Marginal OEM Power Supply

[Robert Myers]

Last summer I bought a box from a very big OEM at a good price.

The box would not boot on delivery because the RAM had been installed
incorrectly, so it's clear that, if the box was tested, it was tested
before the RAM was installed, and now I'm wondering at what stage of
assembly the boxes are supposedly tested (not at all after assembly?).

I increased the memory by 50%, reinstalled the incorrectly installed
memory, and added a second HDD, and the machine did boot.

I haven't used the machine much, and, in particular, I haven't used the
video much, but this summer I started encountering video problems while
trying to upgrade Fedora.

Since I had an extra video card available, I tried changing video card.
My replacement video card drew too much power and the machine would
not even boot. The easy conclusion was that the power supply was either
malfunctioning or marginal to begin with.

The computer store is a fair drive, so I bought both a replacement power
supply and a very low end video card. I don't like replacing power
supplies, so I tried the video card. Works like magic. I'll replace
the power supply if I get around to it. I mostly run the box remotely
and I don't need fancy video. As far as warranties are concerned, I
can't imagine what circumstances would make a warranty repair worth my
time, so I don't worry about it.

Maybe when you pay for installed upgrades, you pay for an upgraded power
supply. Since the memory was added to this box with no further testing,
it's easy to imagine that boxes might be shipped with upgrades such that
the power supply is often marginal or inadequate.

Anybody have any insight?

Robert.

 

[Bill Davidsen]

Last summer I bought a box from a very big OEM at a good price.

The box would not boot on delivery because the RAM had been installed
incorrectly, so it's clear that, if the box was tested, it was tested
before the RAM was installed, and now I'm wondering at what stage of
assembly the boxes are supposedly tested (not at all after assembly?).

The stage where the end user tries to boot the box, at least from the vendor you
used.

I increased the memory by 50%, reinstalled the incorrectly installed
memory, and added a second HDD, and the machine did boot.

I haven't used the machine much, and, in particular, I haven't used the
video much, but this summer I started encountering video problems while
trying to upgrade Fedora.

Since I had an extra video card available, I tried changing video card.
My replacement video card drew too much power and the machine would not
even boot. The easy conclusion was that the power supply was either
malfunctioning or marginal to begin with.

That's a fair guess, many are sized to the need as sold. That's not an unfair
practice, just one which should be more obvious before it bites you.

The computer store is a fair drive, so I bought both a replacement power
supply and a very low end video card. I don't like replacing power
supplies, so I tried the video card. Works like magic. I'll replace
the power supply if I get around to it. I mostly run the box remotely
and I don't need fancy video. As far as warranties are concerned, I
can't imagine what circumstances would make a warranty repair worth my
time, so I don't worry about it.

Unless you have some reason to use a brick and mortar store, places like newegg
and tigerdirect are good sources for parts.

Maybe when you pay for installed upgrades, you pay for an upgraded power
supply. Since the memory was added to this box with no further testing,
it's easy to imagine that boxes might be shipped with upgrades such that
the power supply is often marginal or inadequate.

Anybody have any insight?

Except for gamer machines, vendors expect the cover to stay on. The bad memory
install would have gotten a complaint from me, I doubt anyone ships without a
smoke test, at least not anyone remotely reputable. I bet someone saved a bit of
time by not doing it as s/he should.

 

[Robert Myers]

Robert Myers wrote:

That's a fair guess, many are sized to the need as sold. That's not an
unfair practice, just one which should be more obvious before it bites you.

Never occurred to me that they would size the power supply to the actual
sale. That means that, even if you pay outrageous prices for a
subsequent OEM add-on, it might not work. I'd call that deceptive
marketing.

Unless you have some reason to use a brick and mortar store, places like
newegg and tigerdirect are good sources for parts.

A Microcenter is reasonably close, and, if you shop carefully, their
prices are hard to beat.

**And**, if something goes wrong, I know how to find the store manager.

Except for gamer machines, vendors expect the cover to stay on. The bad
memory install would have gotten a complaint from me, I doubt anyone
ships without a smoke test, at least not anyone remotely reputable. I
bet someone saved a bit of time by not doing it as s/he should.

The default memory configuration for this machine was two sticks. A
third stick was added to this three-channel machine as an "upgrade."

