Arthur Hagen

Daniel James <wastebasket@nospam.aaisp.org> wrote:
> In article
> news:<slrnfkluma.q2l.aznomad.2@ip70-176-155-130.ph.ph.cox.net>, AZ
> Nomad wrote:
>
>> power supply: don't use the crap that is included with most cases.
>> A blowout on that $8 power supply can take out your whole system.
>> Spend $50-90 here.
>
> I'll add a rider to that -- spend those dollars on a good quality PSU
> rather than a lower-quality unit that claims a higher power output.
> Work out roughly how much poser your system will actually use, add
> 10-20% as a margin for error, and buy the best PSU you can that
> exceeds that power by as little as possible.

I strongly advise people to choose a PSU that delivers more than 10-20%
above what you will actually use. The main reason is that PSUs deteriorate
over time, and after a year or two, you /will/ run into problems unless you
have a stronger powersupply that can show wear, but still deliver enough.
And no, buying a quality PSU won't stop the deterioration in output over
time. (Buying a very cheap one might accellerate it, though.)

> My Athlon64 4200+ system in the M2NPV-VM (with integrated graphics --
> faster GPUs eat energy) idles at less than 100W (as measured by my
> UPS) and probably uses less than 200W maximum. My 430W PSU is huge
> overkill! (but I got it cheap)

No, it's not. The total wattage is irrelevant. You need to measure the
current polled /per rail/. When a PSU is "430W", that's the total power for
the powersupply on all rails, and what's unused on one rail won't become
available on another. It may be able to deliver 170W on two 12V rails, 75W
shared between the 3.3V and 5V rails, and 15W for the negative currents.
When new and cold. If old and heated up, it will be able to deliver far
less.

Look at how much power is available /per rail/ when buying the PSU, and
estimate your needs based on that. Usually, the manufacturer hides this
information in plain sight, relying on the customers not to be tech savvy
enough to realise that you get the wattage by multiplying the voltage (V) by
the current (A) and divide it by the number of rails for that voltage. As
an example, if a PSU states +12V : 14A and has two 12V rails (typically one
to the motherboard and one to hard drives and other peripherals), you get
12*14/2 W per rail, or max 84 W. While a HD might be rated at 7W, it may
require twice that for spinning up, and with many hard drives in addition to
other external +12V devices, 84W might not be enough.

Also beware of rounded numbers. If the label says 20A, 10A, 5A for three
rails, it's likely to be numbers rounded *up* to the nearest multiple of 5,
and 20A might in reality mean anything from 16-20A, and for 16A, that's a
48W difference for a 12V rail.

In other words, if you estimate that your rig will at max use 200W, it would
be sheer lunacy to think that a 230W power supply would be big enough. You
may be able to run on it for a while, until you do something that takes all
the power on one of the rails (like smartd spinning up all drives at the
same time, whether mounted or not, while you're burning a DVD), or until the
PSU becomes old or hot.

What's likely to occur if the PSU is slightly underrated is that one of the
voltages will drop to the point that some circuitry that uses it will
experience errors. This may be the RAM, a HD, or just about anything, and
you might not even think of the PSU being the culprit.

That said, there's no point in buying a PSU that's much bigger than what you
will /actually/ need on each rail, keeping in mind that the specs are likely
overrated (yes, even with "top quality" PSUs) and that you add a little
extra for reduced efficiency as the PSU gets older and/or hotter, and any
peripherals you might want to add in the future (like an extra HD or a PCI
card).

Regards,
--
*Art

AZ Nomad

On Tue, 27 Nov 2007 12:10:09 -0500, Arthur Hagen <art@broomstick.com> wrote:


>Daniel James <wastebasket@nospam.aaisp.org> wrote:
>> In article
>> news:<slrnfkluma.q2l.aznomad.2@ip70-176-155-130.ph.ph.cox.net>, AZ
>> Nomad wrote:
>>
>>> power supply: don't use the crap that is included with most cases.
>>> A blowout on that $8 power supply can take out your whole system.
>>> Spend $50-90 here.
>>
>> I'll add a rider to that -- spend those dollars on a good quality PSU
>> rather than a lower-quality unit that claims a higher power output.
>> Work out roughly how much poser your system will actually use, add
>> 10-20% as a margin for error, and buy the best PSU you can that
>> exceeds that power by as little as possible.

>I strongly advise people to choose a PSU that delivers more than 10-20%
>above what you will actually use. The main reason is that PSUs deteriorate

Watch out. I'd prefer a solid 400W power supply over an 8oz 1000W supply
considering your typical system draws less than 200W while running and less
than 300W during startup. A POS 1000W supply may do 1000W according to the
capacity of it's output devices (read: a couple of $3 triacs), but will have a
cheapo fan that'll soon fail and thermal problems with anything more than 150W.
Being built by near slave labor on a too quickly moving assembly line won't help
either.

Daniel James

In article news:<fihj1i$6u6$1@tree.broomstick.com>, Arthur Hagen wrote:
> The total wattage is irrelevant. You need to measure the
> current polled /per rail/.

