Questions on running dual PSU's

[TheGlasMan]

Thanks Mindless,don't think I'm going to get that hardcore!! But I'm always interested innards, so it was fun reading.
If I understand you, I just need a 5V load and that will keep the 12v in line.

The reason I mentioned the hard drives is that that would be the easiest way to create a load, just run the had drives off the secondary psu and problem might be solved, which I already noted wouldn't work with my drives. From crossload tests I've seen, about 3-5 amps is needed.

I mentioned the mother board draw, because it is the only place to get the draw with out adding an outside load. I could wire the 2nd psu in parallel (5v and maybe 3.3v, but I don't know enough to know if that would cause problems.

As you mentioned the resistor is probably the easiest.

Unless you spend top dollar for some decent Antec PSUs

Modern Antecs aren't separate like the old ones were. If they were I would use my Seasonic, who is making the current Antecs. Actually a few years ago a review said the SeaSonics were separate, but it isn't mentioned since.

[mindless1]

How many drives are you talking about, I mean the sum current requirement for the separate loads? For only a few hard drives that use only 12V input, you should be able to get a reasonably priced 12V only switching PSu. Many industrial PSU are rated far more conservatively than PC PSU, and especially if your drive controller is raid type with a staggered spinup function, you won't need more than 1-1.5A per drive (maybe even less, I'm not familiar with these 12V-only hard drives, but manufacturer power ratings on drives can vary per the manufacturer criteria, some are average active state ratings and others are peak momentary ratings).

If you do end up needing 3-5A load on 5V rail, it is significant enough you will want power resistors encased in an aluminum heatsink extrusion and to 'sink them to something (like thick part of the metal chassis or even better to a heat spreader plate which is then mounted with thermal interface to the chassis (or the back of a heatsink). You don't necessarily need to hit 12.00V though, if the 12V rail is down a little you may be fine, with a lesser load on the 5V rail... particularly if you aren't running anything from 5V from that PSU the other option even easier is to just get a PSU with a POT adjustment already built in, and with your 12V load attached just tweak the pot till the 12V is within an acceptible range and you won't care if the 5V is then too high (so long as it isn't tripping the overvoltage protection circuit yet), because this too high 5V range is not powering anything at all.

Do not wire two PSU in parallel to the same device(s), only their ground leads should be common to each other. By that I mean, don't wire each 5V rail to each other then to a load, but you could have separate PSU powering each different voltage to the board or other components, or each PSU supplying the same respective voltage rail but to different components, so long as they have a common ground.

They depend on their own feedback and inevitably, slightly different thresholds to regulate, and will fight against each other. A separate isolation stage and current sharing circuit would have to be devised (or some of it might already be integral to certain PSU such as those designed for redundant server applications). However, if you were to power the motherboard 5V rail from one PSU, and the motherboard 3.3V rail from another PSU (and the second PSU did not have independant regulation for the 3.3V rail as most midrange don't with exception of some Antecs), that 3.3V load is serving to pull down the 5V rail on the 2nd PSU some and reducing (if not eliminating) the need or value of current through an add-on power resistor dummy load.

One final thought - HOw can you know from crossload tests how much current you have to put on 5V rail when you haven't selected the PSU yet? There is some variation in this, particularly if the PSU is a newer model with a very high single 12V rail (or even higher split rail which implies a bit of peak rating for each) you might find you can put quite a load on the 12V rail before it droops too much. Plus, some circuits aren't all that picky about 12.0V power, for example the CPU VRM subcircuit might be quite happy running on 11V of _clean_ power, unless you're trying to overclock by quite a lot on a marginally designed board.

[TheGlasMan]

Thanks for the time you have taken with your posts mindless one. It is appreciated.

Don't worry about the drives, powering them isn't the problem. Sorry abut the red herring. I only mentioned because the were a potential solution that won't work.

Thank you very much about not wiring psu's together, that was the type of problem I was worried about. Unfortunately that also eliminates the usefulness of the idea, since graphics cards are what need the power (I planned to connect the 2nd psu to the auxillary connectors as needed), and they are partially supplied by the primary psu via the main connector. I don't see any way to draw enough power outside the board to make the 2nd psu useful. Darn.
You present the possibily of running the cpu's 12v connectors from psu 2 and the main connector from psu1, as well as the extra graphics adapters. Sounds scary, what do you think?

I think I mentioned earlier, but I'll re post it, I'm planning on running 2 K10's and up to 4 high end AMD/ATI graphics cards, assuming F@H finally gets a client ready for them. All will be overclocked, with luck quite a bit.
Since I will be starting with 1 gpu, I have time before making any decisions, and since new mega psu's are still being released, I may find the "perfect" one yet.

My idea of running the gpu auxiliary connectors from psu 2 would be about a 30-35 amp load, the cpu idea closer to 20 amps, much less. the 3-5 amp # was ball park, it is where jonnyguru* crossload tests, and on some psu's problems begin there, all start to vary. And yes I would get the best regulated one.

What do you think of my new idea, is the cpu power circuit sufficiently isolated from the rest of the board?

Again thank you for the time you have taken.


*example, I read many other good sources, including here

[mindless1]

Thanks for the time you have taken with your posts mindless one. It is appreciated.

Don't worry about the drives, powering them isn't the problem. Sorry abut the red herring. I only mentioned because the were a potential solution that won't work.

Thank you very much about not wiring psu's together, that was the type of problem I was worried about. Unfortunately that also eliminates the usefulness of the idea, since graphics cards are what need the power (I planned to connect the 2nd psu to the auxillary connectors as needed), and they are partially supplied by the primary psu via the main connector. I don't see any way to draw enough power outside the board to make the 2nd psu useful. Darn.