DIMM slots 0 and 2 had been installed correctly and the "upgrade" was
installed to slot 3.

This purchase was from a *very* well-known OEM. It's really hard to
imagine who, other than someone like me or a gamer, would buy such a
box. The only reason I don't name the company is that I know engineers
who work there.

Robert.

 

[Robert Redelmeier]

Never occurred to me that they would size the power supply
to the actual sale. That means that, even if you pay
outrageous prices for a subsequent OEM add-on, it might
not work. I'd call that deceptive marketing.

It might well be. Perhaps this was a batch of computers
specifically built for some megacorp refresh but which proved
out to have a high defect rate -- either as RAM installation
or PSU quality. It was subsequently rejected and dumped on
the resale market where you bought it cheaply.

They ought to have been sold as "debranded", but perhaps
someone was asleep at bigOEM. Or a disobedient reseller.

The Shenzhen suicides and subsequent salary doubling
do not indicate high-quality.


-- Robert

 

[Robert Myers]

It might well be. Perhaps this was a batch of computers
specifically built for some megacorp refresh but which proved
out to have a high defect rate -- either as RAM installation
or PSU quality. It was subsequently rejected and dumped on
the resale market where you bought it cheaply.

They ought to have been sold as "debranded", but perhaps
someone was asleep at bigOEM. Or a disobedient reseller.

Not a reseller. Direct purchase from the OEM.

Your thought that my deal might have been the result of a messed up
corporate order would explain some weird things. For example, why would
anyone put two 2Gb memory sticks in a three-channel box to begin with?

The boxes don't make much sense for a big corporate purchase, but maybe
they weren't the right boxes to begin with. The order was for a more
normal corporate machine with 4Gb of memory and a core 2 dual processor.
4Gb was stuck into the wrong boxes and the boxes were tested before
someone said oops. Test the memory, tell your guys to stick it into the
boxes and offer a special deal. Who knows. The extra stick of memory
seems obviously to have been stuck in. Someone who knew to skip slot 1
would also know to skip slot 3.

Robert.

 

[Bill Davidsen]

Never occurred to me that they would size the power supply to the actual
sale. That means that, even if you pay outrageous prices for a
subsequent OEM add-on, it might not work. I'd call that deceptive
marketing.

A Microcenter is reasonably close, and, if you shop carefully, their
prices are hard to beat.

**And**, if something goes wrong, I know how to find the store manager.

That qualifies as a reason for sure, didn't think "reasonably close," since you
earlier said "The computer store is a fair drive," but I know what you mean, not
too far to go to save a shipping cycle but far enough that you don't want to do
it twice.

The default memory configuration for this machine was two sticks. A
third stick was added to this three-channel machine as an "upgrade."

DIMM slots 0 and 2 had been installed correctly and the "upgrade" was
installed to slot 3.

I would expect that to pass smoke test, then, incorrectly installed as in not
full speed rather than upside down or something, I'm actually surprised that it
didn't show up working on POST and only be obvious with the cover off or using
dmidecode. Or whatever the Windows tool is to get the same information.

 

[Robert Myers]

I would expect that to pass smoke test, then, incorrectly installed as
in not full speed rather than upside down or something, I'm actually
surprised that it didn't show up working on POST and only be obvious
with the cover off or using dmidecode. Or whatever the Windows tool is
to get the same information.

I would have expected it to boot, too, but it didn't. I tried to boot
before opening the machine. It's been long enough that I don't remember
if it got past a power-on POST and stopped at a blank screen. Maybe
that would have constituted passing a smoke test. I can understand why
a tester in a hurry wouldn't want to wait for Vista to boot.

Since I intended to add memory, I didn't hesitate to take the cover off
to see what was going on. Making what I thought would be a correct
memory install fixed the problem.

Robert.

 

[Robert Myers]

I can quite imagine that an OEM box would have the bare minimum power
supply specs require. I mean this must be one of several reasons why an
OEM box is cheaper than any box that you can put together yourself. You
probably would buy a PS with a little bit of extra margin in it, and
it's likely that you probably couldn't even buy a PS as low as what's in
an OEM box.

I had previously noted the puniness of power supplies on OEM boxes, but
not by overloading them to the point where the machine wouldn't run. I
don't think that upping the RAM by 50% and adding a HDD causing a power
supply overload is a reasonable expectation.

