End Power Supply Myths!

[VERTIGGO]

Originally posted by: jpeyton

They quote an poorly written review from ExtremeTech as evidence that their 850W unit can power what competitor's 600W units can't. An FX-60/SLI setup doesn't use anywhere near 600W at load, and any quality competitor's model should power it fine (620HX for example, will power that without breaking a sweat).

Well I know mine runs about 520W - 540W while Folding and running 3DMARK concurrently. I am using the PC P&C 510 SLI and a Thermaltake 250W Express GPU PSU. I know I have it covered, (and of course I did run the system for a while before the additional 250W) but I think it's wise to have a buffer. Especially since I'm going to E6600 and maybe crossfire.

 

[rise]

where are you measuring that Vertiggo?

 

[jonnyGURU]

Originally posted by: magreen
I'd like to know what you think of this data: AnandTech measured power consumption of an X6800 & 8800GTX system and found it to be about 275W.

Like I said (and you quoted. ), I would hardly call running 3DMark a "full load."

I would say that '06 is probably only going to cause the video cards to use about half the power they're capable of drawing, but if you think about the fact that each G80 GPU can only consume 225W simply due to physical limitations, and that CPU's are actually using less power than your GPU's these days, then you would be hard pressed to justify needing a 1000W PSU or greater for anything.

I don't have the facts and figures for the actual power consumption of every CPU and GPU out there. If I did, I'd just build my own PSU calculator. But I often use this example as a benchmark:

I built a quad FX rig (dual FX74 processors) with 4GB of RAM and a pair of 8800GTX cards. For thermal testing, I loaded each of the four cores up to 100% using K8 Burn. I then ran 3DMark'06. The most I could draw from the wall with this unrealistically monsterous build was 1000W. At 80% efficiency, that's only 800W DC. Considering that you're not going to be running each of four cores at full load AND be gaming at the same time (the gaming experience would be unbarable with all four cores in use by another app) there's no reason why even if you did have this kind of rig you would even need an 800W power supply. I had to use at least a 700W PSU because the PC wouldn't even boot with less than that, but even a 700W PSU would allow me to run 3DMark'06 on this rig.

Now consider if you only had one of those CPU's. What would you need then? A 600W?

To put this into even better perspective: I built a QX6800 with a pair of 8800 Ultra cards. The CPU and two GPU's were cooled with CoolIT devices drawing 120W total all on their own. Even with four hard drives, four GB of RAM, Killer NIC, what have you.. the most I saw this rig draw from the wall was 700W. At 80% efficiency, that's only 560W DC.

 

[jonnyGURU]

Originally posted by: rise
where are you measuring that Vertiggo?

That's got to be from the wall.

And considering a PCP&C 510 is not a very efficient PSU, I'd say about 75%, that's only about 400W DC.

 

[RaiderJ]

Did that calculator, my server needs about 307 watts. Perfect for the 450 watt supply I grabbed for it.

All the push towards massive power supplies is silly. I imagine there's about 2 people out there who need a 1200 watt supply.

 

[VERTIGGO]

but isn't the PSU rated by the intake wattage? so it really doesn't matter, it's using up to 540W, and it's rated 510W (albeit @ 50C and with the 250W addition).

 

[NeoPTLD]

Here's my opinion.

It's generally a good idea to stick with a brand known for power supplies that delivers what they put on specs. It's true they might not make their own and outsource the manufacturing, but they are the one who ensures only those that meets the specs are sold under their name.

A no name selling brand might be ok at one time, but their next shipment of lowest bid might not be, so it's a hit and miss.

There's a difference between a power supply that can deliver the rated power continuously and a power supply that is marketed with a 15 second peak power. Generally, there's a relationship between power and weight. The watts/lbs of a power supply doesn't change so much. Components can be pushed harder to achieve more efficiency, but that costs you efficiency.

Don't shop on the wattage alone. The rail distribution should be matched to what your computer needs. This is like buying paint. Say a project needs 20 gallons(10 gallons grey, 5 gallons blue, 2.5 gallons red, 2.5 gallons yellow). so you go out and buy 25 gallons (the equivalent of total "watts" ), 6 1/4 gal of each color. This won't do you good, because despite having surplus in total amount, you're short on grey.

Noise consideration:
Get an 80Plus rated high efficiency power supply. With 300W load, there's a 53W difference in input between 70% and 80%
efficiency. It might not be a big deal from overall power use, but the 53W difference is directly used in the power supply itself.

