"skotl" said:
Hi Paul,
thanks for that.
Think I'm getting there, after a bit of R&D and soldering. There are 4
VIA boards, and their combined rail requirements (plus the disks) are
more than met by the 450w supply I'm putting in.
I could have put 4 ATX supplies, or fed DC-DC supplies (although the
overall power consumption would be higher, given the losses in the
DC-DC tx's) but I'm after a neat, pretty and over-engineered solution.
I bought a brand new 6U server chassis and everything (including the
KVM switch and USB switch for the CDROM) is going in here.
I have also figured out the diode shenanigans; PWROK is just going to
be fed to every board, but PSON# has a 1n4148 flowing *from* the PSU to
each board. This means that as long as one of them is still connecting
their PSON# to ground, current flows thru the diode and the PSU
supplies juice.
When the last of the four raises PSON# (actually, allows it to float)
then PSON# at the PSU will float up and the power will be cut.
Worst effect on the other boards that were waiting is that they sit
with the OpSys saying "it is OK to remove power".
Final trick is that I couldn't figure how to split all the power cables
without using veroboard (doubts over current capacity) or splicing them
together (yuck), so I have invented my own bus with eight 2.5mm rails,
and the cables are looped and soldered to the rails.
There are eight cos that is the number of discrete supplies / signals
from an ATX PSU.
All works, with 2 boards on the bench. I'll let you know when I tie 'em
all up for real!
Cheers,
Scott
Well, I really don't think the diodes are necessary on PS_ON#. But
if you feel it is safer, and it actually works for you, then go ahead.
Page 25 here states the spec for the PS_ON# input:
http://www.formfactors.org/developer\specs\ATX12V PSDG2.01.pdf
The PS_ON# signal should be pulled to 0.8V or lower, to turn
on the supply. The diode in series with the output driver from
the EPIA, will make that spec a bit harder to meet.
I looked at an 8712F Super I/O chip, which has a PS_ON# driver on
it, and it uses an open collector driver when in PS_ON# controlling
mode. Leakage is 10uA (which is negligible and can be ignored).
The driver can sink 8mA on that device. And the ATX spec says you
should be able to sink at least 1.6mA.
Perhaps using a Schottky diode, instead of the 1N4148, would help
reduce the voltage drop. I suppose, as long as it works
for you, then carry on
(Your use of the diode handles the case
where the motherboard uses a normal push/pull driver, instead of
the proper open collector one. I don't think there is anything
to prevent a motherboard maker from using a normal push/pull
driver. Using open collector or open drain is just by convention.
So your use of the diode on PS_ON# does cover off that possibility.)
I'm a bit curious about how the two computers deal with the
common power.
Say we have Comp1 and Comp2...
1) We turn on via Comp1. Both Comp1 and Comp2 power up ?
Does Comp2 complain in any way, that its power control circuit
is not currently asserting PS_ON# and yet the computer is powered ?
2) With Comp1 controlling PS_ON#, as in (1), if you select shutdown
on Comp2 menu, do you see "It is safe to turn off..." message on the
video output of Comp2 ?
3) With Comp1 controlling PS_ON#, as in (1), what happens if you
press the power switch on Comp2 ? I presume if it wasn't asserting
PS_ON#, then it might start to. If it wasn't asserting PS_ON#, how
does it deal with the Power Management event that says "hey, we
are powered, and we just received a request to power up" ?
For this to work properly, you would think a computer that finds itself
powered, would be best advised to start asserting PS_ON# as soon as
it POSTs. I'm curious just how the four computers deal with the
lack of synchronization between power states and the condition of
their logic. (I'm not suggesting something destructive happens,
just that the software on the computer will be in states you might
not normally run into in a one_supply/one_motherboard system.)
As for the pico power converter, I think it is a fairly high
efficiency device, and the lack of visible heat sinks tells you
that. If it is 90+% efficient, and you used a Seasonic 80% efficient
supply as a "brick", I think the 72% efficient overall result, might
still beat a crappy 68% efficient cheap ATX supply
So while
the pico power idea is expensive, it isn't the worst source of
power you could find from an efficiency viewpoint. I have seen
concatenated conversion used in commercial non-ATX power supplies
(one switcher circuit feeds a second switcher, before driving the
load). So it has been done.
Paul
