How does a power supply work?

[TVeblen]

I'm trying to figure out how the math works on a power supply's ratings
regarding wattage and current. Using my 750TX power supply's specs as an
example, here are the listed ratings and my math (V x A = W in parentheses).

Input: 115v US x 10A = (1150 Watts)

Output:
-3.3v @ 24A = (79.2W)
+5v @ 28A = (140W)
Total Above: (219.2W) Combined rating in spec = 180W

12v @ 60A = (720W) 720W in spec

-12v @ .8A = (9.6W) 9.6W in spec

5vsb @ 3A = (15W) 15W in spec

Total all the spec Watts: (924.6W) for a 750W PS.

So, in layman's terms, how does this math work?
I can understand a loss of power between input and output due to
conversion, but not understanding why the math does not add up on the
output side.

Specifically, if someone has a 360W power supply and is trying to run a
high end video card, could you really expect it to reliably supply 12v
at 30A, or is this output going to be reduced by the draw on the other
rails? And how could a layperson determine what amperage to reasonably
expect from that power supply for the video card?

 

[Don Phillipson]

I'm trying to figure out how the math works on a power supply's ratings
regarding wattage and current. Using my 750TX power supply's specs as an
example, here are the listed ratings and my math (V x A = W in parentheses).
. . .
Specifically, if someone has a 360W power supply and is trying to run a
high end video card, could you really expect it to reliably supply 12v
at 30A, or is this output going to be reduced by the draw on the other
rails? And how could a layperson determine what amperage to reasonably
expect from that power supply for the video card?

You may have misunderstood . . . In real installations,
PSU capacity is a practical matter determined by experiment
(not by totalling all possible outputs at theoretical maxima).
This is why most MB vendors supply a monitor program
that measures actual temperatures, fan speeds, voltages
currently in use, etc.

 

[Paul]

I'm trying to figure out how the math works on a power supply's ratings
regarding wattage and current. Using my 750TX power supply's specs as an
example, here are the listed ratings and my math (V x A = W in
parentheses).

Input: 115v US x 10A = (1150 Watts)

Output:
-3.3v @ 24A = (79.2W)
+5v @ 28A = (140W)
Total Above: (219.2W) Combined rating in spec = 180W

12v @ 60A = (720W) 720W in spec

-12v @ .8A = (9.6W) 9.6W in spec

5vsb @ 3A = (15W) 15W in spec

Total all the spec Watts: (924.6W) for a 750W PS.

So, in layman's terms, how does this math work?
I can understand a loss of power between input and output due to
conversion, but not understanding why the math does not add up on the
output side.

Specifically, if someone has a 360W power supply and is trying to run a
high end video card, could you really expect it to reliably supply 12v
at 30A, or is this output going to be reduced by the draw on the other
rails? And how could a layperson determine what amperage to reasonably
expect from that power supply for the video card?

It means the supply is designed with the expectation that one
rail is more heavily loaded than the others.

http://images17.newegg.com/is/image/newegg/17-139-006-Z05?$S640W$

The supply is a 750W total. The 12V rail is rated at 12V @ 60A or
720W. You are allowed to draw that 60A if you want, but if you
did, there would be little remaining total power allowed to be
drawn from the other rails. You'd have 30W left for the other
rails, which for the computer, may not quite cover all the
loads present. It is highly unlikely you could arrange to draw
the 60A, as the other rails may need more than the 30W figure.

That label really spells it all out for you. For example, the
180W limit, says that, although you have 3.3V @ 30A and 5V @ 28A
to work with, the total of those two is limited to 180W. Perhaps
they're sharing a resource inside the power supply. You could draw
99W from the 3.3V side, and have only 81W left from the 5V side,
so that would be 5V @ 16A or so. Either the 3.3V or the 5V rail
could be loaded up to its max, but that would leave less than
the max for the other partner. And obviously, if the 3.3V and 5V
are drawing 180W, then you're no longer going to be allowed to
draw 12V @ 60A, because the sum total would be over 750W.

You'll notice that the label is kinda arranged in "layers". The
750W at the bottom, is an overriding concern. The next level up,
partitions the rails into smaller groups of concern. Finally,
you get to the individual rails and their maximum limits.

To work out what is available on your 750TX, you do a budget
for the computer. I've done a bunch of these in the past for
people, as an illustration of how you can go about it. For
example, I allocate 50W for the motherboard and RAM, and
assume it comes from either the 3.3V or the 5V rail, without
being more precise about it. (The 50W is based on my experience
measuring just two of my motherboards, so there is little basis
for the estimate. I checked an AthlonXP and a P4 board, to come
up with the figure.) So, if we took your 750W, and subtracted 50W
from it, now we have 700W left. If the remaining budget was allocated
to the 12V rail, that would now be less than 60A. We can add in
contributions for hard drives, optical drives, processor and so on,
until the allocation for video is all that is left.

