PDA

View Full Version : How To Check Your Power Supply Voltages


Merc
March 27th, 2007, 12:24:09 PM
Ensuring that you get steady, solid DC voltage to your CPU, RAM and other components is vitally important. The NF4 boards are notoriously picky about power and if the current fluctuates greatly or is below a certain threshold (which is WAY above the +/- 5% ATX spec) then you can get all kinds of weird and seemingly unrelated problems plaguing your system. This is why the first question anyone asks regarding strange system behavior is "What PSU do you have in there and what are its specs?" Total watts and amps on the 12v rail(s) are the primary data they are looking for with this question but the 3.3v and 5v rails are also important, although not usually as critical as most PSU's can supply plenty of DC on those rails.

So how can you tell what current your PSU is supplying and if there are any fluctuations on the rails? Well, the mobo itself has sensors that provide this info and you can read what those sensors are seeing in BIOS or via Asus Probe. Simple enough right? Wrong. Take a look at what Asus Probe is reading on the 12v rail. Is it 11.76v? A 12v reading at that level would immediately get my attention with this board but your system is probably running fine right? The A8N SLI Deluxe sensors are not at all reliable and give false readings that make them fairly useless except for seeing fluctuations.

In order to get a real reading of your voltages you'll need to get a voltmeter and go probing. This is not as daunting a task as it sounds and a voltmeter comes in handy for all kinds of things. Neither does it need to be expensive. You can pick up a pocket size multimeter (mm) at Radio Shack for about $20 http://tinyurl.com/az7hb. LCD readout, small size etc. It's a nice piece of gear and plenty for this job.

The first step is to be aware that you are using metal probes around live traces so BE CAREFUL. The second step is to remember the wire colors and the voltage they relate to:

Orange = 3.3v
Red = 5v
Yellow = 12v
Black = ground

There are two ways to measure the actual voltages:

1. You can plug into an empty molex and get a fairly good reading. Just stretch the empty molex outside the case and test away.
or
2. You can plug right into the back of your 24 pin power connector and get the true enchilada. This is very hazardous though and not worth the risk.

On your empty molex stick the red probe into the the socket with the color wire related to the voltage you want to check, stick the black in black wire socket and if testing a 12v rail set the mm to 20vDC. Now play BF2 or something else that will stress your system (a benchmark or encode a DVD etc) and watch your mm readings as the system works. Expect to see some drop in line voltage but how much is where the judgement call comes in. My PC P&C drops 1/100th under stress but it is a remarkably strong and efficient PSU. Most will drop more than that but you don't want to see more than a couple of 10ths of a volt.

The ATX spec states that +/-5% of the rated voltage is an acceptable amount on the 12v rail. This equates to 11.4v to 12.6v You really want to be at 12.1v. If you aren't sure then write down your readings and post them here and we'll take a look.

Good luck

35Pete
March 27th, 2007, 5:16:40 PM
Merc. In Electrical engineering terms that drop in voltage with respect to change in load current is called "Load Regulation".

The change in output voltage with changes in the supply input voltage (your wall socket) is called Line Regulation.

In this case, the load regulation spec is 5%.

Great article bud.

35Pete
March 27th, 2007, 7:35:17 PM
Question Merc.

How do you know that your multimeter is calibrated?

If it is then ramp up and down the core voltage, read it on the ASUS instruments, and then record a series of data points (voltages) with the multimeter.

The MOBO is certainly using an ADC (analog to digital converter). They may inaccurate, but are usually precise as hell.

The difference between accuracy and precision?

Look at these pics. The one with high accuracy is closer to the bullseye but has a lot of variability. Precision is away from the bullseye (inaccurate) but very reproducible.

High Accuracy-Low Precision
http://i6.photobucket.com/albums/y241/35Pete/150px-High_accuracy_Low_precision.jpg

High Precision-Low Accuracy
http://i6.photobucket.com/albums/y241/35Pete/150px-High_precision_Low_accuracy.jpg

Based on this, I can calibrate my instrument readings with the following:

Collect the data in two points. Warp your core from say...1.30 to 1.50 in .2 volt (200 mV) intervals, if your core is cool with that voltage level.

Then. you have two (x,y) pairs....Guess what? You can find the calibration (correction) constants so that you can distill the actual core voltage from your SW readings.

You'll get:

Offset Voltage. (b)
Multiplier (m)

Trust me. ADC's are VERY linear unless they are ultra-cheap crap. That means, the ACTUAL voltage can be converted from your SW readings with this real simple equation:

V_actual = V_sw_measured * m + b (from y = mx+b).

http://i6.photobucket.com/albums/y241/35Pete/calibrate.jpg

Then you just pull up the ole' Microsoft calculator and do the simple math. Close your case, and check your voltages.

Merc
March 27th, 2007, 10:58:30 PM
You can absolutely calibrate your BIOS readouts to a known good meter but never trust them unless you have. The manufacturers do put monitoring system in the board but they are never calibrated. I have never had a board read right.

As far as the meter calibration, well, i don't know but unless it is significantly flawed it is good enough for these purposes. The main thing you want to watch is the idle voltage on the 12v and then load the rig up and watch the droop. Bgf ugly droops are bad for overclocking so always get a board with good power delivery and, IMHO, 8 phase.

Droop and drop are OCer terms. I try and write to that crowd but anything you can teach is incredibly welcome. Just put it on a business major from Canisius levelhttp://www.smiliemania.de/smilie132/00000290.gif
The only thing I have learned from years of doing this is there is a helluva lot I don't know and it increases everyday. That's why this stuff is is addictive and fun. http://www.vocinelweb.it/faccine/biggrin/11.gif

35Pete
March 28th, 2007, 6:04:03 AM
Cool. But if you ever start reading the tech specs on some of this stuff they may use the term load regulation. Now you are one up on the other OC'ers ;)

Ohh, don't forget drift. If you use my method you need to recal your equation every few months.

And a quick and dirty cal is to measure several power supplies (cell phone, VPN router, portable appliance, ect..)

Record the data and plot it. It should be a one-to-one line. If not, then get the slope and the zero-intercept (offset) and now you've calibrated (so to speak) your multimeter.

BTW. Multimeter calibration from the factory is only good for about a year.

The MOBO manufacturers have a SLOPPY manufacturing process. (Usually the result of ultra-competitive pricing) These things come of the factory line and into a process called "Tune and Test". Someone's not controlling the tuning process tight enough. You see, we engineers EXPECT the ADCs and other circuits to be inaccurate. We do count on them to be precise though. Hence we can "tune" them (mostly with SW) to be both precise and accurate. I've written many a tune and test procedure for circuits that I have designed for the TSE group in the factory (test systems engineering). There is NO EXCUSE for the wrong calibration coefficients to be in that bios. None, nada.

BTW. The method I gave you is standard method in the electronics industry. Trust me, it works. Your cell phone would be FUBAR if it didn't.