At 300W load, the 80% efficient dissipates 75W
70% efficient unit dissipates 128W, which is 70%, so the PSU cooling needs about twice the capacity meaning more noise.

While the efficiency difference only saves you a buck or two on power bill, you will get 15-20% more runtime on UPS and reduces noise. The efficiency on power supplies is the best around 50% load. www.80plus.org site shows good efficient test data for various 80+ certified power supplies.

 

[jonnyGURU]

Originally posted by: VERTIGGO
but isn't the PSU rated by the intake wattage? so it really doesn't matter, it's using up to 540W, and it's rated 510W (albeit @ 50C and with the 250W addition).

No. PSU's are rated for DC output. Not AC input.

 

[jonnyGURU]

Originally posted by: RaiderJ

All the push towards massive power supplies is silly. I imagine there's about 2 people out there who need a 1200 watt supply.

If the budget allowed for it, I would rather have a 1200W power supply with better voltage regulation, less ripple and noise and is more efficient at all loads than a 400W power supply that has mediocre voltage regulation, near spec ripple and noise and is only efficient at 50% and up loads, even if my computer only typically used 200 to 300W.

 

[JEDIYoda]

While the efficiency difference only saves you a buck or two on power bill, you will get 15-20% more runtime on UPS and reduces noise. The efficiency on power supplies is the best around 50% load. www.80plus.org site shows good efficient test data for various 80+ certified power supplies.

That`s not necessarily true!!
Efficiency and run time have nothing to do with quality verses a PSU thats not as efficient but used better quality caps ans such and will be running 10 years from now!!

Verses a highly efficient PSU that yin a year you might be sending back for another one.,..

Peace!!

Where did this notion that build quality and efficiency go hand in hand???

hmmmm

 

[magreen]

Um, maybe read what he wrote again. He said runtime on a UPS. i.e. if the power goes out, your 80% efficient psu will be using up less of the battery on your UPS than a 70% efficient psu, so the UPS can run it longer.

 

[NeoPTLD]

Originally posted by: jonnyGURU
<div class="FTQUOTE"><begin quote>Originally posted by: RaiderJ

All the push towards massive power supplies is silly. I imagine there's about 2 people out there who need a 1200 watt supply.</end quote></div>

If the budget allowed for it, I would rather have a 1200W power supply with better voltage regulation, less ripple and noise and is more efficient at all loads than a 400W power supply that has mediocre voltage regulation, near spec ripple and noise and is only efficient at 50% and up loads, even if my computer only typically used 200 to 300W.

Some power supplies have a minimum load requirement. A 1200W power supply could go out of regulation during very low power state.

Also, it's not that those small PSUs are only efficient above 50%. The efficiency curve is mound shaped, best efficiency around 50%. Did you look at any of the test reports on www.80plus.org? For minimum heat production the power supply should be loaded to about 50% under idle conditions. If your computer has more than 2:1 idle to full load ratio.

I don't think Tom's Hardware's extensive review said anything about ripple level vs relative load. Do you have a link to test data showing ripple and noise goes up with relative power use?

People who have a computer with a need for sustained 1200W of power will need a dedicated 15A branch circuit, but should have a 20A circuit if it's to be shared with anything.

A 1200W PSU with active PFC with with 0.98 PF and 0.8 efficiency draws 12.7A

You'll be slightly over what's permitted by code, even with just the computer on the circuit. You'll definitely need an unshared, dedicated 20A circuit if you were to share with another equipment. Preferebly two 20A circuits on the same duplex if you are ever going to use a laser printer.

You might need yet anther 120V or a 240v circuit if you're going to use a portable AC to cool that place.

" Cord-and-plug connected equipment not fastened in place, such as a drill press or table saw, must not have an ampere rating more than 80 percent of the branch circuit rating [210.23(A)(1)]. UL and other testing laboratories list portable equipment (such as hair dryers) up to 100 percent of the circuit rating. "

 

[jonnyGURU]

Originally posted by: NeoPTLD

Originally posted by: jonnyGURU

Originally posted by: RaiderJ

All the push towards massive power supplies is silly. I imagine there's about 2 people out there who need a 1200 watt supply.

If the budget allowed for it, I would rather have a 1200W power supply with better voltage regulation, less ripple and noise and is more efficient at all loads than a 400W power supply that has mediocre voltage regulation, near spec ripple and noise and is only efficient at 50% and up loads, even if my computer only typically used 200 to 300W.

Some power supplies have a minimum load requirement. A 1200W power supply could go out of regulation during very low power state.

