Going about PSU testing all wrong?

[jonnyGURU]

Ok. This discussion started in OT, but I'm kicking it over here for some educated opinions (or some Anandtech opinions... whatever the case may be.)

I built this out of some 1157 lights, a distribution block and some wood:

http://www.jonnyguru.com/tester

I plan on putting four more bulbs on it so I'm up to almost a 20A load, but for now I've been throwing a couple 300W power supplies on it with the four bulbs to see how effective it is.

Theoretically, the four bulbs should only be a "9A" load. I've got two good 300W power supplies here, an Ultra X-Finity 300W and an Antec True 330, and even though the 12V rails are rated at 12A and 10A respectively. Both of them drop the 12V below the ATX tolerant 11.4V after only 9A and 7A loads, respectively.

I went ahead and put a Powmax Demon 580W on there even though I expected it to only show a change after more than four light bulbs (despite how crappy it is, I expected to last AT LEAST that long) and it actually TRIPPED after only two light bulbs! I never even got a chance to turn the second light ON and the PSU tripped!!!!

Colt45 over in OT seems to think it's because there's no load on any of the other rails. I trust his opinion, but Isn't that sort of old school? Do I need to have a load on other rails if I expect the 12V rail to live up to it's rating?

EDIT: Changed title because after I started loading the 5V, I've been able to get a stable 12V.

 

[PurdueRy]

are those lights wired in series or parallel??

 

[jonnyGURU]

They're in parallel.

I also had these two "Typhoon" 80MM fans with red LED's. I hooked them both up to the 5V of the Antec and much to my enjoyment, they both spun up AND the LED's lit up too! I plugged in the multimeter and got a reading of 11.1 on the 12V with four light bulbs on. I then unplugged the fans and the voltage on the 12V actually dropped to 10.8. I then plugged the fans back in and the voltage went back up to 11.1. Interesting.

So, do you think If I put MORE of a load on the 5V, the 12V may actually go up some more???

What makes the PSU act this way?

Even without a 5V load, is this still a good test? I mean, in a comparative lab environment of course (A versus B versus) since nobody would actually ever run an "all 12V PC" making the results completely synthetic.

 

[PurdueRy]

welll....there is a problem as the impedence of a light bulb changes as it is heated. And there tends to be a huge spike in current when they are first turned on

 

[jonnyGURU]

THAT in itself can be a potential problem, but it doesn't explain the problem at hand. The bulb, cold or hot, is pulling the voltage down below spec even though I'm not pulling as much amperage as the power supply manufacturers say the PSU can handle.

I found a halogen bulb and hooked it up to the 5V and now the 12V on the Ultra is up to 11.5V (a whole .23V!) which is still within spec. Now if I only had another light bulb to hook up to the 12V!

Now I'm just curious why I have to have a load on the 5V in order to get any kind of performance on the 12V.

 

[ribbon13]

When I build a PSU Breakout box (for multiple rails/psus) I use stiffening caps and these to combine all the rails.
http://pcpowercooling.com/products/cool...essories/popups/index1Aloadresistr.htm

And use a car amplifier with some of these to test the 12v rail
http://www.partsexpress.com/pe/showdetl.cfm?&DID=7&Partnumber=019-025

 

[jonnyGURU]

So what's that little guy from PC Power and Cooling do? Just puts a 1A load on the 5V rail?

 

[jonnyGURU]

Oh.. and now that I've hooked up the light bulb to the 5V, I've been able to hook up the Powmax and get it to run, but it's still rather pathetic. It drops down to 11.1 on the 12V rail with only a 7A load. How do they figure this is a 580W capable of 24A on the 12V?? I didn't expect much for a $30 power supply, but I've got 300W PSU's doing better than this thing!

 

[ribbon13]

here's some threads you might find interesting. not sure though
http://www.diyaudio.com/forums/showthread/t-27364.html
http://206.131.241.58/ubb/ultimatebb.php?ubb=get_topic&f=1&t=002544

Yeah, that little guy is an Arcol HS series.

