Hi,

Had a Corsair 1200AX for a couple of months. Ran fine until it died.

Now I'm running my i7-3930K + 2 GTX780's on an ASUS P9X79-WS and I have an E-version of the motherboard on the desk next to me.

The current system is powered by my old AX650.

Just got an E-mail saying the supplier is out of replacement 1200AX's and they'll return my money.

All I need is (two additional GTX780's and)
a recommendation for a new PSU.

I'm considering a 1250W Seasonic model X-1250 from here. I'd get one under 300€.

What should I take in to consideration?
- Single rail amperage
- enough cables for GPUs
- warranty
- ...

I get 2300W from any socket in the wall and the 3x25A 400V~ metering box is a few meters away.
--
Petri

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

I use an Enermax 1350W with this and have no problem.
http://setiathome.berkeley.edu/show_host_detail.php?hostid=7189954

---

I use a 1350W PSU in my 2x690 host, so in theory a gold 1250W is enought to drive 2x780 (2x250W each) + your high capacity MB.

My advice is: if the diference is not so big 1350W vs 1250W use a 1350W, specialy because you pull very hard (OC) from your host as we could see by your RAC.

And BTW actualy your old 1200W PSU already died, so that give you a clue.

---

Hey, wait a minute, i just see now, you are talking to drive 4 x 780 not 2.

You should not use a 1250W or even a 1350W PSU to do that.

The best choice (and the cheaper) in this case is to use 2 PSU instead of one.

Each 780 fully loaded (as we expect when do real crunching) uses up to 250W so is 1 kW just for the GPUs, add to that 300-500W for a fully loaded MB and peripherals, that´s give you 1300-1500W.

But PSU are not perfect they have looses, if you have a very good PSU it´s eficience is in the range of 90% plus a safe margin due the PSU components degradation with the time, you get a 1800-2200PSU, something not common.

So it´s better to use 2 PSU, one for the MB and 2 x 780 (could be the 1350W i mention before, or 1250W if you have few peripherals on it) and one for the other 2 (a 750/850W it´s a good choice for that), all no less than gold and single rail.

Of course all this is considering you want your PSU runs for a long time in a 24/7 cicle.

There is a non writing roule about PSU´s who will be used in a heavy dutty 24/7 cicle, you could use up to 70% of the Total avaiable power, example in a 1000W you could use 700W constantly and you PSU will work for years, more than that and your PSU could work but for how long?

---

So what else is on this branch of the mains? Refrigerator, clothes dryer,
television?

So what else is on this branch of the mains? Refrigerator, clothes dryer,
television?

In my computer room that get its juice through one 16A fuse (230V) there is
a) a laptop 100W
b) my crunching machine now about 650W + LCD display.
Lightning in that room has its own line and fuse.

The over 3.5 kW that I get from the wall is enough for my computing needs.
I know the line is shared with some other room but it is a room with a low power need.

The computer room, the laundry/clothes management room and the sauna department (with a swimming pool now converted to a home theatre room) are in same floor and probably have one of the three phases for its own use. The other floor has kitchen, living room and bedrooms and they each probably take juice from the three different phases of the line.

I have not checked the wiring plan, but my guess is that the house is divided in well balanced segments using all 3 phases evenly.

The biggest electrical gadgets in the house are an electric sauna 6 kW and a 15 kW (output) thermal pump that takes max 4.8 kW in (9 kW backup heating element switched off)). The use all three phases and so does the electric kitchen stove too. They all have 3x16 or 3x20A fuses.

The main fuse is 3x25, 3x36 or 3x63 A. I don't want to go and check.
EDIT: 3x63A just checked.

No brownouts, no burnt fuses. Everything works.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

Thank You all for the replies.

I'll google the mentioned brands/models and make my own judgement regarding to the needed wattage.

I've got a hunch that running Seti does no use that much power for GPU work, since not all parts of the CPU are used. Gaming might be a different story.

My current system is powered by an AX650.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

My current system is powered by an AX650.

A 2x780+MB i7-3930K host running highly OC on a AX650 something not fit. Nothing related if you use only for Gaming or even worst if you do crunching

BY NVidia specs (very conservative BTW) the minimum PSU to drive a single 780 running at stock speed is 600W, and uses 42A from the 12V rail. By the AX650 specs the maximum 12V capacity is about 62A. So if you try to run 2x780+MB on a single AX650 you will overload the PSU. Simple like that, the AX650 have no current capacity to drive even the 2x780 at stock speeds, now imagine when you add OC and a high capacity MB. Sorry but what you have in this situation is a time bomb not a PSU.

