Passive computer power supply questions

[dannyf]

Passive computer power supply

"Passive power supply"? You probably have to build it using passive transistors and passive ICs?

[gxti]

"Passive power supply"? You probably have to build it using passive transistors and passive ICs?

Your joke has already been sniped by Richard Crowley. Better luck next time. The grouch battle is over, but the grouch war is yet to be decided -- don't give up hope!

As for the technical constraints, I don't see any reason why this wouldn't work if you balance the power consumption appropriately. I don't think the concerns about PCI-e vs auxilliary power are well-founded, because after all that is already two separate routes from which it draws power. It has to deal with differences in resistance etc. not to mention multi-rail power supplies. But I do tend to agree with those who are saying that concept overall is misguided. The screaming from my GPU and CPU coolers has always drowned out anything from the PSU.

[akis]

As for the technical constraints, I don't see any reason why this wouldn't work if you balance the power consumption appropriately. I don't think the concerns about PCI-e vs auxilliary power are well-founded, because after all that is already two separate routes from which it draws power. It has to deal with differences in resistance etc. not to mention multi-rail power supplies.

I am not sure I agree. Whatever comes out of the same enclosure PSU , regardless what we call it, multi rail, single rail or whatever, HAS to account for shorts between its own rails of the same "nominal" voltage. So if there are 2 12V connectors coming out of the same PSU, I, as a user, have to be assured, that I can connect then together (in parallel) without blowing out the PSU. In other words I , as a user, do not care how the PSU is made internally, and whether the various 12V connectors feed into the same or completely separate circuits/subcircuits inside.

But noone makes a guarantee that two 12V connectors coming out of separate PSUs will also behave like that. It would be like paralleling batteries or any other kind of PSU and the only way to do it reliably is via (Schottky) diodes and an assured negative current coefficient, so that as one PSU gives out more current, its voltage drops, and then the other PSU takes over so in the end they balance out sharing the load.

And we cannot depend on the internals of a computer, with hard drives, optical drives, expansion cards and whatever else, that if we feed half and half from separate supplies, there will not be paths between the 12V rails. Because as far as the motherboard and busses are concerned there is one 12V rail, not multiple. How would a GPU separate the 12V supplies coming out of the back connectors and the PCI slot without wasting power? And why on earth would it go into the trouble of the additional components required.

 But I do tend to agree with those who are saying that concept overall is misguided. The screaming from my GPU and CPU coolers has always drowned out anything from the PSU.

I thought exactly the same but then noticed that the OP has mentioned somewhere (he was not explicit at the start) that he uses passive cooling for his CPU and GPU. I presume that his PSU is the last bastion of sound pollution inside his case.

[extide]

Isn't that what the AddPSU unit does?

That would require the second unit to be an ATX psu as well. If you went with some sort of industrial 12v only psu, then the AddPSU thing wont work, as they rely on the ATX turn on wire.


Also, most high end video cards actually do NOT pull power from the slot, they are designed to pull it entirely from the PCIe connectors. There are some exceptions to this, but I am fairly sure the 970 and 980's are like this, at least the ref PCB ones.

[RichardF]

Its seems there isn't consensus about whether or not it is safe to try to use a separate PSU to just power the GPUs, via the PCIe connectors. These posts indicate that even high performance GPUs may draw power from both the slot and the connectors:

Nvidia have incorporated a 4 phase power supply with overvoltaging capability. This is given the task of feeding the GM204 GPU with power. An extra power phase is dedicated to delivering power to the GDDR5 memory.

The Geforce GTX980 runs at higher clock frequencies than any other reference Nvidia card before it. They have incorporated integrated dynamic power balancing circuitry to ensure the GPU gets the power delivery evenly across all three power sources (each 6 pin connector and the PCI e slot).

With the dynamic power balancing system in place, the card will direct power from each connector when needed. If one power source is at the limit, the card will draw power from the others to balance it out. This is a feature exclusive to the GTX 980 and GTX 780 Ti.

link

For this test, we measure the power consumption of only the graphics card via the PCI-Express power connector(s) and PCI-Express bus slot. A Keithley Integra 2700 digital multimeter with 6.5-digit resolution is used for all measurements. Again, the values here only reflect the card's power consumption as measured at its DC inputs, not that of the whole system.

MSI GTX 780 Ti Lightning is based on the same new-generation TriFrozr cooling solution as the GTX 780 Lightning. Its underlying PCB is unchanged, too. The card draws power from a pair of 8-pin PCIe power connectors, and conditions it using a gargantuan 20-phase VRM. The VRM uses two separate controllers for GPU and memory+PLL. The GPU and memory both draw power from the power connectors, while only the PLL draws power from the slot. This way, memory OC isn't held back by the PCIe slot power supply.