 

[jonnyGURU]

The second thread was very useful.

Ok.. So the PSU doesn't regulate it's power unless it's minimum load is met. Unfortunately, none of my power supplies have a minimum load rating listed on them! So is it safe to assume that the 1A load provided by the Arcol HS is enough to regulate the 12V rail?

You know what, though... I'm still going to use a light because the light gives me a visual of the test bench being on. Maybe I'll install one 1157 Amber right in the middle.

 

[JimPhelpsMI]

Hi, No power supply will put out the same voltage under all loads. Check the specs on the PS and I think you will find that the tolerance on the 3.3 and 5 v are 5% and all the others are 10%. That's +/- 1.2 V on the +12. Jim

 

[jonnyGURU]

Oh. You're right. You'll never find a PSU that has a 12V that ALWAYS puts out exactly 12V. But the specs are actually 5% on the 3.3V, 5V and 12V to be within Intel ATX spec. The 10% you're thinking of is just for the -5V and -12 which isn't even used by anything but the amplifier portion of onboard sound chips and certain serial port devices. You used to be able to throw "some ISA devices" in that sentence, but honestly.... WHO USES ISA ANYMORE?!?!

Speaking of onboard sound problems.... Any of you guys remember how the MSI 6340 on board sound wouldn't work if you had a PSU with a crappy -5V rail. Aahhh! Memories.

 

[VirtualLarry]

Hey, sounds like a neat project. You might inspire some more of us to build something like it, and start really pushing our PSUs to their limits, to find out what those limits really are. (I assume that's the point of the test jig, right?)

As for the required load on the +5V line necessary in order for the switching PSU to regulate properly - well, I've heard that too. I can't explain why, exactly, but an EE relative told me that. Something about the voltage feedback circuitry.

 

[jonnyGURU]

Aaaak! It's you again!!!!

So Larry... What are you doing after work? Want to swing by Beef O'Brady's and get some wings?

Yeah.. The point is to push a bunch of different power supplies in an effort to make a relative comparison. Currerntly I have (other than the 300's I've put on it already) an Ultra X-Connect 500W, an X-Finity 500W, a Raid Max 520W and a Powmax 580W to load up and document. Maybe even an X-Finity 600W because I'm digging this Raid Max and might throw it in my main rig in place of the Ultra. I've got an Antec Neo-Power and Mad Dog Smart Cable coming so I can have some more modular cable PSU's to compare.

The work in progress will all be documented on my site and once all 6 or 7 power supplies are tested, I'll throw it up on Systemlogic.

I figured out what I'm going to do for the 3.3V and 5V load. I'm going to wire up one filament of an 1157 to the 5V and the other to the 3.3V. That should be enough load for the 12V to begin to regulate.

 

[Zepper]

You want to put a 20% load on the other positive rails than the one that is under test - that's the way Zippy recommends.
.bh.

 

[jonnyGURU]

Only 20%? Well hell. What happens if I put MORE than 20%? Because even 2 bulbs on the 5V and 3.3V is more than 20% of having all 8 of them lit up on the 12V!!

 

[jonnyGURU]

Oh.. and here's a link Eli over in OT dropped me to help explain the concept of loading other rails to regulate another: http://www.quepublishing.com/articles/article.asp?p=30273&seqNum=6 Pretty good read.

 

[Zepper]

Well, the AC side of the transformer limits the total output capacity of the PSU. So you have only so many Amps to deliver across all of the positive rails, the more you load the 3 and 5, the lower will be your limit on the +12 - may be inaccurate for testing purposes and give a falsely low value. Do you know the water analogy of electricity: Amps ~ volume, Volts ~ pressure, resistance ~ diameter of the pipe or other restrictions in the line. I think keeping that in mind can improve understanding.
.bh.