As i said before, overloaded PSU´s could even work, the question is for how long. For a stable and healty 2x780 hightly OC host use nothing less than a 1200W PSU, now for 4x78o host... go for a 2xPSU dessing.

---

One thing you have to remember is that, because CUDA processing does not use the full video engine it derates the power reqirements of the card somewhat.

e.g. I was able to run 3 x GTX580s + o/c'd QX9650 + 4GB memory with an 850W power supply for more than 6 months.

I estimate that when using the cards solely for crunching, it drops the power requirements by at least 20 to 25%

T.A.

Yes you could...i even drive 2x580 or a 690+670 on a 750W PSU either, but not for years, simply the PSU internal component have a life cicle (specialy the capacitors/inductors) so they degradate when the time pass, add that happening even faster when you use more power from your PSU due the high currents. Unfurtunately there are some roules like the ohm law we can´t break.

This host for example has a 780FTW and runs only 2 MB WU at a time at stocks speed, and it´s GPU ussing right now about 75% of the TDP or about 190W, with 3 it reaches 90% of total power, so it´s a urban mith who say CUDA does not use a lot of power, it uses, specialy if you OC the GPU.

Of course as allways YMMV and each one has his own roules, what i try to explain is a safe way to determinate the PSU needed for a 24/7 host who will crunch with no interruption for years, not just months.

There are some very interesting calculator avaiable to do that, for an example of one take a look at:

http://www.thermaltake.outervision.com/

it´s not perfect (it´s still a little conservative - to be safe you need to add > 25% of safe margin to it´s calculations, look at the notes at the end of the page) but gives you a good path to follow. Try it and change the capacitor aging, and you will see what i talk about.

---

BY NVidia specs (very conservative BTW) the minimum PSU to drive a single 780 running at stock speed is 600W, and uses 42A from the 12V rail. By the AX650 specs the maximum 12V capacity is about 62A. So if you try to run 2x780+MB on a single AX650 you will overload the PSU.

Be careful that you're not confusing system power with GPU power. I just got a GT 640 and the box says "Min. power req. 300W", but that's for the whole system; The Wiki page gives the TDP as 49 W, as did another source (perhaps even drilling down on Nvidia's site, I don't recall). The aforementioned Wiki page gives the 780's TDP as 250 W.

---

Yes, it all seems to work, until it doesn't work.
I asked because a friend had no problems with his computer until he
upgraded to a top end, gtx 690 card, then it would blue screen occasionally.

Short story was his freezer would kick in and draw the voltage down long
enough, or cause an inductive spike, or both, to cause his computer to
crap out. Switching to a different mains circuit cured his problem.

Just something to keep in mind.

I probably didn't have-to, but I have already rigged one computer with two power supplies and have the cases to do it with two more (if I can find the time).

The cost of the case to accomplish that is less than the difference between the price of two 850w, two 750w, or two 950w PSUs compared to those mammoth >1250w PSUs. ...not to mention that two 850w PSUs are running more "efficiently" than one 1250w running full-out.

I'd love to have a nice big $700 power supply. I'd just rather spend $300 and also have more power available.

And many of the cases that will let you do over/under power supplies are huge, with big fans everywhere, so they also help cure the heat problem that four GPUs in one rig produce.

It may not be the best solution, but that's the way I'm going.

Let´s see from another point, if each 780 uses up to 42A to run, how a 62A PSU could safely drive 2 of them + plus the MB itself? Simple ohm law.

As i said, could work for some time, but the point is, if you spend 1000´s of $ in high capacity GPU´s, MB, etc. and do a lot of OC on them, why not spend a little more in a PSU who could safely drive your hosts for years?

A 690 uses up to 300W, the 780 up to 250W, not sure about if the TI uses the same 250W too., watts of electric power (actualy the right way to measure must be in VA not watts) thats comes from the manufacter specifications, when you crunch sometimes you use a lot less power (IE if your AP have too much blanks, MB with very low AR, etc) but your PSU must have the capacity to drive the GPU at 100% or some wierd things could happening without any further notice, just look how many complains about that are in the forums. Not confuse this with the TDP who is used normaly to dessing the cooling system.

