Article on PSU size & overkill; irresponsible journalism

[mcoleg]

that and the fact that the ambient will be the same for both psu's.


speaking about the ambient, do you think it would be of any interest to introduce some heatload for the psu's intake, just to see it sweat?

[jessekopelman]

speaking about the ambient, do you think it would be of any interest to introduce some heatload for the psu's intake, just to see it sweat?

Of course. Do you really think the PSU fan would have trouble keeping up even with 100% load at an ambient < 30C? When SPCR tested the S12-330, the worst intake-exhaust delta they got was 8C. As I've said before the only possible concern one can have running a PSU at > 50% utilization is that the fan can't handle the internal heat and the heat from other components in the case. Unless you try and simulate such inside the case conditions, what's the point?

[DyJohnnY]

is it just me or is this written backwards?


The CPU and the video card are the biggest power draws, the rest don't consume all that much. Here's what PSU draws from the wall (I am rounding up):

At idle: 250W with 2A
At load: 380W with 3.2A

////snipped///

But nagging doubts remained. And so, I started collecting info and making the calculations. The 750W Silencer hits 80+ efficiency above 150W and keeps it till 600W load.

Round up the efficiency to 80% and that will give us:

380W / 5 * 4 = 304W

of real power being consumed by the system under full load. Give it a bit of a buffer to stay on the safe side and it looks like all I need is a 380W PSU for that system.


gents, the power ratings for the CPU and GPU are the actual consumption of the devices, the devices transforming the desired power from 12V to whatver vcore, vdimm, have their own efficiency characteristic!!!!
meaning that if component A, draws 100W, voltage regulator R with say 80% efficiency will draw in total 100W*100%/80% = 125W. 100W is useful power, for component A, 25W is wasted power, heat [preferably].

this power is taken from the PSU itself. who in turn must also convert from 220/110V to 12V, for example, with it's own efficiency. let's say PSU efficiency is also 80%.

so, we need 125W, from the PSU with 80% efficiency.
125W * 100% / 80% ~ 156W.

let's sum up the formulas shall we?

A = required power [let's assume all regulators in the PC are identical]
e = efficiency of the regulator system [percentages]
E = efficency of the PSU [percentages]
x = total power to be used by the system [regulator + powered component]
X = total power drawn BY THE PSU FROM THE OUTLET

so X = A * (1/e) * (1/E) = A / ( e * E)

in the case of the article writer: [let's assume e =0.8 and E = 0.8]

X = 380 / (0.8 * 0. = 380 / (0.64) = 593 W this is the power the PSU draws from the outlet,

offering the power

x = 380/0.8 = 475W - to be used by regulators to offer the 380W of power desired for the components.

my point: - the values he mentioned in the review are erroneous.
i'm not trying to say that system needs 750W of power, but i am trying to say that if you know FOR SURE [not read on the internet, not calculated with god knows what calculator] all your PC components may be drawing 380W under full load from your PSU - don't get a 400W PSU.

I'm pretty sure the items he tested weren't stressing EVERYTHING to the limit so he could even be close to the 380W limit [that looks a bit inflated itself to me, but i can't give out a number for what he described there].

[jessekopelman]

DyJohnny, maybe I am misunderstanding you, but I think it is you who are backwards. The rating of the PSU is how much DC it provides, not how much AC it draws at the wall. If the DC draw of all your components combined is < 380W, a 380W PSU will provide sufficient power even though it has to draw 475W (380W/80%) or more at the wall to do so. The issue we are discussing goes beyond the idea of choosing a PSU based on DC Wattage requirements. Some people are saying you must not use a PSU at over 50% load, so as to be safe from overheating. Others, myself included, say that is a bunch of BS.

[jaganath]

gents, the power ratings for the CPU and GPU are the actual consumption of the devices

that's not strictly true (TDP is just a requirement for thermal solution manufacturers, many chips use less than TDP at full load) but DyJohnny has a point, there will be extra efficiency losses associated with the CPU and GPU voltage regulators that we have not accounted for. I don't think it invalidates the general SPCR theme that people need less power than they imagine, but it's something that is relevant to mcoleg's calculations.