 

[jonnyGURU]

Well, I understand that concept. My dilema is "what's too high? What's too low?" I noticed that the more load I put on the 5V (so far) the better my 12V regulation has been (at least up to 9A. Still waiting for more 1157 sockets.) But, as you point out, if I put too much load on the 5V, I start taking power away from the 12V unnecessarily. The "pipe" is only so big around.

Now, your typical PC usually has about a 20A load on the 5V rail, so maybe it's important that I do load the 5V mroe than just 20%. This might actually make the test more realistic. Who cares if I can load the 12V rail to 28A if I can only load the 5V rail to 2A at the same time?

 

[Zepper]

Well, most switching PSUs have minimum load ratings for the + rails that should be posted on the mfr's web site. For the zippy 460, it is 2.5A on the 5V, 1A on 12 and 3.3, and 0.1A on 5VSB... And here are the words direct from zippy.
*** WHEN PERFORMING CROSS REGULATION TEST, IT IS REQUESTED TO SET THE HIGHER OUTPUT CHANNEL AT 90% MAXIMUM AND THE LOWER OUTPUT CHANNELS AT 20% MININUM OF RATED SPEC. I interpret that to mean that if you run the other rails at 20% capacity, you should be able to expect to get 90% of the rating out of the rail under test. IAC, I used the little Powmax PSU tester (all passive components so they couldn't screw it up too badly ) on those two Zippys I had here and that has under a 1A load on the +12 and +5 and the Voltage error on all 3 of the +rails was in the second decimal place by DMM... (e.g. 5.04, 3.34, 12.04)

.bh.

 

[jonnyGURU]

Hmm... I should expect 90% with the other rails at 20%, eh? I think I'll move in that direction.

 

[VirtualLarry]

Originally posted by: jonnyGURU
Oh.. and here's a link Eli over in OT dropped me to help explain the concept of loading other rails to regulate another: http://www.quepublishing.com/articles/article.asp?p=30273&seqNum=6 Pretty good read.

Wow, that's an excellent read, thanks for the link.

One characteristic of all switching-type power supplies is that they do not run without a load. This means that you must have the supply plugged into something drawing power for the supply to work. If you simply have the power supply on a bench with nothing plugged into it, either the supply burns up or its protection circuitry shuts it down. Most power supplies are protected from no-load operation and shut down automatically. Some of the cheap clone supplies, however, lack the protection circuit and relay. They are destroyed after a few seconds of no-load operation. A few power supplies have their own built-in load resistors, so they can run even though no normal load is plugged in.

The above is why I don't generally advise people to use the "paperclip trick" to power-on their ATX supply, since you generally don't know beforehand if it's a good supply or not, and one could be damaging it by doing so.

I think the part about the built-in load resistors is what you were alluding to in that other thread, when you were talking about the "continous power" feature of the X-Connects.

That feature could actually be highly useful for people using ATX PC PSUs to upgrade/replace the power-supplies used in older arcade cabs, and when converting into a MAME cab, stuff like that.

 

[jonnyGURU]

Originally posted by: VirtualLarry
The above is why I don't generally advise people to use the "paperclip trick" to power-on their ATX supply, since you generally don't know beforehand if it's a good supply or not, and one could be damaging it by doing so.

True. But think of HOW CHEAP that power supply must be to not even have simple dummy loads on them? I've done the paper clip trick to Codegens, Deer/Allied, Leadman. I've never had one not turn on and stay on and none suffered any detrimental effect. And apparently, this is a rule of thumb for any switching power supply, not just ATX. So this applies to AT power supplies as well. How many AT power supplies have you just walked up to and hit the power button on? Did any of those power supplies blow up soon after? I'm not saying the guy is full of crap. I'm just floored that in this day and age, there could still be a power supply out there that you could blow up by turning it on without a load. "Other cheaper clone supplies".... Where?

 

[jonnyGURU]

Ok. I've got one filament wired up to 3.3V pulling exactly 1A and the other filament wred up to 5V pulling 1.3A. Then I've got a Youngyear PSU tester hooked up (which was hooked up all along, but apparently didn't provide enough load to make much difference) which has two 15W/10 Ohm resistors in it (don't know how much load this produces) and now my 12V readings are looking a lot better.

