Hello, do computer power supplies always draw as much wattage as they are rated for? ie: does a 750 watt power supply draw 750 watt while sitting idle in Windows?

I am building a new computer with a 750 watt power supply, but I don't want to waste a ton of electricity (and hurt my elect bills) over its life span. I'm oversizing the psu because my current machine suffered greatly from a weak 350 watt psu (hdd randomly powering off, random power cycling).

Or does wattage even matter toward running up my elect bills? Should I be concerned with amperage and voltage? Oh wait, if amperage is "the size of the pipe" and voltage is constant, will I even be able to get enough wattage to power the machine?




You may ask why I chose 750 watt. The machine has an SLI motherboard anticipating two gpus in the future. For now it has an nVidia 8800 GT 512, which is pretty power hungry. Also it has a 68 watt low power Athlon 5200+, two (IDE) to three hard drives, and a DVDR (SATA). I don't ever EVER want my hard drives to randomly spin down then back up while using the computer, which happened frighteningly often with my current crappy 350 watt psu.

750W means it is capable of supplying something near that number. It doesn't draw it constantly. Get one that is high efficiency.

Manufacturers mostly lie when they're advertising power supplies. They add up what each of the various rails can supply individually and not what the power supply can do with a load on all the rails.

No.

Most desktops (average video card, one/two hard drives, etc..) run in steady state at perhaps 140W or so, and that's what the PSU draws.

There are PSU differences, though. Any new PSU should really have at least an 80% efficiency rating -- check it before you buy.

The real trick is finding a PSU that is 80+% efficient in the typical 140-200W range. Having a 750W PSU that is 80% efficient at 600W is useless, and is also quite typical of such over-spec'd units.

Another trick the makers/sellers use, is to publish efficiency ratings that are only valid when run on 220-240VAC, with no comparable numbers for 120VAC. (The higher the input voltage, the more efficient the PSU, to a degree).

Cheers

Thanks for the confirmation. I've read articles at <URL="http://www.anandtech.com">Anandtech</A> which rather scientifically monitored power consumption. They say a system like mine will run 146 Watts at idle and 269 Watts under load (speaking mainly of the CPU and video card). They say that with two cards in SLI, they've measured 211W at idle and 327W under load. I want to be ready for that.

So, if a "750W" PSU doesn't draw that all the time, than at least it's ready for whatever load I throw at it. No harm in oversizing it, except cost, which is negligible compared to the damage done to my hard drives from under powering them.

Actually, there *is* some harm potential in oversizing it by that much: it is unlikely to be as efficient in the 140-200W range as a lesser rated PSU could be. This means that you'll be contributing more to global warming than need be -- quite wasteful, given a choice.

Another advantage to choosing efficiency, is that the PSU should run cooler (using less energy), and therefore may also run *quieter* (lower fan speed).

Cheers

For anecdotal evidence of PSU sizing, I have here a 2.4GHz Intel Quad-Core box, 4GB of PC2-6400 RAM, on an Asus P5B-VM motherboard. It has five 7200rpm SATA drives installed, a PCIe SATA add-in controller, a floppy, DVD-RW, and one more random drive in a removable bay. Onboard Intel GPU + DVI-D adaper, two case fans, and an Antec SU-430W PSU.

Full off (back panel PSU switch), it draws 0.000 amps.
Turned "off", it draws .038 amps @ 120V, roughly 4.5W.
Steady idle-state gives 0.95 amps @ 120V, roughly 114W.
Full load(*) gives 1.62 amps @ 120V, roughly 195W.

(*) All four CPUs 100% busy, all five SATA drives 100% busy.

Cheers

Hi,

For many switching power supplies, it is recommended to have at least 20% minimum load. If that is not provided, the power supply may operate in discontuous conduction mode (the loop response is not ideal (lower in frequency)). This mode actually can draw additional current but instead of being at the base switching frequency (say 400kHz), it might be between 20kHz to 100kHz.

Power supplies might be based off of a master switcher and post regulates the other voltages. Those may be switchers too or linears. Bottom line is each switching power supply should meet minimum load requirements. That said, not all power supplies need a minimum load to guarantee regulation or prevent discontinuous conduction mode. Other topologies help with both of these issues.

The minimum load is required for each switching regulator output. It is important to make sure that each supply voltage has enough for the demands of your system components. You already indicated that suspect you may not have enough and your disks would spin-up and spin-down. You might have enough for the 3.3V, but lack +5V or +12V capacity.

I'm not advocating placing load resistors on minimally loaded supplies (being energy conscious), but, I also know that a minimally loaded or overloaded supply voltage can cause difficult to locate system problems. I have used power resistors on lightly loaded supplies (often, they resulted in poor regulation and transient response) to fix flaky problems especially under temperature and voltage margins.

Match up the system requirements to the power supply capabilities (including future upgrades) and you will have a more robust system.

Just my $0.02.

Ross

That's very interesting. How did you measure this amperage draw?

Here, I thought hard drives were a big draw. Running three IDE drives, the drives would power off and crash under disk load. One drive eventually died from the abuse. Running two IDE drives, I had occasional power cycling. I am now running one IDE drive and the system is stable.