Minimimum load requirements tend to be only 1A per rail which is easy for a computer to accomplish. And most quality power supplies are independently regulated, so cross loads aren't a problem.

Originally posted by: NeoPTLD
Also, it's not that those small PSUs are only efficient above 50%. The efficiency curve is mound shaped, best efficiency around 50%. Did you look at any of the test reports on www.80plus.org? For minimum heat production the power supply should be loaded to about 50% under idle conditions. If your computer has more than 2:1 idle to full load ratio.

I never said that small PSU's are only efficient over 50%. I just said it was a possibility. You have to look at each PSU on a case by case instance. Yes, the efficiency curve is mound shaped, but that mound varies from unit to unit and the mound might start earlier or later on some units than others.

Originally posted by: NeoPTLD
I don't think Tom's Hardware's extensive review said anything about ripple level vs relative load. Do you have a link to test data showing ripple and noise goes up with relative power use?

My name is jonnyGURU. I help companies design power supplies and I occasionally review them: http://www.jonnyguru.com.

Originally posted by: NeoPTLD
People who have a computer with a need for sustained 1200W of power will need a dedicated 15A branch circuit, but should have a 20A circuit if it's to be shared with anything.

A 1200W PSU with active PFC with with 0.98 PF and 0.8 efficiency draws 12.7A

You'll be slightly over what's permitted by code, even with just the computer on the circuit. You'll definitely need an unshared, dedicated 20A circuit if you were to share with another equipment. Preferebly two 20A circuits on the same duplex if you are ever going to use a laser printer.

You might need yet anther 120V or a 240v circuit if you're going to use a portable AC to cool that place.

"Cord-and-plug connected equipment not fastened in place, such as a drill press or table saw, must not have an ampere rating more than 80 percent of the branch circuit rating [210.23(A)(1)]. UL and other testing laboratories list portable equipment (such as hair dryers) up to 100 percent of the circuit rating."

I don't think there's really anyone reading this that actually "needs" a 1200W PSU. The two biggest rigs I've built so far this year were a Quad FX with a pair of 8800GTX cards and a QX6800 with a pair of 8800 Ultras cooled with CoolIT systems.

The Quad FX, with a load on all four cores and running 3DMark'06 would only draw 1000W from the wall at the most. That translates to about 800W DC based on the efficiency of the PSU I was using.

The QX6800 was only drawing 700W from the wall despite the fact that the CoolIT coolers alone used 120W DC. I could've easily run that machine off of a 600W or 700W PSU.

 

[magreen]

Originally posted by: jonnyGURU

Originally posted by: NeoPTLD
I don't think Tom's Hardware's extensive review said anything about ripple level vs relative load. Do you have a link to test data showing ripple and noise goes up with relative power use?

My name is jonnyGURU. I help companies design power supplies and I occasionally review them: http://www.jonnyguru.com.

LOL. That's funny!

Anand: The newer 1333 FSB Intel chips run on different internal timings from the 1066 chips.

Questioner: I don't think Tom's Hardware Guide mentioned this about Intel chips. Do you have a link to support your claim?

Anand: My name is Anand Lal Shimpi. I have this tech review site...

 

[JEDIYoda]

Originally posted by: NeoPTLD
<div class="FTQUOTE"><begin quote>Originally posted by: jonnyGURU
<div class="FTQUOTE"><begin quote>Originally posted by: RaiderJ

All the push towards massive power supplies is silly. I imagine there's about 2 people out there who need a 1200 watt supply.</end quote></div>

If the budget allowed for it, I would rather have a 1200W power supply with better voltage regulation, less ripple and noise and is more efficient at all loads than a 400W power supply that has mediocre voltage regulation, near spec ripple and noise and is only efficient at 50% and up loads, even if my computer only typically used 200 to 300W.


</end quote></div>

Some power supplies have a minimum load requirement. A 1200W power supply could go out of regulation during very low power state.

Also, it's not that those small PSUs are only efficient above 50%. The efficiency curve is mound shaped, best efficiency around 50%. Did you look at any of the test reports on www.80plus.org? For minimum heat production the power supply should be loaded to about 50% under idle conditions. If your computer has more than 2:1 idle to full load ratio.

I don't think Tom's Hardware's extensive review said anything about ripple level vs relative load. Do you have a link to test data showing ripple and noise goes up with relative power use?

People who have a computer with a need for sustained 1200W of power will need a dedicated 15A branch circuit, but should have a 20A circuit if it's to be shared with anything.