This should make for some good reviews.

 

[jonnyGURU]

Ok. Apparently I got the attention of someone who knows a hell of a lot more than I do.

A broker in the industry whose brother engineers power supplies dropped me an email and basically told me, "I saw your rig. You look like you're having fun, but you're going about this all wrong. Do you want to go about it right? Call me."

So I did.

He told me that I need to evenly load the rails in order to get an accurate reading.

So do I do what Zepper suggested from a Zippy manual? Make sure the 3.3V and 5V are 10% of the 12V?

To test basic voltage stability, yes. But it's still not completely accurate because you have some power supplies that regulate more to the 12V than the 5V and visa versa. Too many variables come into play to just throw a solid number like "10" out there. At best, It's a good starting point. It's a good test for "is this power supply defective?" But not good to test if it's a decent unit.

For example, let's say you have a PSU that can do 34A on the 12V. That's 408 watts. That would mean you'd put only an 8A load on the 5V if you were to fully load the 12V. So now you're putting a "real world" load on the 5V and an unrealistic load on the 12V. The 12V may or may not go out of spec. The same is true the other way around. Say you have a PSU that only does 20A on the 12V and does 38A on the 5V. You load up the 12V to 240W and load up the 5V to 4.8A! Now your 12V is WAY out of spec because the power supply is obviously engineered to move most of it's power on the 5V rail!

So light bulbs are out. Too difficult and expensive to properly regulate.

I took a couple days off from work and drove up to his office where he showed me an actual PSU testing machine (SunMoon SM-268) and we did a series of tests on some different power supplies.

He had told me that the only "good" PSU test online was the one that Tom's hardware just did. But I was feeling a bit let down about that opinion because they only "benchmarked" efficiency at different loads, gave an aesthetic overview of the PSU and then concluded "none of them blew up." How is that a good review?

He told me that if you actually know how to use the test equipment and what a good "balanced" number for a power supply's given specs is, you can load a PSU up to 100% load for eight hours and not have a single problem. Tom's review was only trying to be "fair and balanced." Nobody isn't going to load a power supply to 100% load for any prolonged period of time. And all of the power supplies tested, ANY power supply that is UL listed, is GOING TO live up to it's label.

So what aspects of a power supply do you test? Features and efficiency.

So he's saying to buy a power supply based on features and the specs on the label and if efficiency is important, look at that as well?

Hmm.... Not good enough for me. I like blowing stuff up.

I sketched up three hypothetical systems. One with a high 5V (like a Socket A CPU on a motherboard without 12V CPU regulation, like a Biostar M7NCD Ultra 400, a couple hard drives, a couple optical drives, some USB devices, a few fans, etc.) and calculated a hypothetical full load. I then did the same for a similar system that regulates CPU voltage off of the 12V rail and then a super machine that smacked down on the 12V rail because of a Prescott and SLI, or an AMD64 with SLI and a four drive RAID array.

I showed off my numbers and got the nod that this would be a good idea because it would test full loads distributed correctly across the rails. If a power supply blew up, it wouldn't be because it was a crappy power supply, but because it was not the correct power supply for the application if worst case scenario, maximum load was every required.

We tried to knock a few tests out.... We did an OCZ, an Enermax, a PCP&C, Antec and Ultra X-Connect and X-Finity before we ran out of time (it was getting late and I had to get home so I could go to work in the morning.)

Much like THG's review, all of the aforementioned power supplies handled the full loads I hit them with as per their label... AND THEN SOME. Some characteristics are worth noting and that's what I'm doing here. ALL of these power supplies are VERY GOOD despite smell, heat and small puff of smoke. NONE of these power supplies should be frowned upon because in the end, they were ALL run OVER label specs, and like I said, none blew up.

Test rigs:

(Basic description, 12V/5V/3.3V. I won't get into exact specs because this post is already long enough.)