[RichardF]

there will come a point further increases in fan speed = diminished returns. so alot more noise but not really alot more heat removed

This can be seen in many ATX power supply noise/fan speed/wattage load graphs. Many newer units will not even run a fan until they're loaded above 250 watts or so (something around 400 for the Corsair RM 750 & 850). At a certain load rans will rapidly ramp up, greatly reducing the efficiency of the cooling in terms of noise generated. Some go above 50 decibels. Anandtech said the Seasonic SS 1050 could be heard "rooms away".

My take on the charts is that the ATX form factor, even when lengthened a bit, is not designed well for handling such high wattages quietly. I think much bigger form factor would help, especially if waste heat would be exhausted out of the case. But, since nothing like that yet exists in the marketplace, I thought about the possibility of using two fanless power supplies rather than just a single power supply. The dual sourcing of power (PCIe slot and connectors) for graphics cards seems to be the biggest obstacle.

While figuring out how to put them in an enclosure with adequate cooling is one issue, the one that concerns me more is danger to the equipment and possibly having "dirty" power that causes instability or stress on the equipment. I'm not nearly as concerned about cooling strategy because I don't have significant space limitations. I would like to have them in an enclosure to reduce EMI and dust exposure, but the passive cases I've seen are just cases with a lot of holes in them. That doesn't seem to be terribly complicated. Plus, I can always add a large very slow-moving fan to supplement the convection.

so imo, its a matter of good PSU (or rather thermal dissipation) design. and at that, my take is use a oversized PSU so it never have to ramp up its fan if noise is such an important factor.

That was my reasoning behind a double-wide PSU or something similar. It would have twice the surface area for cooling. Since 520 watt fanless models are offered from reliable vendors, then why can't doubling the space given to cooling make it possible to increase a fanless PSU to, say, 750 watts? Or, if you want a higher wattage then put something like a 200mm fan on it and run it at a low rate.

come to think of it ... almost all fancy heatsinks in the market are of low thermal efficiency ... because almost all of them are so glossy like a mirror  . i also tried painting an entire PCB (not motherboard, just a test PCB to see thermal spread) black, between the unpainted and painted, there is actually quite a noticeable difference too ...

Is this because the paint provides more surface area? If so, then would matte paint with grit in it (silica or something) be even better? At what point, though, does the paint become a heat insulator?

[ajb]

Its seems there isn't consensus about whether or not it is safe to try to use a separate PSU to just power the GPUs, via the PCIe connectors. These posts indicate that even high performance GPUs may draw power from both the slot and the connectors:

The concern isn't that a card may draw power from two or three sources, the concern is that the card may directly connect those sources together, which can cause currents to flow between those sources through the card.  If the card described in your quote can independently regulate the the portion of it's operating power that comes from each source, that suggests that the different sources supply different phases of the main buck converter(s).  They mention five phases, so probably two come from each PCIe aux connector and the fifth comes from the PCIe slot.  Using a polyphase converter to balance power draw from multiple sources is interesting, I'd be curious to see that circuit. 

It's pretty trivial to check if any given card ties its power sources together, just need to measure continuity among the 12V inputs.

While figuring out how to put them in an enclosure with adequate cooling is one issue, the one that concerns me more is danger to the equipment and possibly having "dirty" power that causes instability or stress on the equipment.

"Dirty power" is a pretty squishy term.  Ripple on two supplies won't be any worse than on one supply.  Under/overvoltage is not permitted by the ATX spec, so shouldn't be a problem on a decent supply.  Both supplies will be required to meet the hazardous power limits of the ATX spec, so should shut down on overcurrent.  I think as long as you use two high quality, ATX-compliant power supplies, suitably bond the DC grounds together, and ensure that all of the positive rails are mutually isolated, the only real concern I would have from looking at the ATX spec is that whatever mechanism you use to tie the power on/off of the two supplies together doesn't violate the risetime requirements, and ensures that if one power supply shuts down the other immediate follows suit.  I don't think the little gizmo you linked on Amazon would do that.  If I were to do this myself I'd probably build an adapter that ANDs the PWR_OK signals to the motherboard, and routes the PS_ON# signal to both PSUs via some logic that immediately deasserts PS_ON# if either PSU misbehaves.

[RichardF]

The concern isn't that a card may draw power from two or three sources, the concern is that the card may directly connect those sources together, which can cause currents to flow between those sources through the card.

That's what I was referring to. That's why the issue of whether or not a card is going to draw from both the slot and the connectors seems critical.

It's pretty trivial to check if any given card ties its power sources together, just need to measure continuity among the 12V inputs.