I must not understand your project, thought it was a super-folding box which has no need for externally powered video cards (even if you had them as spare cards, it would be better to buy an ultra efficient *GPU* like ATI Rage Mobility, they're about $0-15 per depending on who has some they're getting rid of. If you linked to a concise complete list of the configuration of this box, it might make the overall power scheme easier to overview.

You present the possibily of running the cpu's 12v connectors from psu 2 and the main connector from psu1, as well as the extra graphics adapters. Sounds scary, what do you think?

I said nothing about the graphics because I didn't think think such a card would be appropriate unless you're gaming with the system. There is no problem using a different PSU for the mainboard ATX connector and the CPU 12V connector, so long as there is no continuity between the two connectors' positive pins. In the vast majority of motherboards this is the case, but to rule out rare exceptions you could try powering on the system without the 12V connector, connected... if it runs like that there is definintely continuity and two PSU couldn't be used in this arrangment (though safest would be to just use a multimeter and measure continuity, being sure you're measuring the positive pins since there should be continuity between ground (there definitely will be) but not between the positive 12V pins of these opposing connectors.

In short, you can power anything from a separate supply per each voltage, so long as there is no loop connecting the two components power rails together. For example when a hard drive is powered by a different PSU, it still does have an electrical bus connection to the mainboard, same as multilple parts on the mainboard do, but a separate power regulation stage just as the CPU does on most boards with the 12V connector.

I think I mentioned earlier, but I'll re post it, I'm planning on running 2 K10's and up to 4 high end AMD/ATI graphics cards, assuming F@H finally gets a client ready for them. All will be overclocked, with luck quite a bit.
Since I will be starting with 1 gpu, I have time before making any decisions, and since new mega psu's are still being released, I may find the "perfect" one yet.

Are you able to use the GPUs' processing power to fold? I overlooked that possibility with my prior comment but if not, I would not use higher powered video cards. The best split of power would be as I've just mentioned, check continuity between the CPUs' 12V connector and the ATX connector's 12V pin, and if no continuity, power the CPU from one PSU and the mainboard plus video card from another. Some video cards also have independant isolated power from the mainboard, and likewise you could check continuity on it's power pins to determine this, but also keep in mind that the more power sharing there is going on, the more beefy you will want the two PSUs' grounding to each other to be to minimize ground current differences in potential.

My idea of running the gpu auxiliary connectors from psu 2 would be about a 30-35 amp load, the cpu idea closer to 20 amps, much less. the 3-5 amp # was ball park, it is where jonnyguru* crossload tests, and on some psu's problems begin there, all start to vary. And yes I would get the best regulated one.

IF these figures do end up being roughly accurate, I would use added dummy loads on the 5V rail, OR just get a 12V-only industrial PSU instead (again tying it's ground to the other PSU ground).

What do you think of my new idea, is the cpu power circuit sufficiently isolated from the rest of the board?

Again thank you for the time you have taken.


*example, I read many other good sources, including here

It is a rare situation when the CPU 12V power isn't isolated, I don't think there is any semi-modern board where it isn't isolated because that shared, non-isolated scheme can cause problems with modern processors opposed to early low speed, higher voltage and lower current P4 and Athlons a few years ago. Really it was just a few boards at the very beginning of implementing the 12V CPU connector that didn't have it isolated, but certainly instead of relying on the high odds it's isolated, the conservative test is the continuity check with a multimeter.

Remember that for each power rail suppling a lot of current, you should have roughly same ground return wiring. In a predesigned power delivery from one PSU, this is usually true but if you were to start grafting various PSu together it is more significant to remember that for each fairly loaded lead from one PSU, there needs to be a similar low impedance common ground between all parts including both PSU.

[MrNovi]

I must not understand your project, thought it was a super-folding box which has no need for externally powered video cards (even if you had them as spare cards, it would be better to buy an ultra efficient *GPU* like ATI Rage Mobility, they're about $0-15 per depending on who has some they're getting rid of.

From I can tell based on this comment of

and up to 4 high end AMD/ATI graphics cards, assuming F@H finally gets a client ready for them.

that he will be GPU folding which would really cause these vid cards to draw a lot of juice. I'm guessing he's referring to the newer X29xx cards.

[TheGlasMan]

Sorry thought, that was understood, Doh!!!
A couple of them for gaming now and then, natch.
I hope for a better performance/power ratio by then, March?, but I don't expect any power reduction. Back up would be 1950xt(x) and they take a bit as well.

[ivanmmj]

Any details how it killed it, and how did you rig it?

Dunno, but it blew a contact on the molex connector on the 9800pro. It looked like someone set off a miniature firecracker on it.

I only had it wired to always be on (using the power switch in the back to turn it on.) It was only plug into my 7900GS. It's 5v rail looked horrible when I tested it, and the 12V rail was running more than it should've. I don't know why...

[mindless1]

The problem with using a different PSu for a video card than for the motherboard is when the motherboard is also supplying that voltage rail to the card, in which case the (somewhat) higher voltage PSU of the two will case this voltage difference X current used by all devices using that rail, to flow across that slot, card, and connector. The weakest link will give away first from the heat, and the PSU with higher voltage will be supplying more current while the other PSU will keep trying to re-regulate down to a lower voltage than it's sensing.

Thus, the ground connection between the two PSU has to be good, and there can't be any rail loops causing the rail of one PSU to be continuous to the rail of the other.