1 computer, 2 PSUs - feasible?

[pawihte]

Is it possible to use two cheap low-power PSUs instead of a
single premium unit?

Even in the remote area where I live, computers have become an
essential part of everyday life, but many computer owners have
had to dig deep into their savings to buy an entry-level machine.
They either don't know about PSU quality and actual capacity, or
can ill afford to spend that extra $50-150 for a good PSU. It's
hard to blame them as I've personally known literally hundreds of
$10 PSUs last for years. (The $ here meaning the equivalent in US
currency). There's nothing available in between the $10 types and
the >$50 models.

Forget about build quality and load regulation, etc. for the time
being. Is it at all feasible to use one cheap PSU for, say, a
power-hungry graphics card and another one for the rest of the
system? What about these factors -

1. Syncing. It shouldn't be hard to make a simple adaptor to
switch two PSUs from the same motherboard. But there may be a
difference of a fraction of a second in the times taken by the
two PSUs to power up and power down.

2. Partial PSU breakdown. If one PSU breaks down while the other
is still operating, could that cause damage to the rest of the
system?

 

[Paul]

Is it possible to use two cheap low-power PSUs instead of a
single premium unit?

Even in the remote area where I live, computers have become an
essential part of everyday life, but many computer owners have
had to dig deep into their savings to buy an entry-level machine.
They either don't know about PSU quality and actual capacity, or
can ill afford to spend that extra $50-150 for a good PSU. It's
hard to blame them as I've personally known literally hundreds of
$10 PSUs last for years. (The $ here meaning the equivalent in US
currency). There's nothing available in between the $10 types and
the >$50 models.

Forget about build quality and load regulation, etc. for the time
being. Is it at all feasible to use one cheap PSU for, say, a
power-hungry graphics card and another one for the rest of the
system? What about these factors -

1. Syncing. It shouldn't be hard to make a simple adaptor to
switch two PSUs from the same motherboard. But there may be a
difference of a fraction of a second in the times taken by the
two PSUs to power up and power down.

2. Partial PSU breakdown. If one PSU breaks down while the other
is still operating, could that cause damage to the rest of the
system?

Find a copy of the latest ATX spec from formfactors.org (in the
Specifications and Guides section).

PS_ON# is an input, and is wired-OR type. It has a pullup, and
a driving device pulls down with an open collector. If you join
two power supplies together, you'd put their PS_ON# signals
in parallel. The motherboard would then end up driving the
pullup resistors on both PSUs, which is twice the load.

PWK_OK is an active high output from the supply. The ATX spec doesn't
specify it is wire-OR logic with open collector drive (which is essential
for safely tying together a set of outputs without damaged). At least,
I couldn't see that suggested there. The specification also seems to be
worded, as if they expect you to tie a pulldown to the circuit (to
guarantee PWR_OK is logic 0 when the power supply is not working at
all). And that suggests totem-pole drive, with an external resistor
pulldown to guarantee logic 0 when the supply is off.

If I look at the schematic on this site, it looks like PWR_OK
comes from an LM393 comparator, with a 680 ohm pullup, so this
supply happens to implement it in such a way, that it can be
tied together to another supply. This design fails to include the
1K to common, to force the output to logic 0 when the supply is
off. Again, there is the issue, that the LM393 would be driving
5V/340 ohms = 15ma. The datasheet doesn't suggest it works that
well at high current (poor logic 0).

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

http://www.national.com/ds/LM/LM193.pdf

The DC output characteristic of the ATX supply, is a voltage source,
and it is a "push-only" design. It is not "push-pull". For example,
if you short the 5V rail to the 3.3V rail, the 3.3V rail would be
lifted, and the power supply would not actively fight it. If the supply
had OVP (and not all of them do), the supply would shut off once
the 3.3V went too far out of spec. But the supply doesn't actively
pull down on that rail, to try to bring it back into spec.

Since the power supply is a voltage source, you can't tie it in
parallel with another supply. You would get current hogging if
you did (effectively, one supply can take all the load, while
the other one just sits there). The outputs can be kept independent
from one another, and be used to power things known to be independent
inside the computer. (Redundant power supplies, have a different
output design, and do current sharing, so are safe to connect in parallel.
The designers know in advance, that the supplies will be paralleled. When
both supplies are present, they each deliver 50% of the current, plus/minus
about one percent.)

The ATX12V used by the motherboard, is independent (except on
some Biostar motherboards, where it was discovered their
motherboards would run with only the main power connector in
place). The PCI Express external connectors (2x3, 2x4) are supposed
to be independent as well. Those are examples of things you could
power. The independent things, all seem to be using 12V.

(Independence of PCI Express rails is on page 9.)

http://www.pcisig.com/developers/ma...c_id=fa4ec3357012d69821baa0856011c665ac770768

An example of a power supply designed to operate as a "second supply",
is the Fortron Booster. It senses the presence of 12V, and enables
its 12V output milliseconds later. The 12V it produces, is suitable
for independent load inputs. It does not tie into PWR_OK, neither
does it use PS_ON# for control purposes. The wiring provided,
is for powering the PCI Express auxiliary connectors on a pair
of video cards. So if you had some 150W video cards, you might
run two of them off that thing (leaving a little capacity left over).
If the unit failed, it could potentially remain on, while the
other supply is off. But PCI Express video cards, as far as I know,
are supposed to be able to take it.

http://www.newegg.com/Product/Product.aspx?Item=N82E16817104054

It may be acceptable to try your experiment on your own home rig, but if
you do it on a paying customer's machine, remember you're
responsible for the insurance liability. If someone's house
burns down because of some fault with the two supplies, there
will be a knock at the door, and some lawyers to answer to.
The only thing that should go into a customer machine, is approved
circuit configurations. Where you can claim, "everything I've done
is standard industry practice". Then, if they want to sue someone,
it'll be the power supply manufacturer.

