Buying computer hardware

[Ossa]

Errmmmmm guys..... thats all very well. All I wanted to know was if my PSU 360W was enough. :lol

Well, ive baught it now. How do you tell if the PSU is to low in wattage?

Don't worry, yours is fine... you have an 89W CPU and your graphics card is 75W peak. With the other stuff, you're doing fine with 360W.

As typical efficiencies are about 70%, ...

There are sugestions (on the internet) that efficiencies greater than 80% are on the way!

Wow... didn't realise that they were so bad. I did a switch mode power converter course last year at MIT (6.334), where we took a field trip (the only one I've ever heard of at university!) to a server power supply manufacturer: http://www.synqor.com/ They seem perfectly capable of making 90% efficient power supplies (actually just looked up the spec for a 360W half-brick, 95% is achieved for a good range of loads), which are admittedly much more expensive than PC PSUs, but from what I learned, 80% should be fairly easy to do for PC PSUs, for not all that much money.

Note: I realise that PC PSUs are AC/DC converters and synqor makes DC/DC converters, but many of the same principles apply.

[edit] Seems thay aren't that bad at all actually, for instance this gives 78% minimum (this was the first manufacturer that I looked at). [/edit]

Ossa

[Mark Jones]

Furthermore the efficiency varies with load, tyipically being highest at less than full load and lower at full load and at light load. The difference may be as much as 10% - 20%. So running 500W PSUs at lower power outputs is not quite as good as it seems.

Hi Roger, you've hit on some of the physical limitations of the engineered topology itself here. But aren't today's supplies better than 80% efficiency? Maybe that is a minimum figure? In any case, putting a 600W supply in a tiny PC is just a waste of power, while putting a 150W supply in a PIV 4GHz might "let the smoke out." :wink

[MichaelW]

I have not checked recently, but Storage Review used to be a good site for hard disk information.

[hutch--]

In the days when I made transformer based power supplies for audio, power consumption was based on load and an idling transformer power supply used almost no power at all but I just asked my older brother who is a whiz in electronics after having been at it for 50 years and he said the later choppers are also very efficient and are smart enough to cut right down on power consumption if the load is very low so I cannot see a reason not to run a big power supply as they are a lot more reliable and don't use any more power than is drawn by the sum total load from gizmos in the computer.

[Mark Jones]

Huch, if my memory serves me correctly, PC supplies are quite different to transformer action. An ATX supply simply will not operate without a load attached because of several factors. PC supplies are switchmode-type supplies which operate by switching a series of inductors "into" the line voltage for brief periods, then "tapping" the stored energy slowly. This is done very rapidly, between 50-500 thousand times per second. The current flowing into the inductors creates an electro-magnetic field there, and physics dictates that the field can either be released quickly (by switching in a large electrical resistance, which results in a higher voltage / lower amperage,) or released slowly into a tiny resistance, which is seen as a lower voltage / higher amperage. A CPU has very little resistance, so the line voltage is essentially "stepped down" into exactly 3.3v or whatnot. The frequency and/or duty cycle of the supply is autonomously corrected to create the precise voltage outputs and changes with load. For example, if all the hard drives are spinned down and system power usage is little, the supply will decrease the "on" pulse period which effectively lessens the power put into the magnetic induction process. When the drives spin up, more power is needed and the "on" period is increased, creating greater magnetic induction and more resultant power available for the devices.

The end result is that there are losses involved with operating outside the supply's engineered "sweet spot." For argument's sake, lets say that a 500W supply, ran at say 85% load, might yield 80% efficiency. But that same supply, ran at 30% load, might be only 55% efficient.

Here's some great data on supplies, along with a shoot-out of various 300-800W models (be sure to see the results!): http://www.tomshardware.com/2005/07/11/stress_test/

[Roger]

Hi All,

Note: I realise that PC PSUs are AC/DC converters and synqor makes DC/DC converters, but many of the same principles apply.

Its the high volt AC/low volt DC that tends to have lower efficieny.

There are sugestions (on the internet) that efficiencies greater than 80% are on the way!

Mark: I read  this article dated last year.  http://www.trustedreviews.com/article.aspx?art=1014  See page 23 for efficiencies.
It is about 5 months earlier than yours and so the products are not quite so efficient. You will see that it is refering to required  minimum efficiencies as low as 60% (light load) and only recommends 80% for normal load. Conclusion: Check the specs because not all manufacturers will go for the best.

. . . an idling transformer power supply used almost no power at all

As I remember the lectures on transformer design major losses = Iron loss + copper loss.  Iron loss also called magnetisation loss is nominally constant over the whole usable power range whereas the copper loss or I²R loss is dependant on load. In those days (over 40 years ago) small transformers could strugle to get 60% efficiency and, because the Fe loss vector was normal to the Cu loss and load vectors, the no-load input power could actually be higher than the full-load input power. This may be why we now all use transformerless designs  :wink

In any case, putting a 600W supply in a tiny PC is just a waste of power,

Not so much power, its more a waste of money. Then again they say 'money is power' so you might be right. :wink :wink

Regards Roger

[RedXVII]

Waste of power? Whaaat.