A 1200W PSU with active PFC with with 0.98 PF and 0.8 efficiency draws 12.7A

You'll be slightly over what's permitted by code, even with just the computer on the circuit. You'll definitely need an unshared, dedicated 20A circuit if you were to share with another equipment. Preferebly two 20A circuits on the same duplex if you are ever going to use a laser printer.

You might need yet anther 120V or a 240v circuit if you're going to use a portable AC to cool that place.

" Cord-and-plug connected equipment not fastened in place, such as a drill press or table saw, must not have an ampere rating more than 80 percent of the branch circuit rating [210.23(A)(1)]. UL and other testing laboratories list portable equipment (such as hair dryers) up to 100 percent of the circuit rating. "

What I lik best is Johnny`s rsponse---My name is jonnyGURU. I help companies design power supplies and I occasionally review them: http://www.jonnyguru.com.

Besides being the most respected PSU GURU on the internet!@!

 

[jonnyGURU]

I try not to be too cocky, though.

All it'll take is an EE to come along and bitchslap me into next tuesday.

 

[JEDIYoda]

Originally posted by: jonnyGURU
I try not to be too cocky, though.

All it'll take is an EE to come along and bitchslap me into next tuesday.

Thats why we all likw yaz!!

 

[NeoPTLD]

<div class="FTQUOTE"><begin quote>Originally posted by: jonnyGURU


Minimimum load requirements tend to be only 1A per rail which is easy for a computer to accomplish. And most quality power supplies are independently regulated, so cross loads aren't a problem.

Any data? I have not been able to find this information listed. I have seen a 200A 5V 1kW server power supply with a 15A minimum load requirement though.

<div class="FTQUOTE"><begin quote>
I never said that small PSU's are only efficient over 50%. I just said it was a possibility. You have to look at each PSU on a case by case instance. Yes, the efficiency curve is mound shaped, but that mound varies from unit to unit and the mound might start earlier or later on some units than others.
</end quote></div>

You're right they vary from unit to unit. 50% seems like a good median point though, but we can't say for sure without a statistical analysis.

<div class="FTQUOTE"><begin quote>
My name is jonnyGURU. I help companies design power supplies and I occasionally review them: [L=http://www.jonnyguru.com]http://www.jonnyguru.com</end quote></div>.

I have had power supply shut down on me from plugging in a load to Molex connector while it was powered on. It must not have been able to handle the sharp rise in current.

Transient response is yet another factor outside of "steady state" wattage that's important in ensuring stability. What sort of relative peak maximum and minimum you rate as acceptable for highly transient loads, such as 10% to 90% load change on 12v bus which could occur if all four processors were in sleep state and put into full load on a moment's notice.

What would you specify as the maximum allowable dI/dt to keep the notching within spec?

I have a laser printer with a very high dI/dt and the wiring to the outlet has a relatively high impedance. It causes enough notching in the power line to trip my UPS over to battery and get it flagged for "line notching" not to mention a very annoying flicker to lighting.

If the output side impedance isn't low enough, PC supplies can have the same problem with highly transient load and cause other components to malfunction.

 

[jonnyGURU]

Originally posted by: NeoPTLD
Any data? I have not been able to find this information listed. I have seen a 200A 5V 1kW server power supply with a 15A minimum load requirement though.

You're trolling. I'll let you pull up the PDF's of several PSU's on the market. Just grab a random 10. I'll bet you at least 7 or 8 will substantiate what I'm saying.

Originally posted by: NeoPTLD
You're right they vary from unit to unit. 50% seems like a good median point though, but we can't say for sure without a statistical analysis.

50% is probably the best median point when talking "in general." Do I have statistical analysis'? Yes. Every time I do a PSU review and I state the efficiency at each load. Lately, most PSU's have been a very consistant efficiency from 15 to 20% load all of the way to 100% load, with the exception being at very low loads or when the PSU is crossloaded.

Originally posted by: NeoPTLD
I have had power supply shut down on me from plugging in a load to Molex connector while it was powered on. It must not have been able to handle the sharp rise in current.

Transient response is yet another factor outside of "steady state" wattage that's important in ensuring stability. What sort of relative peak maximum and minimum you rate as acceptable for highly transient loads, such as 10% to 90% load change on 12v bus which could occur if all four processors were in sleep state and put into full load on a moment's notice.

What would you specify as the maximum allowable dI/dt to keep the notching within spec?

