

Re: PSU Gurus: PSU Ripple?

Something is definitely off here, a member at jonnyguru tested a delta dps-300kb to find less than 30mv of ripple on all rails.

How are you connecting the power supplies to the scope? If the connections are loose, it is going to cause some problems.

[ratdude747]

Re: PSU Gurus: PSU Ripple?

sorry to revive an old thread, but i have a related concern:

a psu i did a nearly full recap on (i didn't have enough so i had to leave in 2 brand new GL caps) is showing ripple on a multimeter... i was getting a difference of .01 to .02 v out of it.

would that explain why a system running off of it is running hotter than it did in another case/psu combo?

btw, it is an ultra V 500w w/ better primary caps and all but 2 10v 680uf caps recapped. i also shorted the 115v switch, got rid of all mid-wire splices, and replaced the 12v+ plug/wire. it is powering a socket 939 althon64 x2, no OC, asus a8n32-sli... it is running so hot that under med-high loads it will thermally shut down... maybe the hot weather is the culprit, but i just want to be sure.

is tht amount unsafe? if so, should the psu meet the scrapheap??

[Toasty]

Re: PSU Gurus: PSU Ripple?

How are you seeing ripple on a multimeter?? Ripple is at switching frequency (30kHz+) and is seen with a scope, not multimeter.

Perhaps what you are seeing is voltage fluctuation and not ripple? Could it be poor regulation under low load condition, or rail loading imbalance?

Toast

[ratdude747]

Re: PSU Gurus: PSU Ripple?

i thought ripple WAS voltage fluctuation... whoops, not thinking.

it did it on all rails.

afaik the regulation is good. beefy guts were attached to the heatsink.

is my fluctuation out of spec? the voltages are well in spec.

sounds like the psu is not my problem.

[370forlife]

Re: PSU Gurus: PSU Ripple?

The voltages can do whatever they want as long as they are in the + or - 5% tolerances (except the -12v which is + or - 10%) so long as they are there don't worry about that.

[Behemot]

Re: PSU Gurus: PSU Ripple?

The voltages can do whatever they want as long as they are in the + or - 5% tolerances (except the -12v which is + or - 10%) so long as they are there don't worry about that.

True, but still, the more stable the voltages are, the better for your system, if you want to do major overclocking, stable voltage helps significantly. After burning my Seasonic I am stuck with recapped Eurocase and had to lower OC from 2,8 to 2,7 GHz immediatelly after connecting the PSU.


As Toasty said about unconnected connectors, does it have some impact on ripple in real system or does it only make harder to measure it? I know it made problems e.g. with SCSi for that reason it was ended with those terminators. Could something similar help the ripple or the cables are so weak antennas they have no actuall impact?

[Behemot]

Re: PSU Gurus: PSU Ripple?

I'd open this thread again…it was not that bad before (quite OK in some cases actually), but now when guys with the equipment I am using have moved, I am picking some terrible stuff on the ground cables in the new room. Even two ground wires in the same molex read around 40 mV ripple one to other.

What I will do is make myself a small board with BNCs and the decoupling capacitors on it directly which will be connected between the probe and O-scope itself. Jonny at JonnyGURU has it that way, only that he has the caps on his load directly, at least it seems from this tiny photo:

and I will have them on the O-scope end of the probe. I have succesfully proved the ripple is at minimum when I connect probe over some load resistor or bulb. So with the caps on probe (I will also get 100% same which Intel reccomends), I will just measure ripple directly from Main ATX connector as everywhere else I am getting bullshit.

Also I have been thinking about ending the free cables with terminators as good ol' SCSi has been using. So some 100k resistors on free cables, whaddaya think? Problem is I can hardly end all liek this as getting SATA males is some difficult task. Maybe making my own ones from some PCB? Those are basically some printed contacts, nothing more. Do you think I need to end every single connector, or ending the ends of each cable is enough?

[Behemot]

Re: PSU Gurus: PSU Ripple?

OK. I've made myself them terminators with 1k resistors (realized 100k is too much) and the neighbour grounds have calmed down.

Now I am still not sure about the capacitors. Placing them in front of O-scope just filtered out every signal making it 400 uV ripple so that's not the way

I've measured one PSU with probe directly on the decoupling capacitors on my loader. The results were not incredibly high, more like incredibly low, 40 mV top and it was not THAT good PSU. Especially the +12 V rail was suspiciously low. I am using 0,1uF ceramics with 22uF Nichicon PW per rail.