If you have a 600W supply and only use 200W in your machine, surely the total power consumption by the machine is 200W, not 600W.

[Mark Jones]

Not exactly. If a 600W supply is 50% efficient (overall) when powering a 200W load, then that supply takes in 400W/h to create the 200W/h delivered. The excess 200W is lost as heat into the room. (This is what is meant by "50% efficiency.") That's why PSU's have fans and cooling panels and radiator fins... to keep them from burning up. :bg

If the same supply was 80% efficient with a 500W load, total consumption would be (500W * (1 / 0.80)) = 625W/h

If it were 72% efficient with a 600W load, total consumption would be 833.33W/h.

A toaster, electric hot-water heater or hair-dryer (generally considered to be the most energy-consuming devices) are about 1.5KW each in comparison. An electric range heating element or microwave oven is about a kilowatt. So the computers of today are becoming big-time energy hogs, especially if they operate continuously.

[hutch--]

Roger has an interesting point here, a transformer that is powered on one side and open circuit on the other still has to magnetise the transformer core so there is clearly a load even when there is none added from the other side. To some extent the hysterisis characteristics of the laminations while the power is being cycled at normal AC rates will compensate for the load which will be different from its startup load but there is still a load involved at idle.

The other factor is transformer design has been subject to both better materials and basic design and here I am thinking of torroidal transformers on some of the later silicon steels used as core material. I used to measure the input and output of transformers when building audio and crappy little I and E core transformers produced some strange looking stuff on the output end. The power here in OZ is 240 volt 50 cycle and its reasonably close to a sine wave and with a class transformer the output did not look much different to the input on a dual trace oscilloscope where poor laminations in junky I and E core transformers produced some really weird wave forms.

I used to be able to put a small 2 amp torroidal transformer in the same case as an audio pre-amp and the field leakage was so low you did not have to shield them and they are really quiet in comparison to the best I and E core stuff.

[Ossa]

Mark is right here... transformers have very little to do with the way modern PSUs work. They essentially have two parts (although ones even a few years ago had a very different first part, so many are still in use).

Let me just say that there is a measure of how close the current waveform is to the voltage waveform in terms of both shape (i.e. how close it is to a sinusoid) and in phase. The term for this value is power factor and it can be split down into one part for the distortion and one for the phase. The first part is determines the harmonic distortion, and I won't go into the mathematical definition, but suffice to say that the closer it is to 1, the less distortion there is. The phase part is defined as simply the cosine of the phase angle (so if the current and voltage are in phase, this takes on the value of 1). These two values are multiplied together to get the overall power factor. Large industrial companies are often charged by the power companies for having power factors that vary from 1 (as this causes merry hell with the power grid).

Although it's not needed, many modern PSUs still have a transformer at the input before the rectifier... so losses there DO occur, but these are not hugely significant.

The first bit is a rectifier... but now-a-days it's no ordinary rectifier. Lets first look at what used to happen though... basically the PSU would only draw current during the peak part of the voltage wave... this meant that it was in phase, but it had a horrible pulse train type current waveform... so its harmonic distortion was very high. It is for this reason that if you ever (DO NOT TRY THIS UNLESS YOU KNOW EXACTLY WHAT YOU AR DOING) look at the mains voltage waveform on an oscilloscope, you will see that it is a trapezoidal waveform... all these old power supplies are still in use. The way that it varies the power delivered is the length of the current pulses... now, as well as the power factor, the efficientcy of such systems as awful... but it is the power factor that is causing government agencies to introduce new standards that forbid the sale of power supplies that use this type of front end.

New power supplies attempt to control their current waveform, so that it is a sinusoid... I won't go into how, but if you are interested, get hold of a copy of Principles of Power Electronics by J.G. Kassakian, M.F. Schlecht and G.C. Verghese. Let me also say that these rectifiers are very efficient, little of the loss is from this component.

The second is a DC/DC converter (which Mark described nicely a few posts back)... These supplies are called switch mode, because they are split into two circuits, each of which has access to an inductor (or other storage device - a capacitor - but an inductor is the most common), but they cannot "access" the inductor at the same time, so the two circuits are switched in and out of operation very fast... and the ratio of time that each circuit has access to the inductor is called the duty ratio... by varying this,
the amount of power delivered to the load varies enormously. What the system attempts to do is keep the output voltage constant by varying the duty ratio.

So where is the power loss? In several places: the switching components and the storage components... basically the switch has two types of loss: switching loss (from allowing both current to flow and a potential difference accross the device) and conduction loss (from conducting when the switch is meant to be off). The loss in these
is highly dependent on on the power output, but I must make a point here: when switching loss is worst in one component, it often the best in another. The result? switching losses are lower for lower loads, but are proportionally higher, i.e. efficientcy decreases for low loads. The inductor behaves much as you might expect (with copper and core losses) and once again, efficiency will decrease for lower loads.

There is a big note here: the amount that efficiency decreases for low loads does very from PSU to PSU... so if you can get your hands on the spec sheet for the PSU, it will hopefully show you the efficiency versus load graph. But as I said before... because PSUs are made for as little money as possible, they are far less efficient than is possible - 90% efficient (min... max is more like 95%) are possible... just no one makes them.

Ossa