One other small benefit of the $50 supply, is the potential for
it to be 80% efficient, versus the 68% efficiency of the $10
supply. Over the years, that will give enough savings on the
power bill, to pay for the $50 supply. So it isn't all bad.

80% efficient supply for $45
http://www.newegg.com/Product/Product.aspx?Item=N82E16817371033

Paul

 

[pawihte]

Find a copy of the latest ATX spec from formfactors.org (in the
Specifications and Guides section).

PS_ON# is an input, and is wired-OR type. It has a pullup, and
a driving device pulls down with an open collector. If you join
two power supplies together, you'd put their PS_ON# signals
in parallel. The motherboard would then end up driving the
pullup resistors on both PSUs, which is twice the load.

PWK_OK is an active high output from the supply. The ATX spec
doesn't
specify it is wire-OR logic with open collector drive (which is
essential for safely tying together a set of outputs without
damaged). At least, I couldn't see that suggested there. The
specification also seems to
be worded, as if they expect you to tie a pulldown to the
circuit (to
guarantee PWR_OK is logic 0 when the power supply is not
working at
all). And that suggests totem-pole drive, with an external
resistor
pulldown to guarantee logic 0 when the supply is off.

If I look at the schematic on this site, it looks like PWR_OK
comes from an LM393 comparator, with a 680 ohm pullup, so this
supply happens to implement it in such a way, that it can be
tied together to another supply. This design fails to include
the
1K to common, to force the output to logic 0 when the supply is
off. Again, there is the issue, that the LM393 would be driving
5V/340 ohms = 15ma. The datasheet doesn't suggest it works that
well at high current (poor logic 0).

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

http://www.national.com/ds/LM/LM193.pdf

The DC output characteristic of the ATX supply, is a voltage
source,
and it is a "push-only" design. It is not "push-pull". For
example,
if you short the 5V rail to the 3.3V rail, the 3.3V rail would
be
lifted, and the power supply would not actively fight it. If
the
supply had OVP (and not all of them do), the supply would shut
off
once the 3.3V went too far out of spec. But the supply doesn't
actively
pull down on that rail, to try to bring it back into spec.

Since the power supply is a voltage source, you can't tie it in
parallel with another supply. You would get current hogging if
you did (effectively, one supply can take all the load, while
the other one just sits there). The outputs can be kept
independent
from one another, and be used to power things known to be
independent
inside the computer. (Redundant power supplies, have a
different
output design, and do current sharing, so are safe to connect
in
parallel. The designers know in advance, that the supplies will
be
paralleled. When both supplies are present, they each deliver
50% of
the current, plus/minus about one percent.)

The ATX12V used by the motherboard, is independent (except on
some Biostar motherboards, where it was discovered their
motherboards would run with only the main power connector in
place). The PCI Express external connectors (2x3, 2x4) are
supposed
to be independent as well. Those are examples of things you
could
power. The independent things, all seem to be using 12V.

(Independence of PCI Express rails is on page 9.)

http://www.pcisig.com/developers/ma...c_id=fa4ec3357012d69821baa0856011c665ac770768

An example of a power supply designed to operate as a "second
supply",
is the Fortron Booster. It senses the presence of 12V, and
enables
its 12V output milliseconds later. The 12V it produces, is
suitable
for independent load inputs. It does not tie into PWR_OK,
neither
does it use PS_ON# for control purposes. The wiring provided,
is for powering the PCI Express auxiliary connectors on a pair
of video cards. So if you had some 150W video cards, you might
run two of them off that thing (leaving a little capacity left
over).
If the unit failed, it could potentially remain on, while the
other supply is off. But PCI Express video cards, as far as I
know,
are supposed to be able to take it.

http://www.newegg.com/Product/Product.aspx?Item=N82E16817104054

It may be acceptable to try your experiment on your own home
rig, but
if you do it on a paying customer's machine, remember you're
responsible for the insurance liability. If someone's house
burns down because of some fault with the two supplies, there
will be a knock at the door, and some lawyers to answer to.
The only thing that should go into a customer machine, is
approved
circuit configurations. Where you can claim, "everything I've
done
is standard industry practice". Then, if they want to sue
someone,
it'll be the power supply manufacturer.

One other small benefit of the $50 supply, is the potential for
it to be 80% efficient, versus the 68% efficiency of the $10
supply. Over the years, that will give enough savings on the
power bill, to pay for the $50 supply. So it isn't all bad.

80% efficient supply for $45
http://www.newegg.com/Product/Product.aspx?Item=N82E16817371033

Paul

Thanks, Paul. I've been down with a severe case of the flu for 3
days now. Still don't have the energy to go through the points
you raised and continue the discussion. Will be back later.