I have a laser printer with a very high dI/dt and the wiring to the outlet has a relatively high impedance. It causes enough notching in the power line to trip my UPS over to battery and get it flagged for "line notching" not to mention a very annoying flicker to lighting.

If the output side impedance isn't low enough, PC supplies can have the same problem with highly transient load and cause other components to malfunction.

My load tester does not "ramp up" loads unless I ramp them up. If I jump from test to test, the change in load is immediate. Most PSU's, unless they are very poorly built, do not shut down. Typically when they shut down, it's because the voltage drops below spec and the PSU's PWM is programmed to put the PSU into protection if the voltage drops below a certain threshold. This can be witnessed on an Oscope because typically the drop is very sudden. I haven't attempted to capture this drop out myself, but fortunately our good friend Paul Johnson will begin doing transient load testing on units he reviews for Hard OCP. He has a manual switch that drops a sudden load on a particular connector and then he captures the resulting drop in voltage with a screen shot of his USB Stingray in action. He's currently done this only with the Turbo Cool 1kW, which did not shut down, and will be doing this test from this point forward.

 

[The Boston Dangler]

nobody mentioned www.spcr.com?

Power Supply Fundamentals
Recommended Power Supplies

 

[JEDIYoda]

Originally posted by: The Boston Dangler
nobody mentioned www.spcr.com?

Power Supply Fundamentals
Recommended Power Supplies

why mention SPCR? who are they anyways???..lol
Personally you can`t take seriously a company that does real nice PSU reviews yet will rate a PSU that is quiet higher than a PSU that is built better but not as quiet......
thats why I say SPCR who???

Thats another reason why i would rate johnnyGURU`s site over the SPCR site and also over ard|Forum...yet I would rate ard|Forum over SPCR....

Peace!!

 

[rise]

lol, in case you didn't know, the "s" in spcr stands for "silent". it stands to reason that they would value silence highly. jonny does the same thing, weighting performance, aesthetics, etc differently in his conclusions. for me, aestetics are useless yet he rates them 40% i think. that doesn't mean i dismiss his reviews (just his conclusions, lol) seriously, his scores mean nothing because of that but i love his reviews.

so i know jonny is your new bestest pal but lets not just dismiss others becasue you don't like their conclusions.

edit- i stand way corrected on jonny's aeststics value, i guess i was so disgusted it weighted in the score at all that i remembered it as worse than it is anyway, 10% aesthetics and he has the audacity to define "value" to me, at 30%

 

[JEDIYoda]

Originally posted by: rise
lol, in case you didn't know, the "s" in spcr stands for "silent". it stands to reason that they would value silence highly. jonny does the same thing, weighting performance, aesthetics, etc differently in his conclusions. for me, aestetics are useless yet he rates them 40% i think. that doesn't mean i dismiss his reviews (just his conclusions, lol) seriously, his scores mean nothing because of that but i love his reviews.

so i know jonny is your new bestest pal but lets not just dismiss others becasue you don't like their conclusions.

edit- i stand way corrected on jonny's aeststics value, i guess i was so disgusted it weighted in the score at all that i remembered it as worse than it is anyway, 10% aesthetics and he has the audacity to define "value" to me, at 30%

Actually no Jonny does not do the same thing...you didn`t read what i posted...

Johnny does not rate a lesser quality PSU over a better quality PSU just because the leser quality PSU is quieter....nope jonny does not do that!! Sorry!!!

 

[rise]

woosh

 

[jonnyGURU]

Originally posted by: rise
edit- i stand way corrected on jonny's aeststics value, i guess i was so disgusted it weighted in the score at all that i remembered it as worse than it is anyway, 10% aesthetics and he has the audacity to define "value" to me, at 30%

LOL... Well, don't go out shopping with me anytime soon. The constant "you can get that for $$ at XXX, don't buy that here" will drive you up the wall.

Ratings are subjective no matter what site you visit because every one has an opinion and no matter what that opinion is going to show through a review.

If you're the kind of person to read the first page and last page of a review and nothing in betwen to find out how the conclusions were made, then please, PLEASE, don't read my reviews, reference my reviews, link to my reviews, etc. Some of the worst recommendations for and against certain products on the web are done because things are taken out of their context. I can give countless examples of how and where this has happened.

The fact that I give 10% of the score to aesthetics and 30% of the score to value bothers you, then you must be one of those that goes straight to the last page and looks at only the conclusion, because if I otherwise pan a PSU because it has poor regulation, excessive ripple and noise, poor efficiency, etc. it's going to be in there from page 2 and on.