So, if I don't put ENOUGH draw on this "750W" psu, I will have problems? Great, then I should have gone with "500W" or "600W".


This was my intention, but I found it hard to find honest data about power usage. I couldn't find how much power the motherboard drew. And I wasn't sure how to trust the power supply manufacturer ratings. I've never seen (noticed) any mention of "efficiency".

Well, I've got a good source for buying PSUs. Maybe I'll have to review this even closer. Thanks for the tips.

With an AC ammeter, of course, clamped around one leg (conductor) of the 120VAC power cord. This enables use of the Volts x Amps approximation for power, which is apparently more than the actual real power draw.

Cheers

Modern-ish ones seem to run to about 12W for 7,200RPM, 20W for 15,000RPM.

It sounds more like your power supply isn't really supplying 350W, or is failing and no longer supplying it, or isn't able to supply it on the right voltage rails. (That is, it's no use having a 350W PSU that can supply, say, 200W at 3.3V, plus 100W at 5V, plus 50W at 12V -- and that is how these things are measured -- if what your PC needs is 100W at 3.3V plus 100W at 12V.)

Peter

Following on from the discussion about energy efficiency, here's a website with listings "80plus" certified PSUs... http://www.80plus.org/manu/psu/manu_psu.htm

http://www.extreme.outervision.com/psucalculator.jsp looks like a pretty neat PSU calculator. I haven't used it myself but for $1.45 you can get a version that breaks the PSU requirement down in amperes for the major supply rails (+5v, +12v, +3.3v).

Here's another, older system I have here: 2.4GHz P4, 2GB DDR SDRAM, Intel mobo, Radeon 7000/VE PCI graphics card, one firewire card, a lone older 160GB IDE hard disk, two optical drives, and a floppy drive, ps2 mouse + keyboard.

Full off (back panel PSU switch), it draws 0.017 amps, ~2W.
Turned "off", it draws .05 amps @ 120V, roughly 6W.
Steady idle-state gives 0.55 amps @ 120V, roughly 66W.
Full load(*) gives 0.94 amps @ 120V, roughly 113W.

(*) CPU 100% busy, hard disk 100% busy, mouse moving windows rapidly around the screen.

Mmm.. I don't like the fact that this (Enermax) PSU doesn't *really* turn everything off when the back button is off.. maybe time to rewire it.

Cheers

One more: This my main system, in use for 8+ hours daily (and in-suspend the rest of the time), so power usage here really matters. Dell Inspiron 9400 "notebook", Core2Duo 2.1GHz, 3GB DDR2, 160GB 2.5" 7200rpm SATA, DVD-RW, ATI X1400 graphics card w/256MB, 17" backlit LCD, external bus-powered USB2 hub & mouse.

Full-off measures .01 amps @ 120V, roughly 1W.
Suspended-to-RAM measures .021 amps @ 120V, roughly 2.5W.
Steady idle-state gives 0.26 amps @ 120V, roughly 31W.
Full load(*) gives 0.51 amps @ 120V, roughly 61W.

(*) Both CPUs 100% busy, hard disk 100% busy, mouse moving windows rapidly around the screen, battery fully charged.

Numbers for this system include the display, whereas the two systems reported on earlier were sans display.

Cheers

As many others have said, power consumption is not fixed.

To do some calcs on total power draw in a couple of server rooms, we use one of these fairly cheap Kill-a-Watts. I've tested ours side-by-side against a Fluke DMM with clamp meter extension and it was "pretty close (TM)". Before we put a new box in production, we take readings at rest and with one cpuburn program per core. I haven't found a good program to exercise hard drives, power-wise, so would be interested if anybody knows of a "cpu-plus-diskburn"-type program.

For dual power supply boxes we put the Kill-a-Watt on one leg and unplug the 2nd power supply so we can what UPS loads we would get if a particular branch circuit or UPS failed.

Anyhow, for $25 it is pretty handy.

Prime95 (or MPrime for Linux) with huge data sets. Run one instance per CPU, and make sure that the data set size is going to cause plenty of swapping.

This also has the advantage that it'll exercise your RAM pretty well. If I've got some suspect RAM, I run it (without the swapping) for a couple of hours, then I run memtest86 once the chips have warmed up a bit.

Which reminds me, that I was going to put a random seek test into hdparm someday, to supplement the -t timings. Someday..

What I was using here, was to simply start a kernel build with "make -j7", and once it gets going well, do an "updatedb" in another window, to saturate the primary drive with seek activity.

Cheers

Mark,

Here's what extreme came up with for the first system you posted.

That's the second system I posted for (not the first).
The Enermax PSU in use there is rated at 375W, I think.

Cheers

Hi,

The power supply vendor usually indicate what current per output voltage it is capable of. Sometimes it is expressed in watts which can be converted dirctly to current by dividing the watts by the voltage, ie: 100W / 3.3v = 30.3 Amps.