From the little I've read, it seems like the slot may also load the 3.3 rail.

suitably bond the DC grounds together

How would I do that?

and ensure that all of the positive rails are mutually isolated

Does this require an additional modification?

I don't think the little gizmo you linked on Amazon would do that.

That's unfortunate. It seems to me that if someone is going to market a device for tying ATX supplies together they should cover all the important bases.

If I were to do this myself I'd probably build an adapter that ANDs the PWR_OK signals to the motherboard, and routes the PS_ON# signal to both PSUs via some logic that immediately deasserts PS_ON# if either PSU misbehaves.

It seems I'm out of luck, then. The one friend I have who has this background is back in Finland.

Thanks for the info.

[extide]

The concern isn't that a card may draw power from two or three sources, the concern is that the card may directly connect those sources together, which can cause currents to flow between those sources through the card.  If the card described in your quote can independently regulate the the portion of it's operating power that comes from each source, that suggests that the different sources supply different phases of the main buck converter(s).

Yes, this is exactly the issue right here.

[ajb]

The concern isn't that a card may draw power from two or three sources, the concern is that the card may directly connect those sources together, which can cause currents to flow between those sources through the card.

That's what I was referring to. That's why the issue of whether or not a card is going to draw from both the slot and the connectors seems critical.

Except it's possible for a device to draw power from more than one source without tying those sources together.  This is a necessary condition of being able to independently regulate the amount of power drawn from each source.  From the description of the cards that you quoted, manufacturers of high end graphics cards are accomplishing this by splitting the sources among separate phases of a polyphase buck converter, which means the sources are isolated from one another by their own dedicated switches and inductors.

From the little I've read, it seems like the slot may also load the 3.3 rail.

As long as the slot is the only source of 3.3V available to the card, then it's no concern.

suitably bond the DC grounds together

How would I do that?

Since you'd need to interface to both PSUs' ATX connectors to get proper startup/shutdown behavior, probably build your interface gizmo with two 24 pin ATX headers and have all of the 24 pin connector's grounds tied together there.  The supply rails from ONE PSU would head out to the motherboard from there, along with the PS_ON and PWR_GOOD lines from the interface logic. 

and ensure that all of the positive rails are mutually isolated

Does this require an additional modification?

Nope, just proper connections to your peripherals, as we've been discussing in this thread.

[RichardF]

As long as the slot is the only source of 3.3V available to the card, then it's no concern.

I seem to recall reading that the slot provides both 3.3V and 12V power. Unless that 12V power is handled by a separate phase it seems like there could be issues with a second PSU.

Since you'd need to interface to both PSUs' ATX connectors to get proper startup/shutdown behavior, probably build your interface gizmo with two 24 pin ATX headers and have all of the 24 pin connector's grounds tied together there.  The supply rails from ONE PSU would head out to the motherboard from there, along with the PS_ON and PWR_GOOD lines from the interface logic.

and ensure that all of the positive rails are mutually isolated

Is there a way to modify the AddPSU device to accomplish this since it already handles the startup/shutdown issue? I am looking for a solution that involves as little DIY soldering and the like as possible. It is not my area of expertise.

Nope, just proper connections to your peripherals, as we've been discussing in this thread.

Proper meaning what? The big potential problem appears to be the GPUs due to the power being provided by the slot. If GPUs are getting both 3.3V and 12V power from the slot, is it still OK to power them via another PSU?

[ajb]

Proper meaning make sure none of your peripherals will directly tie a supply rail from one power supply to a supply rail on another.

The AddPSU is nothing but a 4 pin drive connector, a 24 pin ATX connector, and a relay.  It's a far cry from what you'd need to ensure proper startup/shutown behavior with any confidence of staying within the ATX spec's timing requirements.  At a minimum you need a handful of transistors or a logic gate or two to implement proper startup/shutdown behavior.  If you wanted to be defensive you'd add a window comparator for every supply rail.   

The fact that the PCIe slot provides 3V3 is no concern because there's no other source of 3.3V available to the card, so there's no chance that the card may connect one PSU's 3.3V rail to another PSU's 3.3V rail.  The fact that the slot provides 12V IS a concern because the PCIe Aux power connectors also provide 12V, so if you have the motherboard (and hence PCIe slot) coming from one PSU and the Aux connectors coming from the other PSU, it would be possible for the card to connect all of those 12V inputs together, which could mean that you have the 12V rails from two different power supplies shorted together.  HOWEVER, it seems (as shown in links you've provided) that high power video cards typically segregate the 12V supplies from the slot and the PCIe Aux connectors onto different phases, so as long as your card does that (you can check with a multimeter) there's no concern there.

[RichardF]

I appreciate the information and clarifications. 

Now, all we need is for someone to make a better product than AddPSU and sell it on Amazon.