There are also web sites that do power calculations, but
I don't like any of the current sites that are available,
due to their habit of hiding details. I feel it is important
for the estimation process to be completely exposed (as in,
a spreadsheet format), as then it is easier to verify
just how bad the estimation method is. For example, Asus
offers a web page for power supply calculations, but it
works in increments of 50W. It is quantized. You could
have 49W worth of load, and it would say "zero". You could
add 1W more load and then it would read "50W". Real stupid.
By doing so, and rounding, it isn't possible to accurately
see what each component is contributing to the total. Such an
approach, to me, does not encourage a sense of trust. As in,
the ability to verify the math is being done correctly.

The Takaman site in Japan was the only site I liked, and it
closed years ago. I think they used to sell power supplies,
and that was their reason for offering a calculator.

One web site people used to use, offered estimates that
were roughly double the real value. I think at one time,
it may have estimated a stick of RAM as 25W. When I
see stuff like that, I encourage people to use their
imaginations. For example, imagine a 100W light bulb
where your four sticks of RAM are. Would you burn
yourself on the RAM, the same way you would on the
light bulb ? If the amount of power computed, doesn't match
your experience with component temperatures, that is a
very rough way to do a reality check. Modern RAM sticks,
according to Kingston data sheets, are somewhere around
2W a piece now. Which is why I can easily hide them
in my "50W for chipset and RAM sticks" figure, without
really worrying about it.

As for "how does a power supply work", here is one schematic
to illustrate. What is shown here, is very old, but still
gives you some idea how it works. There is a text description,
so no point in me repeating it

http://www.pavouk.org/hw/en_atxps.html

The nice feature of those, is they're transformer isolated for
safety. The "Hi-POT OK" on the label of your power supply,
means someone tested the transformers for a certain level of
breakdown protection. So even that little thing, is there
for a reason. You assume they're Hi-POT tested, but it is
nice to see it in print.

Some details on Hi-POT testing are mentioned here, in the
"Hi-Pot Requirements" section.

http://www.ramtechno.com/medsafety.php

http://en.wikipedia.org/wiki/Hipot

HTH,
Paul

 

[TVeblen]

It means the supply is designed with the expectation that one
rail is more heavily loaded than the others.

http://images17.newegg.com/is/image/newegg/17-139-006-Z05?$S640W$

The supply is a 750W total. The 12V rail is rated at 12V @ 60A or
720W. You are allowed to draw that 60A if you want, but if you
did, there would be little remaining total power allowed to be
drawn from the other rails. You'd have 30W left for the other
rails, which for the computer, may not quite cover all the
loads present. It is highly unlikely you could arrange to draw
the 60A, as the other rails may need more than the 30W figure.

That label really spells it all out for you. For example, the
180W limit, says that, although you have 3.3V @ 30A and 5V @ 28A
to work with, the total of those two is limited to 180W. Perhaps
they're sharing a resource inside the power supply. You could draw
99W from the 3.3V side, and have only 81W left from the 5V side,
so that would be 5V @ 16A or so. Either the 3.3V or the 5V rail
could be loaded up to its max, but that would leave less than
the max for the other partner. And obviously, if the 3.3V and 5V
are drawing 180W, then you're no longer going to be allowed to
draw 12V @ 60A, because the sum total would be over 750W.

You'll notice that the label is kinda arranged in "layers". The
750W at the bottom, is an overriding concern. The next level up,
partitions the rails into smaller groups of concern. Finally,
you get to the individual rails and their maximum limits.

To work out what is available on your 750TX, you do a budget
for the computer. I've done a bunch of these in the past for
people, as an illustration of how you can go about it. For
example, I allocate 50W for the motherboard and RAM, and
assume it comes from either the 3.3V or the 5V rail, without
being more precise about it. (The 50W is based on my experience
measuring just two of my motherboards, so there is little basis
for the estimate. I checked an AthlonXP and a P4 board, to come
up with the figure.) So, if we took your 750W, and subtracted 50W
from it, now we have 700W left. If the remaining budget was allocated
to the 12V rail, that would now be less than 60A. We can add in
contributions for hard drives, optical drives, processor and so on,
until the allocation for video is all that is left.

There are also web sites that do power calculations, but
I don't like any of the current sites that are available,
due to their habit of hiding details. I feel it is important
for the estimation process to be completely exposed (as in,
a spreadsheet format), as then it is easier to verify
just how bad the estimation method is. For example, Asus
offers a web page for power supply calculations, but it
works in increments of 50W. It is quantized. You could
have 49W worth of load, and it would say "zero". You could
add 1W more load and then it would read "50W". Real stupid.
By doing so, and rounding, it isn't possible to accurately
see what each component is contributing to the total. Such an
approach, to me, does not encourage a sense of trust. As in,
the ability to verify the math is being done correctly.