If you look at the power supply specification tab on the website that was supplied, it can easily be seen that the current supplied to each voltage vary widely and do not scale with the overall advertised wattage of the supply. Some are targeted towards motherboard/RAM/Video voltages (3.3 & 5V) and some are targeted towards drives and other items. This was what I was referring to as matching the supply to your system needs (including future upgrade needs).

What I was trying to say is that each voltage may not be loaded if say the +12V has the capacity of 50 Amps. If that was properly loaded at 20% that would be 50 X 0.2 = 10 Amps. I doubt that many 3 disk drive (sets), fans, etc, would draw 10 Amps. A load resistor on that voltage might insure minimum load regulation.

The 20% load on the full 750W is not necessary, only on those that do not meet the minimum load that you suspect may cause problems. This can be located by observing the output ripple with an oscilloscope and a spectrum analyzer (with proper set-up (line LISNs, attenuators, etc)) - definitely not common place instrumentation, but, useful for locating these types of problems.

The advantage to a larger capacity supply is that if you wanted to provide minimum load resistors, they could be removed when that additional power is required in the future as you upgrade.

If you do provide minimum load resistors, be sure that you size them at double their wattage rating on the resistor. For example: if you need 20% of a 50 Amp (10 Amps) load on +12V:

1) Subtract out the existing current of the devices already in the system (say 6 Amps)
2) Subtract this from the 20% current minimum load (10A - 6A = 4A)
3) Divide 12V by the 4A = 3 Ohms
4) Calculate the wattage of the resistor Power = 4A X 12V = 48Watts. For safe operating area of the resistor, double the wattage (48W X 2 = ~100Watts) - it will draw only the 48Watts not the 100 Watts.
5) Obtain a 3 Ohm 100 Watt resistor and mount it so that it will be properly cooled (either conduction through the chassis or convection via the fan). Make sure the resistor mounting will not create a fire risk. Twist the wires together to reduce EMI. Make sure that it does not cause too much of a thermal rise in the system as well.

When you add resistors like these, make sure that it is needed to solve a problem, as it will burn constant wasteful power.

Ross

Are the minimum loads on PC power supplies really that high? Problems with underloaded PSUs are familiar on this board as the "whining Rio Central" problem, but that's a ~60W PSU being run at 1W or less. And even then only one of the two suppliers of Rio Central PSUs suffered from it.

Peter

Would a disk-defragging problem do the trick? That works the hard drive as hard as any program I know of.

I have been impressed with Auslogic's free defragger...

tanstaafl.

To maximise the power consumption you'd have to make it constantly spin up and down I guess. Not really something you'd want to do. Read the label on it instead for the power requirements at spin up

Peter,

You are correct. The amount of minimum load depends on the regulator topology. There are Forward Converters, Buck Regulators, Boost Regulators, traditional linear, low drop-out linears, those using Synchronous Rectification, etc

Each have different loop responses based on the compensation circuitry, magnetizing inductance, amount of input and output bulk capacitance, current mode control loop or voltage mode control loop, etc.

So, the topology makes a difference in what is required as a minimum load. Some require less, some require more. Typical Forward Converters are regulating larger loads and might need the full 20%. I have also seen Buck Regulators require as low as 5% and up to 25%. It all depends. Usually, the 20% makes most of them behave.

The whine is typical of an underloaded supply as the control loop is running at a much lower frequency than the switching frequency. Sometimes that is also the sign of an improperly "gapped" magnetics design or poor construction of the magnetics (Switching transformer and smoothing inductor(s) if used). The gapping is used to reduce the risk of saturating the core.

In a minimally loaded supply...

The control loop is sampling the output voltage, comparing it to a known reference. If the voltage is correct, the correction is not required for long periods of time. Either the pulse width is too narrow if it is a Pulse Width Modulated (PWM) or if it is a pulse position type (inserts a pulse when the energy is needed), there is no correction required. The control loop is at a much lower frequency. The transformer is built to be a higher frequency device (targeted at the switching frequency) and doesn't know what to do with the lower frequency demand. If the bulk input capacitance is minimal (cost and weight savings), the energy is not there to supply the lower frequency. So, it gets it from the power source (line voltage) and can actually reflect the low frequency back into the power source. This can sometimes cause EMC problems with other things attached to the same Power Mains as conducted emissions and high frequency flicker.

I know, I know, too much information...

Ross

That's a cool toy!
A little googling got me info about the successor, which includes the ability to hold info during power-off.
NewEgg.com has them both. The original for just $16 and the 'EZ' for $40.
Oh, the EZ will calculate the usage cost directly if you give it your rate from the power company.
Here's a review of the Kill-A-Watt EZ.

Then there's the Watt's Up Pro(Amazon) for those who want to fiddle with the data on your computer. It's $115-$150, though, so the original P3 Kill-A-Watt is a definite must have at $20.

Wow.

I still find it difficult to accept that people still say stuff like this (from the review text of the EZ model):

It's not a belief, people. It is happening, and far more quickly than originally thought.

Cheers

That was why I used the term wasteful in my response. I also believe that we are accelerating the damage to our planet.

Interestingly, 60 Minutes aired a piece last night that further validates that is happening with forest fires.

Ross