Other important thing, when you measure your power drained from the input of the PSU you measure mean power (humans timing), not high frequency spike power used by GPU´s(computer timing)so you could see 150Watts in you multimeter or watt-o-metter for example but the actual peak power could be easely 1.5x that, an that spikes of current must be supplied by your PSU too. Few have a instrument who could realy measure this kind of current spikes.

Maybe some other guy (Jason are you there?) could explain that better for you all, I just try to show a safe way to choose the PSU of a big cruncher who could last for years.

BTW I just see today a host with a 780 using a common 500W PSU who don´t even have the VGA power conectors and works perfect from the user side, but i´m sure the molex connectors will be melted in months if not weeks.

---

...
Power supply efficiency has been mentioned. That should be taken into consideration also.

Power supply efficiency? If a 1000 watt power supply has an efficiency of 90%, then *it* is consuming 100 watts to make the power supply conversion (and dissipating it in heat) and will provide 900 Watts of power at its output.
...

No, power supply watt ratings are on the output. When running at 90% efficiency and providing 1000 watts of output, the power supply would be drawing 1111.1... watts of AC power and therefore dissipating 111.1 watts as heat.

                                                                   Joe

if each 780 uses up to 42A to run

Doubt that, that would be 504W. No card is using even nearly that much, that must be nVidia's recommendation for the whole system.

There's an easy method to figure out the maximum power consumption for a graphics card:
- for the PCI-E slot you count 75W
- for each 6-pin PCI-Express power connector you add 75W
- for each 8-pin PCI-Express power connector you add 150W

Fow lower end GPUs, specially for those without any PCI-Express power connectors, you even should not need to add any safety margin, for high end GPUs add maybe 20%. So for example if you have a 300W GPU count with 360W.

---

...
Power supply efficiency has been mentioned. That should be taken into consideration also.

Power supply efficiency? If a 1000 watt power supply has an efficiency of 90%, then *it* is consuming 100 watts to make the power supply conversion (and dissipating it in heat) and will provide 900 Watts of power at its output.
...

No, power supply watt ratings are on the output. When running at 90% efficiency and providing 1000 watts of output, the power supply would be drawing 1111.1... watts of AC power and therefore dissipating 111.1 watts as heat.

                                                                   Joe

I recall a couple of unexpected concerns that came up, back in the day when Steve (ScimanSteve) built his 2 x GTX480 rig. Probably a long anecdote & Steve may remember some things differently than I do, though it went something like this.

One point first, some manufacturers weren't then, I'm guessing aren't now, entirely honest with using the convention of output power like most these days. I think one particular manufacturer involved at the time, while figuring out Steve's machine, might be out of business, or at least not making PSUs anymore.

Because the 480's were such a huge and new draw, to try to understand, large PSUs not as readily available, there was some deeper analysis involved.

Firstly something was affecting overall system stability, but initially we weren't sure how. A fair bit became clearer once I installed my own 480, and managed to toast a 700Watt Seasonic Gold PSU in a short space of time.

On replacing that PSU with a large corsair gold (future proofing and overkill included), I noted the inclusion of chokes on the PCIe power leads. Curious, the penny didn't completely drop until after looking at the 12V rails with an oscilloscope under idle and load, with and without the chokes in place.

Taking copious readings, it worked out that the chained efficiencies of the PSU, GPUs, and PSU->Motherboard->GPU each have a current slew, dependant on their quality. The efficiencies multiply, and the slews and various back EMFs from the buck converters generate some amount of AC on the 12V.

The short of it was, In the end it worked out that better gear, with some decent amount of headroom, had nearly no ripple at all. Ultimately we pay for, and expect with DC, Watts (real power), while the sizing really needs accommodate a highly variable amount of reactive power.

In these contexts, probably the trends to recommend reasonable headroom and spec out from current specs as a minimum, (instead of Watts as nominal), are a personal risk choice to some degree. For each pair of 780's I personally wouldn't be putting anything less than a really good 100Amp single rail 1200W PSU, even though the RMS draw would likely be less than half that. It's the coincident peak current draws that aren't specified for multi-gpu, and resulting required effective slew.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Hey, wait a minute, i just see now, you are talking to drive 4 x 780 not 2.

You should not use a 1250W or even a 1350W PSU to do that.

I do and it works, for now it is good enough.

---