[gmat]

Yes TDP only reflects the characteristics of the chip itself, not its surroundings. So CPU alone and GPU alone, not the whole motherboard and not the whole graphics card.
How much current draws a northbridge chip ? What's drawn by a SCSI RAID card ? What are the power requirement of sound cards ? Who knows ?
Finding the real power needs of a system, apart from assembling one and actually getting a hold of proper equipment to measure power in each part, is kinda tricky. Add up the known TDP's ? Add the power ratings of HDD's and fans ? and wave hands, add a nice round random figure and make a ritual dance ? That's how some people end up with imaginary power needs.

To get back to my "sweet spot": If you get to know the REAL power needs of the system (which are varying, no doubt) you have to center the main use mode of the system around the peak efficiency of the PSU. That's what i meant the efficiency being a curve, try to stay around its top.
PSU's that power industrial equipments 24/24, are used at near peak rated power... and they dont really fail that much (usually it's a cooling related problem ! - friends who do maintenance on those witnessed that a lot, overheating always lead to premature failure)

[mcoleg]

sry, didn't notice there were replies here.

DyJohnnY - thanks for reading the article and thanks for using numbers.

you have probably noticed two causality lines in my assumptions and that's what thrown off your calculations.

allow me to point a couple of things that i consider of interest here:

-the psu is rated 380 watt continues output
-imagine the power consumption in blocks or objects where solidity is divided by the conversion from greater power to lesser one. then you will see where the dc/dc conversion efficiency loss belongs.
-check dc/dc conversion efficiency numbers, btw.

another thing - "items he tested weren't stressing EVERYTHING to the limit" - if you re-read the article, you'll see what tests i've done. i am pretty sure it's the highest my system can draw.

if you got any other ideas for the tests, i'll try them.



jessekopelman - there's more than a backward calculus here. there's an extra component to the equation that's not suppose to be here since it's already calculated in as a part of a whole system's power consumption.

jaganath - it's irrelevant; see point 2 above. for example, if you think of how much power a motherboard burns, do we not calculate the dc/dc conversion loss into the whole?

gmat - no matter how many times you repeat it, i just can't get behind the "sweet spot" idea. power consumption is a dynamic process that involves power, time and change. the simplest i can imagine is a graph... "sweet spot" gives me a vision of a dot .

[gmat]

I agree that my wording wasn't the best.. so here's a graph


Try to have your most typical system uses into that "sweet spot" zone The problem with users of 700W (or more !!) PSU's is their system places them on the very left of the curve, so they waste a lot of efficiency as compared to a lower powered PSU.

Of course, the efficiency curves look different for each PSU, some manufacturers publish it, some don't (but there's SPCR !), so usually you can devise a best PSU to put on a known system.

What would be VERY interesting and enlightening, is actual power measurements on each rail (3.3V, 5V, etc) on typical systems:
* SFF
* HTPC
* entry level home PC (<$70 CPU, <$150 GPU)
* gamer's PC with single GPU
* gamer's PC with dual GPU
And get a min/max of these power draws.
Maybe i missed something but i've never seen such tests, anywhere, done in an extensive manner.

[EsaT]

What would be VERY interesting and enlightening, is actual power measurements on each rail (3.3V, 5V, etc) on typical systems:
* SFF
* HTPC
* entry level home PC (<$70 CPU, <$150 GPU)
* gamer's PC with single GPU
* gamer's PC with dual GPU
And get a min/max of these power draws.

In highend PCs CPU and graphic card(s) are absolutely biggest factor affecting to consumption numbers.
X-bit Labs has measured consumption of these parts quite accurately in their reviews so I think those should be good values for making estimates.

[VanWaGuy]

dyJohhny's point is valid for the CPU though. If you look at the power usage by the CPU, it does have some loss on the motherboard where the voltage is stepped down to Vcore. This conversion though is probably in the mid to high 90% range though, so the other rounding up and margins that are being added in the calculations would also accomodate this power supply to Vcore loss.

[santacruzbob]

well, as an overclocker, i do get the idea of an overkill. also, as an overclocker, i'd like to choose my components carefully and with consideration.

so far it did not include the psu, apparently. a 500w unit would have been more than sufficient for the components and would have had a buffer to withstand the drop in efficiency from the additional heat. yet my last two psu buys were 700 and 750 watt.

lol. 500W is almost certainly overkill. I run a 235W channel well tech PSU w/ a 5v fan powering a 939 3200+ @ 1.1v, and a 350W sparkle PSU, also with 5v fan powering a e6300@3ghz w/ 7950GT. Quality engineering makes all the difference. I couldn't even imagine tossing a 700w PSU in anything... I think marketing hype has blinded consumers.