Socket A w/ 5V regulation, AGP, two HDD, 10A/30A/3A

Prescott, AGP, two HDD, 20A/20A/3A

Prescott w/ SLI and two HDD or A64 w/ SLI and four HDD, 30A/20A/3A


The OCZ ran well, but started getting really hot. At the heavy 12V load and heavy 5V load, the wattage meters on both the SunMoon and our AC Outlet analyzer started bouncing all over the place. He told me that was a sign of instability. The rails were within spec. Again, not a knock on the PSU. I don't see anyone putting a 30A load on the 12V rail or a 30A load on the 5V rail for such a prolonged period.


The Enermax was very stable. But it almost got hotter than any of the other power supplies (second to the Antec) and the whole office started to smell like burnt electronics. Despite this, it really did keep on trucking under load and the rails didn't deviate from spec. Only every once in a while would we see some instability and that was only while the 12V was up at 30A.


The PCP&C sat there like I didn't have a load on it. Seriously. The rails never fluctuated and it never got hot. I actually laughed out loud. Even though the power supply was 510W, it was doing 600W easy and the 12V was at 12, the 5V was at 5, etc. Too bad the power supply costs $225!


With the heavy 12V load (yes, the tester can do dual rails,) the Antec got really hot, but never smoked or stunk. When I say hot... I mean REALLY hot. There was hot air coming out of the back and when I picked up the PSU, the bottom of the PSU where the PCB is bolted was almost difficult to touch. Despite this, the rails were fairly stable. Too bad those cables are so fugly. At the heavy 5V load, the PSU was really stable, but then again the PSU has an unrealistically high 38A capability on the 5V rail. Why would you need that much juice on the 5V? I wish they would have put more on the 12V where we need it for today's PC's.


The X-Connect was the most stable on the heavy 12V, but had the biggest +/- going from idle to full load. Always within spec, though. The 5V test sent the 12V into a serious dip, but the Ultra is engineered with a high 12V for Prescotts and SLI, so that didn't surprise me. There was a slight smell, but not as bad as the Enermax. When we shut the machine off, I did so suddenly after the full load. I guess I learned that I need to gradually let the PSU down because the sudden stop of the fan caused a sudden concentration of heat, which resulted in a small puff of smoke. Now I can see why Raidmax's ECASO (Enhanced Cooling After System Off, where the fans spin for three minutes after power off is so important. After the PSU cooled back down, I fired it back up and it seemed to be OK.


The X-Finity 600W acted like a buffalo being beaten by a pine needle. Not as stable as the PCP&C, but only dropped from 12.1 to 11.72 going from 0 to 30A on the 12V rail. Being that we were only hitting it with a 477W load, I looked at my new friend and asked if it was OK if I tried to blow up the PSU. He nodded and with the 12V already at 30A, I started taking the 5V up from 20A. I don't remember exactly where the amperage was at for the 5V, but I remember seeing 700W on the load tester and just over 900W at the AC outlet when the lights went out in the office. I guess between the PSU and all of the other test equipment on the same outlet, we pushed the wiring in the building beyond IT'S capability and the breaker tripped. My friend then told me that that was to be expected because the X-Finity 600W is essentially two 400 or 450W power supplies mounted on top of each other and then "underclocked" to a total 600W so it didn't get too hot. Nice.


Anyhoo.... We ran out of time and I was going to leave. My friend unplugged all of the test equipment and asked if I wanted him to help me put the stuff in my car. HUH?!?!

He said back in Taiwan, they have racks loaded with these so they could test several units at a time. He could lend me one unit as long as I reported to him with any findings, problems, etc. So now I'm back at work with a SM-268 in a cardboard box sitting in my trunk. I will have to take the family to the fair on Saturday, but come Sunday I'll be hooking all of this stuff up and will continue testing power supplies.

If anyone has any questions or wants me to try anything, please ask! And Anand... if you want to do a PSU shoot out, let me know!