You do indeed -- and I'm sorry I forgot to say so.

However, my own over-spec'd PSU gives at least twice what's needed on
all rails ... and that really *is* more than is necessary.

What I was trying to warn the OP of is that -- just as buying a cheap
PSU is a false economy (which is to say no economy at all) -- spending
more to buy a higher power PSU of no better quality is a complete waste
of money.

Yes, the PSU does have to be sufficiently powerful to run the system,
but I know of several people who advocate 480W and 600W (and even more
powerful) PSUs for systems that, on closer inspection, would run
perfectly well on a good quality 350W (or even smaller) unit.

You *can* buy 1kW PSUs ... but it's fairly hard to spec a system that
needs anything like that (having two top-whack graphics cards would get
you almost half-way, but using the other half would be hard).

> Also beware of rounded numbers. If the label says 20A, 10A, 5A for
> three rails, it's likely to be numbers rounded *up* to the nearest
> multiple of 5, and 20A might in reality mean anything from 16-20A,
> and for 16A, that's a 48W difference for a 12V rail.

That's something I haven't come across. I think that the Trades
Descriptions Act (or whatever stands in its place, nowadays) here in
the UK would mean that anyone selling a PSU claiming to provide 20W on
some rail when it only provided 16W would quickly be required to
withdraw it from sale.

> That said, there's no point in buying a PSU that's much bigger than
> what you will /actually/ need on each rail, keeping in mind that the
> specs are likely overrated (yes, even with "top quality" PSUs) and
> that you add a little extra for reduced efficiency as the PSU gets
> older and/or hotter,

So ... how much overcapacity would YOU allow on each rail, when
spec'ing a PSU?

> ... and any peripherals you might want to add in the future (like
> an extra HD or a PCI card).

Allowing slack for expansion is a different matter, of course.

Cheers,
Daniel.

Arthur Hagen

AZ Nomad <aznomad.2@PremoveOBthisOX.COM> wrote:
> On Tue, 27 Nov 2007 12:10:09 -0500, Arthur Hagen <art@broomstick.com>
> wrote:
>
>> I strongly advise people to choose a PSU that delivers more than
>> 10-20% above what you will actually use. The main reason is that
>> PSUs deteriorate
>
> Watch out. I'd prefer a solid 400W power supply over an 8oz 1000W
> supply considering your typical system draws less than 200W while
> running and less than 300W during startup. A POS 1000W supply may do
> 1000W according to the capacity of it's output devices (read: a
> couple of $3 triacs), but will have a cheapo fan that'll soon fail
> and thermal problems with anything more than 150W. Being built by
> near slave labor on a too quickly moving assembly line won't help
> either.

Oh, I agree totally. Plus that cheap PSUs with very high ratings are more
likely to have a *peak* rating stated, which is only sustainable for a very
short time, instead of a honest sustainable delivery wattage. Think of it
as PMPO ratings on a stereo -- it has nothing to do with reality.

Overpowered PSUs also tend to have a single 12 V rail. While this allows
more power to be pulled, it is bad for various reasons, among them because
it means that a voltage drop due to a temporary high power draw will affect
everything on the system -- in other words, when your drives spin up, your
motherboard may get lower voltage. Needless to say, this is Bad for your
system.

Also worth noting is that overpowered PSUs tend to draw more power. While a
400W and an 850W power supply both may have an 80% efficiency rating, that's
under full load. If you draw 200W from each of them, the 850W one will in
all likelyhood draw far more power from the mains, which generates excess
heat and costs you more.

So err on the side of caution, but not on the side of overkill. If you
/think/ you may need 200W, a good quality[1] 350-430W PSU may be just about
right.

[1]: One good indicator of a bad PSU is the absense of UL, GS and TÜV
certification markers. Another good indicator of quality is whether it has
been on the market for some time; bad products don't survive for long, but
may reappear with new labels and product names. Price is also an
indicator -- while there /are/ good deals to be had, ask yourself /why/
something is sold for a low price. "Because they want to be nice to you"
isn't the answer.

Regards,
--
*Art

Arthur Hagen

Daniel James <wastebasket@nospam.aaisp.org> wrote:
> In article news:<fihj1i$6u6$1@tree.broomstick.com>, Arthur Hagen
> wrote:
>
> So ... how much overcapacity would YOU allow on each rail, when
> spec'ing a PSU?

If having to state a rule of thumb, I'd say at least 50% extra.
It depends on what you run on each rail too -- it might be prudent to have a
higher safety margin for the rails that provide bus power or power a high
end graphics card -- a voltage drop due to a temporary peak draw can be more
critical there (and much harder to troubleshoot as a PSU problem).

I'm going to have to replace the PSU on this box pretty soon, because with
everything I've added over time, the +5VSB (standby power) doesn't deliver
enough juice, meaning I can no longer start the system after shutting it
down unless it's been powered all the way down with the hard switch on the
PSU.

Regards,
--
*Art