[kaange]

To all those who think mcoleg is pushing the envelope of the EarthWatts, could they perhaps suggest a series of applications/tasks that mcoleg could run which would cause his system to overload his PC?

I think the testing he did was pretty realistic test of how far he could push his power consumption.

JonnyGuru, I really enjoy your website and admire your testing methodology but to suggest that mcoleg is guilty of "irresponsible jounalism" is irresponsible in itself and all the articles that suggest you need a 700W+ PSU to power a system like mcoleg's are far more irresponsible (not saying you would suggest this but other less technically capable site). A well built PSU should run at it's CONTINUOUS rated output without problems - it just may be noisier at that level than a PSU with twice the rating but your reviews show that isn't always the case either. And almost no PCs run near their max power draws for hours upon hours anyway.

I installed an HX520W in a friends machine. He was running a QX6700 and a pair of X1950XTX cards. We've all seen HX520W's run more than that, right? Well, it wasn't my first choice for PSU, but it's what I had handy and the build worked fine so I let it go like that. God damn Thermaltake Soprano case didn't have much airflow or something because the PSU overheated and shut off and I couldn't get her to come back on again no matter what. Did I blame the PSU? No. It's cram packed in that case and there's only one intake and one exhaust fan and it's really kind of tight in there when you have two video cards, two hard drives and two opticals. I replaced the HX520W with a Seasonic M12-700 and I haven't heard back from him (this was 8 months ago) so I assume everything is fine. Does his build need 700W? No way. I would say that open air, his PC could probably run forever on a good 500W power supply. But his build wasn't open air and his Soprano case was tucked under a desk so I would expect temperatures to get a little hotter than 40 or 50C around the power supply.

I would think that in this case, it wasn't the greater capacity of the M12-700 that allowed this PC to operate more reliably but the extra 60mm cooling fan that the M12 series has which the HX series lack. I would bet good money that a M12-500 would have worked just as well.

[DyJohnnY]

@mcoleg

I'm not sure what you mean by my calculations being off.
If we assume the wattages you listed are correct then the numbers are corect.
I mean at least the way i understand it, what you are saying in your article is:
if component X has nominal power P, the power drawn from the PSU is P' - lower than P. that physically impossible, u can't use 100W to get 150W for example

There may be one or 2 flaws in my logic.
1 is if the TDP numbers you listed INCLUDE the "<1" efficency of the CPU regulators, and i find that highly questionable since there are a gzillion ways to provide regulated voltage to those parts, with various efficiencies.
The second may be the fact that GPU TDP mai include the DC>DC conversions since the board has integrated voltage regulation, while the CPU does not.

with this argument in mind i still stand behin what I wrote.

As far as you stressing everything on that PC to it's INDIVIDUAL limit I am still sticking to what I said, i don't think you were able to push EACH component individually to the MAX LOAD.

I will also give you at least one reason why not:
let's take a simple scenario:
CPU
GPU
HDD that's all you have on your PC.

Let's say for argument's sake the GPU rarely polls the CPU for data, using IRQ's, so we can assume they both can run at 100% at the same time, for a sustained amount of time. This is highly ideal but let's assume it is like this.
However for the HDD things are in reality far from ideal. when a hdd need to do something it sends an IRQ to the CPU, the CPU stops, listens and performs what it is asked to. If you are runing a test that also stresses HDD you will get a gzillion IRQ's to the CPU, telling him to stop whatever it's doing and serve the drive with requested operations. This inevitably means that CPU is not on 100% all the time and overall system load decreases due to the high number of IRQ's. you all know this is true, try running a video benchmark while copying a lot of files and you might see "frames" in an otherwise smooth benchmark run.

So at least as far as a HDD a GPU and a CPU being each under 100% LOAD generating the maximum load EACH of them can ever be stressed at, that is quite questionable in my view.
Add more drives, copying between hdd's, optical drives and things will look the same way.

Yes i don't know exact DC>DC conversion ratings but i assumed a fair value i believe, or at least i put things in plain sight, it's not something to be overlooked.