Now Intel specifies to use 10uF tantalum cap with 1 ohm ESR, 0,39 A ripple. I selected the 22uF because it is closer both ESR and ripple wise. But it may be actually filtering the signal more because of it's double capacity hence the low ripple. Everybody else uses aluminium electrolytic cap, Jonny even general purpose NCC KMG series. So the question is, does the ESR and ripple matter here? Or should it be just 10 uF and who cares the rest?

[PeteS in CA]

Re: PSU Gurus: PSU Ripple?

The typical scope probe ground is several inches in length with an alligator clip on the end. It's an great antenna!

A great test to see how much of the noise you "see" is just radiated into the ground lead is to connect the alligator clip to the probe tip and hold it near where you plan to measure. There are sockets available into which one may plug a Tek (or similar) scope probe that has the ground lead and "witch's hat" removed. Or you can fashion something like what Behemot did. That little collar on the probe just back from the needle point is ground. It isn't coax all the way to the signal point, but it's close enough. Bent and formed bus wire can work quite well, and is a LOT cheaper than buying little sockets. One thing Toasty didn't mention in his post near the beginning of this thread is that ripple is typically specified with a scope bandwidth of 20MHz. Personally, I've never used the 10uF + .1uF cap thingy.

[Behemot]

Re: PSU Gurus: PSU Ripple?

Well I am not usually picking strange things when having clip connected to tip, not more than several mV. But if I connect both to where it should not be taken readings from, I see straneg things. I'd say it's obvious as they write it in white paper, but you know…man asks why it should matter? It does

The caps do incredible difference. I have experimented with that today. Basically, without the caps, ripple can reach 100 mV and much more. Those are just false readings and the same you can obtain by measuring somewhere else than you should. With caps it drops to 20-40 mV level. I am gaving a thought to the idea of re-measuring PSUs I've called bad and not complying with ATX spec so far as just by measuring on caps directy I have differences sometimes way greater than 100 %.

It is really difficult and a bit of alchemy and I think nobody can say „I measure the actuall ripple!“ and you can more or less just get closer or further from the spec and compare between PSUs measured the same way. But I think I am pretty close to it now as I am following the Intel's specification as much as I can:

• as of today, using 0,1uF ceramic and 10uF Nichicon PW (10uF SMD tantal specified)
• measuring on the caps (which is almost as Intel specifies it - on loaded path with probe tip in front of the caps and ground behind the caps)
• 20MHz bandwidth limiter

I only lack differential probe now. In future I will try to get rid of extra noise by using extra filtration in front of measured PSU, using more shielded probe, extra shielding the PSU and the load tester and also galvanically separating PSU from mains.

With reading taken from caps directly, I am quite surprised of how low the ripple is even on PSUs I'd personally value as mediocre. The reason why for example Jonny and others have it higher is because they most likely have some wires connected to the loaded paths and these wires have bypassed with caps. ANd there they measure. Or they even measure ripple on un-loaded conenctors!

I am not saying you cannot measure that, but ripple is specified somehow. If you measure it somewhere else somehow else, it is something, but not ripple according to ATX specification.

One question though, can the capacitor type choice effect it somehow by great margin? I experimented with 10- and 22uF caps and the difference was neglible when measuring free cables, when soldered 10uF instead of 22uF, I have some ripple a bit higher, some actually 50 % lower. As I measure most of the ripple at aprox. 2,5 MHz level which are only some strange harmonics, I think it is more affected by ceramics on motherboards rather than aluminium lytics, they just cannot filter this out. That's why the results are overall +- the same. But still, Intel was using tantal.

Can tantals filter ripple/noise at 2 MHz level? Or do they filter even less than lytics as they have higher ripple current, but way higher ESR also?

[Th3_uN1Qu3]

Re: PSU Gurus: PSU Ripple?

Tantalums are better than lytics. They *do* tend to explode or catch fire if abused, but they are available in capacities comparable to small lytics without costing a fortune (like ceramics would). So yes, they are perfectly capable of filtering 2MHz.

As switching frequencies got higher, they mostly use ceramics in VRMs and other small/low voltage DC-DC converters nowadays, but a few years ago, tantalums were commonplace. Just look on an older laptop board (2000-ish) and it's going to be full of them.

[Behemot]

Re: PSU Gurus: PSU Ripple?

OK, so it is possible that I'll get lower ripple on the O-scope if I'll use tantalum caps?