Power Consumption Revisited

Question

I was reading an article on tomshardware.com the other day in which they described a process by which they were able to measure the power consumption of various video cards using a simple device that plugs into a standard wall socket and then displays the number of watts the currently plugged in device is using.  In light of the ever-increasing PSU recommendations that tend to show up here (I recall seeing one poster recommend "a PSU with 24 or more amps on the +12V rail for anything other than a barebones system"), I decided that it might be beneficial to these forums if I did a little empirical study of my own.  So anyways, I shelled out ~$30 for the device shown here:
http://www.supermediastore.com/kilwateldet1.html
...and ran some tests of my own.  The results:
Preliminary Testing -
To verify that my power meter would give reasonably accurate readings, I first hooked it up to a 3-way lamp with a 50/100/150 watt bulb installed.  The readings returned for the 50/100/150 watt settings (respectively) were 44/94/142, so it would seem like my power meter is at least reasonably accurate.  Some other stuff I measured just for the hell of it...my speakers use 3 watts of power in stadby mode, and 30 watts when active (haven't yet tested when active and playing at full volume), and my monitor uses about 70 watts when on, and about 2 watts in standby mode.
Results -
Satisfied that I had not just wasted my money on an inaccurate power meter, I then went and hooked it up to my PC (the one described in my sig) and measured the power consumption under a variety of circumstances.  It is important to note that these readings reflect the total amount of power drain being applied to the wall socket, not the amount of power that is actually being demanded by the system.  This is because no PSU is 100% efficient (a good one will be maybe 80% efficient, if even that much), so the amount of power that is actually needed by the system is actually about (at least)20% *less* than the recorded values.  Anyways:
During startup, the power usage spikes very briefly at 197 watts, then averages 152 watts over the rest of the boot cycle.
The system uses 134 watts of power when idling.
Under full CPU load, the system uses 168 watts.
Running 3d Mark 2001 the power usage is 169 watts.
Playing Far Cry (high detail settings, 1024x768x32), the power usage is again 169 watts.
Conclusions -
So, let's now assume a worst-case scenario, in which the extra 34 watts recorded during full CPU load came entirely from extra CPU power drain (a reasonable assumption), and in which the extra 35 watts recorded during 3d Mark and Far Cry came entirely from extra video card load (a much less reasonable assumption), and in which we have a PSU that is 90% efficient (greater efficiency means that the system would actually have to demand *more* power in order to get the total power drain up that high).  In this case we see that if an application were developed that fully taxed the video card and CPU continuously, the total power drain would be 134 + 34 + 35 = 203 watts (which actually correlates rather nicely with the 197 watt spike observed during the boot cycle), meaning that the system is demanding about 183 watts from our unrealistically efficient PSU (note that with the PSU efficiency set to a more realistic 75%, the system would only be demanding a mere 153 watts of juice at full CPU and video card load).  
Admittedly, the video card in my system is relatively weak, so let us again take the worst case scenario and assume that if I were to be using a 6800 Ultra, the total power drain would be 100 watts greater (this is above what the actual difference should be given the results posted on tomshardware.com regarding the power use of the 6800 Ultra), so our video card now consumes an astounding 135 watts of power, and our total power drain (in our unrealistic situation where we have some application which is capable of 100% CPU and video card utilization for a sustained length of time) is now 303 watts.  With our unrealistic 90% efficient PSU, it would mean that the system is demanding about 273 watts from the PSU (about 228 watts with a 75% efficient PSU).
Note that aside from the weak video card, I have a fairly robust system (which also happens to be slightly overclocked), with 4 HDD's (two of which are WD Raptors), 2 optical drives, several PCI devices, and two large 120mm case fans, and yet the power demands of this system, even in an unrealisticly demanding situation, are *well* within the ability of a quality 380W (or even 300W) PSU to deliver.  In this case even if all the power happened to be being sucked off of the +12V rail (which is not the case), any PSU with 18 amps at +12V could still handle it.  Furthermore, even if I were to add a needlessly power-hungry video card into the mix, the power demands are *still* safely within what any decent 380W PSU should be capable of (and even what a quality 300W PSU should be capable of, although this may be pushing it a little, though it should always be noted that the numbers indicate a hypothetical worst-case power drain that should be beyond the maximum drain possible in any real-world situation).
So, we can therefore conclude that the power demands of a reasonably robust Athlon64 based system are not astronomical by any means, and that they do not justify a minimum recommendation of a 465W PSU with 24+ amps on the +12V rail for any system which is not "barebones," and that there is no observational evidence to support the idea that a PSU with 18 or fewer amps at +12V is categorically inadequate for use in an Athlon64 based system.
...anyways, I guess that's all, I hope you found this interesting, or at least informative.  I'm off to see what else I can do with my power meter thingy...

Answer 1

Really?  Do you have measured data which clearly supports your claims, or are you just holding up an opinion as a matter of fact?
My point was, my measured results show that the total power demand of an Athlon64 based system across *all* of the rails is fairly low, even at 100% system load.  So, let's recalculate things assuming a 75% efficient PSU, with 75% of all load being at + 12V (which is still probably higher than the actual value), and let's leave the hypothetical 6800U inside of my system.  We get .75 * 303 = 227 watts in total that the system is demanding.  Of these 227 watts, the system is demanding .75 * 227 = 170 watts over the +12V rail.  170 watts / 12V gives us a total demand of 14.2 amps on the +12V rail.  Note that this is with the hypothetically demanding 6800U card installed and is still likely to be at least a couple amps higher than what a *real* system would ever use, and any *quality* PSU capable of 18 amps at +12V should still be perfectly adequate for use in the system.
Furthermore, PSU efficiency dropping to 60% in real world situations supports my results, as it means that the actual system was demanding substantially *less* power than the system in my hypothetical example, making things even *easier* for the PSU.  Re-running the above equation with a 60% efficient PSU and 75% of all power demand coming from the +12V rail, we see that the system is only asking for 11.4 amps at +12V at full load with a 6800U installed (and also at full load).
If you want to disagree with my results, that's fine, but don't expect me to take your argument as credible unless you have some actual, measurable data to back up your claims.  Saying "this is the way things *really* work because I say so" doesn't cut it, so until you want to break out a multimeter and measure the amps your PSU delivers to the MB on the +12V rail at boot, idle, load, and gaming and then report your results and discuss whether or not they are consistent with your "amps are what counts" hypothesis, I hold my results and conclusions up as being valid, and as soon as I see any measured results which contradict mine, I will gladly stfu about PSU recommendations being needlessly high.