I'm also sorry about distracting everyone with my mention of the AC socket power draw, i figured it would be rather unimportant by the time i reached the end of the post, but after all that writing i was not going to delete it .
It was however a bit distracting from the main point i was trying to make, i admit that, that being "consider DC>DC voltage conversions taking place on the MB/GPU", but I am glad some people have seen and acknowledged the issue.

Regarding the power draw figures measured....

jaganath - it's irrelevant; see point 2 above. for example, if you think of how much power a motherboard burns, do we not calculate the dc/dc conversion loss into the whole?

I'm not sure you can put the DC>DC conversion ratios in that category since the CPU has a separate power supply connection, and every CPU has a different TDP, hence the conversion ratio is applicable to different TDP's resulting different loads on the regulators.
i'm not sure whether the GPU TDP takes into account the DC>DC conversions, that's a different story, the regulators are located on the board itself, so it is quite possible, but the CPU does not, it relies on the board.

As far as measuring real power draws: i can only think of one way:

Devise a suite of stress tests.
Use your current method to repeat the tests and see if the numbers you get match, just so we know the tests can be reproduced with the same results.

Measure current draw of each component by taping into it's supply line. [canibalize a PSU to get a new set of cables you can extend the current ones with, in between the cables you insert one Ampermeter, multimeter, osciloscope is even better, but whatever]

You run each suite of tests and measure current on each voltage line. multiply that by the line voltage and you get power drawn from the PSU, simple "straight from the horse's mouth" - no more calculations necessary.

You do the same for voltage lines for CPU, GPU [if you can do that somehow], HDD, optical drives, etc, etc.

Then you add all the numbers up and they should be the total power draw from your PSU in your stress test enviroment.

I think this is the best way one can measure DC power draw from the PSU, that's how reviewer's do it, maybe with advanced tools [like EM field measurements around the DC voltage lines] but the principle is the same.

disclaimer: anyone attempting this needs to understand the logic behind what i'm talking about, the risks and must have the expertise to perform the tasks....kind of, don't blame me if you short your 12V line with the ground and get a vegetable computer.

[kaange]

DyJohnnY, you are going back to the "add the max possible power draw for every single component" approach for determining system power requirements. This is just plain silly as it almost impossible to do in the real world - a very thoroughly written application for a particular system configuration may be able to do it but I doubt it. You, yourself have stated a simple example of why this is the case - so why cater for this impossible situation?

The rest (especially DC->DC conversion) really is irrelevant.

And as jessekopelman, pointed out, your PSU rating calculation was reversed.

[jessekopelman]

DyJohnny, I think you are missing the point of the overall discussion in this thread. We were not debating the most accurate way to measure DC load, we were debating whether a PSU is capable of dealing with the heat involved in running at > 50% utilization. The crux of this argument is really about the ability of the PSU to dissipate heat produced outside of the unit from other components in addition to heat produced inside the unit from AC-DC transformation inefficiency. (Hopefully) No one is saying you need a huge amount of Wattage overhead for its own sake -- instead some (like jonnyguru) claim that higher wattage rating = better ability to dissipate/survive heat from other components. Going back to mcoleg's original article, the point was not that he stressed his system to the theoretical limits, but that he was unable to come up with anything even remotely approaching the rated draw of his "low-end" PSU even though many people would advise using a much higher rated PSU with that same system. The only thing that needs to be taken from that original article is that it is very hard to build a system that has more than 400-500W of DC draw and thus the central issue of PSU selection returns to the question of whether a higher rated PSU is able to better handle the heat of a poorly ventilated case (as it is becoming difficult to find a PSU not rated for 400+W) -- i.e. should you shoot for jonnyguru's suggestion to have a PSU rated for double the required load, just so you don't have to worry about whether your case is properly ventilated?

[mcoleg]

ok, i'll play. one last time though. i'll be using the numbers from the article, check it to see what the numbers stand for.