The Takaman site in Japan was the only site I liked, and it
closed years ago. I think they used to sell power supplies,
and that was their reason for offering a calculator.

One web site people used to use, offered estimates that
were roughly double the real value. I think at one time,
it may have estimated a stick of RAM as 25W. When I
see stuff like that, I encourage people to use their
imaginations. For example, imagine a 100W light bulb
where your four sticks of RAM are. Would you burn
yourself on the RAM, the same way you would on the
light bulb ? If the amount of power computed, doesn't match
your experience with component temperatures, that is a
very rough way to do a reality check. Modern RAM sticks,
according to Kingston data sheets, are somewhere around
2W a piece now. Which is why I can easily hide them
in my "50W for chipset and RAM sticks" figure, without
really worrying about it.

As for "how does a power supply work", here is one schematic
to illustrate. What is shown here, is very old, but still
gives you some idea how it works. There is a text description,
so no point in me repeating it

http://www.pavouk.org/hw/en_atxps.html

The nice feature of those, is they're transformer isolated for
safety. The "Hi-POT OK" on the label of your power supply,
means someone tested the transformers for a certain level of
breakdown protection. So even that little thing, is there
for a reason. You assume they're Hi-POT tested, but it is
nice to see it in print.

Some details on Hi-POT testing are mentioned here, in the
"Hi-Pot Requirements" section.

http://www.ramtechno.com/medsafety.php

http://en.wikipedia.org/wiki/Hipot

HTH,
Paul

Hey, thanks a lot for that info. It does help.
That pretty much confirms what I "thought" was the case. But I don't
know that much about electrical engineering, so what do I "know"!?
Now I can pretend I know what I'm talking about with much more conviction.
Thanks again.

You'd think that the power requirements of every PC component would be
needed, useful information for everyone to have - industrial or
consumer. It is curious why that data is not standard in the component
specs. Or is it only provided in the technical data sheets?

 

[Paul]

Hey, thanks a lot for that info. It does help.
That pretty much confirms what I "thought" was the case. But I don't
know that much about electrical engineering, so what do I "know"!?
Now I can pretend I know what I'm talking about with much more conviction.
Thanks again.

You'd think that the power requirements of every PC component would be
needed, useful information for everyone to have - industrial or
consumer. It is curious why that data is not standard in the component
specs. Or is it only provided in the technical data sheets?

Power consumption of things, is a closely guarded secret
I'm not kidding. At work, one department would not tell
another department, what the power consumption was for something
they were building. Even after a product is shipping, and there
are tons of units to take measurements from, you still would
find the info closely guarded.

It has to do partly, with the liabilities of releasing the
information, and the error bounds on the info. Take a video
card for example. Xbitlabs measures video cards, but they
might only measure one sample. Say the power consumed was
100W while running 3DMark on the Xbitlabs test jig. Well,
if an engineer at Nvidia did a worst case analysis, perhaps
a reasonable estimate, assuming reasonable activities on the
card, might be 200W for one card in a thousand (the "worst
card of the bunch"). Should that engineer quote 200W in the
documentation ? The marketing department would have him killed,
if he did that You'll notice on the latest Nvidia video
card launch, that power is the single most important issue
with their new cards. A very touchy subject for them.

So rather than release "marketing negative" information,
or release information that turns out not to be true,
it is simply easier not to release any information at all.
Then, outfits like Xbitlabs can expose a ballpark figure,
and generally, people won't end up buying a too-small supply
as a result.

As for the video card example, the onboard regulator has
power limits as well. It might turn out, that the regulator
would overload, before any theoretical peak could be met.
The video card might well be designed to "throttle" if that
happened. Or, the driver might be modified to prevent
problems. That happened with the "Furmark" utility, which
had the potential to damage video cards. The driver has
something added, to detect that Furmark is running, and
prevent it from ruining the hardware (introduce throttling).

The marketing department usually has control of information
release, and probably have decided there is no up-side to
releasing power numbers. And so we lurch along on guesses
and hunches.

The optical drive industry solution to this problem, was
to use the "boiler plate" approach. They would stick
"12V @ 1.5A, 5V @ 1.5A" on all the various products, which
is a way of saying "we're not telling you, and here is
some crap information to munch on". So that is another way
to handle it. When things like BluRay came out, the
boiler plate was changed for them, but you're likely
to run into the same thing. No manufacturer would want
to use an honest number, because a competitor could
come out with a drive drawing 2 watts less

Hard drives are a bit more honest. Why, I don't know. Low
power consumption on a hard drive, isn't always a good idea,
as some of the "greener" products compromise performance
for power. But the choice is there, if you want to shop by power
number. There is occasional dishonesty, like not listing
spinup power on 2.5" laptop drives. Which is actually
an important number to know, to make USB enclosures
for 2.5" drives work properly.