(17A+17A)x12V=408W - no limiters on 12V rails; this is a peak number so we don't care about it. maximum continues load on 12V rails is 324W, according to specs.

you were assuming that DC/DC conversion is 80%:

(132W+135W)/0.8=333.75W

alright, let's see. 132W is for the whole video card, not just gpu alone. the voltage converters are on the card itself so what we have here is:

132W+135W/0.8=300.75W

DC/DC conversion worth it's salt is 90-95%; google it. that gets us between:

132W+135W/0.9=282W and
132W+135W/0.95=274.1W

DC/DC conversion circurtry is located on the motherboard (not supplied directly from the psu) and the video card so the numbers above cover the end-point load plus step-down loss. in addition to 12V rail, the 5V and 3V rails supply the motherboard, the drives, the memory and the chipsets but compared to 12v rail's their load is pretty negligible. whatever's left till 380W (note that 380W is a continues, not a peak value), is plenty to cover all the extra bits that are not as power-hungry as the cpu and the video card. the numbers that we'll get then are pretty close to what i've got from testing.

also, just to alleviate yet another misconception, at maximum load (not peak) the psu will draw from the wall 380W/0.8=475W



and still, all the above is pointless and irrelevant since what i am interested is how much the system as a whole consumes. meaning i take measurements at the wall before the power supply and i have the efficiency curve to calculate how much the psu outputs. the system tested is presented as a complete closed system that draws a certain amount of energy. that includes all of the components' loads and all the heat and conversion losses.

to summarize - watts from the wall * 0.8 = watts the psu outputs. watts the psu outputs = watts the whole system draws from it. individual components' maximum power consumption is only a rough estimate to get a ball park figure as to how much psu needs to output at maximum load of the system. internal DC/DC conversion is only a part of the whole system's power consumption.


if i have to explain it any simpler, i'll have to start drawing pictures :p .


good idea about taking a psu apart and measuring all the rails separately. when you get that done, please post your findings.

[dhanson865]

So the high power opinion is that if the PC can draw 300 watts I should use a 620W PSU instead of the 430W PSU so I have more overhead.

Lets throw some other numbers in this discussion:

Code: Select all

Model       Output (W)  200     250     300     400     500
        Efficiency
     S12-430            81.8%   80.5%   79.6%   75.5%*
     HX620W             83.6%   84.5%   84.7%   83.9%   81.9%

         Temp Rise (°C)
S12-430                  5       7       8       8*
HX620W                   9      12      12      12      11

         Noise (dBA@1m)
S12-430                 25      29      32      37*
HX620W                  22      22      22      29      43

So the 430 watt choice is staying cooler as measured by the difference in temperature between the intake and exhaust air as mentioned here http://www.silentpcreview.com/article683-page4.html

I'm going to say that the 430 watt version will be just as reliable if not more so just because of that temp difference. But at that noise difference and efficiency difference no serious gamer would be happy with the 430 watt PSU

Now look at this comparison:

Code: Select all

Model       Output (W)  150     200     250     300     400     500
        Efficiency
     HX520W             81.0%   84.5%   85.2%   85.1%   83.7%   81.3%
     HX620W             79.9%   83.6%   84.5%   84.7%   83.9%   81.9%

         Temp Rise (°C)
HX520W                   7       9      12      13      12      11
HX620W                   7       9      12      12      12      11

         Noise (dBA@1m)
HX520W                  22      22      22      22      29      43
HX620W                  22      22      22      22      29      43

What reason is there for a gamer to believe that the 620 watt will do any better for him than the 520 watt? The 520 watt version has better efficiency at loads under 400 watts and the temps and noise are practically identical.

Why not save the extra money and just get the HX520W?


OK, so what about the Corsair HX versus the Seasonic M12:

Code: Select all

Model       Output (W)  150     200     250     300     400     500
        Efficiency
     HX620W             79.9%   83.6%   84.5%   84.7%   83.9%   81.9%
     M12-700            80.3%   83.5%   83.8%   83.6%   82.8%   81.5%

         Temp Rise (°C)
HX620W                   7       9      12      12      12      11
M12-700                  2       2       6       7      10      12

         Noise (dBA@1m)
HX620W                  22      22      22      22      29      43
M12-700                 21      21      21      25      35      42

The M12 stays cooler under most circumstances and noise is a mixed bag. Given the previous comparison you would expect a M12-600 to rate just as well as the M12-700.

Since the M12 stays cooler I'd expect it to last longer but then again the Corsair is cheaper by about $30 for the same wattage so you get what you pay for...

[jessekopelman]

What reason is there for a gamer to believe that the 620 watt will do any better for him than the 520 watt?

As Rick James so famously said, "Cocaine is a hell of drug." Or, if you prefer; the philosopher Ilkka Niiniluoto's question, "is it rational to be rational?" I would also refer you to the concept of magical thinking. Any of these three could be your reason.