On processors, there may be a listed TDP, but that Thermal
Design Power might only be useful for heatsink design purposes.
My 65W processor, draws 36W max at 100% load, and 6W at idle.
I've had other processors, where the measured power was
a little bit higher than TDP. There may be datasheets, with
more precise numbers shown, but try and figure out how to
make a power number out of them. Simply multiplying
Vcore by Icore, doesn't give reasonable results (and that
is partially, because the actual value of Vcore is
something different - processors use load lines and
have SpeedStep/Cool N' Quiet to complicate matters).
So, the next best thing, is to just use TDP and hope
the real power is lower than that.

Paul

 

[TVeblen]

Power consumption of things, is a closely guarded secret
I'm not kidding. At work, one department would not tell
another department, what the power consumption was for something
they were building. Even after a product is shipping, and there
are tons of units to take measurements from, you still would
find the info closely guarded.

It has to do partly, with the liabilities of releasing the
information, and the error bounds on the info. Take a video
card for example. Xbitlabs measures video cards, but they
might only measure one sample. Say the power consumed was
100W while running 3DMark on the Xbitlabs test jig. Well,
if an engineer at Nvidia did a worst case analysis, perhaps
a reasonable estimate, assuming reasonable activities on the
card, might be 200W for one card in a thousand (the "worst
card of the bunch"). Should that engineer quote 200W in the
documentation ? The marketing department would have him killed,
if he did that You'll notice on the latest Nvidia video
card launch, that power is the single most important issue
with their new cards. A very touchy subject for them.

So rather than release "marketing negative" information,
or release information that turns out not to be true,
it is simply easier not to release any information at all.
Then, outfits like Xbitlabs can expose a ballpark figure,
and generally, people won't end up buying a too-small supply
as a result.

As for the video card example, the onboard regulator has
power limits as well. It might turn out, that the regulator
would overload, before any theoretical peak could be met.
The video card might well be designed to "throttle" if that
happened. Or, the driver might be modified to prevent
problems. That happened with the "Furmark" utility, which
had the potential to damage video cards. The driver has
something added, to detect that Furmark is running, and
prevent it from ruining the hardware (introduce throttling).

The marketing department usually has control of information
release, and probably have decided there is no up-side to
releasing power numbers. And so we lurch along on guesses
and hunches.

The optical drive industry solution to this problem, was
to use the "boiler plate" approach. They would stick
"12V @ 1.5A, 5V @ 1.5A" on all the various products, which
is a way of saying "we're not telling you, and here is
some crap information to munch on". So that is another way
to handle it. When things like BluRay came out, the
boiler plate was changed for them, but you're likely
to run into the same thing. No manufacturer would want
to use an honest number, because a competitor could
come out with a drive drawing 2 watts less

Hard drives are a bit more honest. Why, I don't know. Low
power consumption on a hard drive, isn't always a good idea,
as some of the "greener" products compromise performance
for power. But the choice is there, if you want to shop by power
number. There is occasional dishonesty, like not listing
spinup power on 2.5" laptop drives. Which is actually
an important number to know, to make USB enclosures
for 2.5" drives work properly.

On processors, there may be a listed TDP, but that Thermal
Design Power might only be useful for heatsink design purposes.
My 65W processor, draws 36W max at 100% load, and 6W at idle.
I've had other processors, where the measured power was
a little bit higher than TDP. There may be datasheets, with
more precise numbers shown, but try and figure out how to
make a power number out of them. Simply multiplying
Vcore by Icore, doesn't give reasonable results (and that
is partially, because the actual value of Vcore is
something different - processors use load lines and
have SpeedStep/Cool N' Quiet to complicate matters).
So, the next best thing, is to just use TDP and hope
the real power is lower than that.

Paul

Well that explains things well and makes sense (or nonsense, depending
on how you look at it!).

One thing you said that caught my attention: video cards regarding
"onboard regulators" and "throttling". The problem under discussion
involves many ATI 5000 series cards running under Windows 7 that are
"cutting out" to blackscreen and only a reboot can unfreeze the system.
No bluescreen crashes or minidumps, no error events, no TDR events. All
the usual suspects are of little help: driver upgrades & proper
re-installs, overclocking, overheating issues. The one issue that caught
my eye was the users power supply (DEll system): 360W and running an ATI
5770 (min 450W spec). But someone posted that Dell PS's are
"under-rated" and the power should not be an issue. This post was to
check that idea against my limited understanding of how PS's work.

Thanks again Paul.