Something else about this machine that is shaved very close is the
thermal design. It will not run for long with the side cover removed.
That wouldn't surprise me if the side cover included a duct, but there
are only holes strategically placed in the side panel. The CPU has an
enormous heat sink, but no fan of its own!

Robert.

 

[Robert Redelmeier]

I had previously noted the puniness of power supplies on OEM
boxes, but not by overloading them to the point where the machine
wouldn't run. I don't think that upping the RAM by 50% and adding
a HDD causing a power supply overload is a reasonable expectation.

Agreed. HDDs (especially 10k) can cause a heavy motor-start
load on the 12V but should otherwise be fine. One problem with
big OEMs is they are also aiming at EPA EnergyStar targets
which cause them to tight-size PSUs for max efficiency.
The German TuV may also have powerfactor targets.

Something else about this machine that is shaved very close is
the thermal design. It will not run for long with the side cover
removed. That wouldn't surprise me if the side cover included a
duct, but there are only holes strategically placed in the side
panel. The CPU has an enormous heat sink, but no fan of its own!

There is your clue, no CPU fan -- the cover is a duct. With the
cover off, air can bypass the CPU heatsink and slip straight to
[from] the PSU fan. I've built machines without CPU fans, and
you have to be very careful about airflow.


-- Robert R

 

[Robert Myers]

Agreed. HDDs (especially 10k) can cause a heavy motor-start
load on the 12V but should otherwise be fine. One problem with
big OEMs is they are also aiming at EPA EnergyStar targets
which cause them to tight-size PSUs for max efficiency.
The German TuV may also have powerfactor targets.

Energy efficiency is the new wild card. It's the only consideration I
can imagine that would justify cutting it so close.

Something else about this machine that is shaved very close is
the thermal design. It will not run for long with the side cover
removed. That wouldn't surprise me if the side cover included a
duct, but there are only holes strategically placed in the side
panel. The CPU has an enormous heat sink, but no fan of its own!

There is your clue, no CPU fan -- the cover is a duct. With the
cover off, air can bypass the CPU heatsink and slip straight to
[from] the PSU fan. I've built machines without CPU fans, and
you have to be very careful about airflow.

It's misleading for me to say that the CPU has no fan. The case fan
draws air in such a way that the predominant flow is through the many
horizontally stacked, spaced plates of the heat sink. It wouldn't seem
that it would matter so much where the air comes from, but apparently it
does. The heat sink stack is so tall it extends nearly to the edge of
the case, so that air entering through the side holes (highly turbulent
because it is a collection of small jets) encounters the top of the heat
sink almost immediately. I suspect that those holes behave more like
vorticity generators than like a duct. Take the cover off, and the
relatively laminar flow through the heat sink doesn't create enough heat
transfer.

Robert.

 

[Bill Davidsen]

I would have expected it to boot, too, but it didn't. I tried to boot
before opening the machine. It's been long enough that I don't remember
if it got past a power-on POST and stopped at a blank screen. Maybe
that would have constituted passing a smoke test. I can understand why
a tester in a hurry wouldn't want to wait for Vista to boot.

Since I intended to add memory, I didn't hesitate to take the cover off
to see what was going on. Making what I thought would be a correct
memory install fixed the problem.

Glad you like it, I have been thinking of a 930 for a KVM server, drop in 12GB
of RAM and 4TB of cheap disk and put all the boring little 512m servers on Earth
on it.

 

[Bill Davidsen]

Agreed. HDDs (especially 10k) can cause a heavy motor-start
load on the 12V but should otherwise be fine. One problem with
big OEMs is they are also aiming at EPA EnergyStar targets
which cause them to tight-size PSUs for max efficiency.
The German TuV may also have powerfactor targets.

Energy efficiency is the new wild card. It's the only consideration I
can imagine that would justify cutting it so close.

Something else about this machine that is shaved very close is
the thermal design. It will not run for long with the side cover
removed. That wouldn't surprise me if the side cover included a
duct, but there are only holes strategically placed in the side
panel. The CPU has an enormous heat sink, but no fan of its own!

There is your clue, no CPU fan -- the cover is a duct. With the
cover off, air can bypass the CPU heatsink and slip straight to
[from] the PSU fan. I've built machines without CPU fans, and
you have to be very careful about airflow.

It's misleading for me to say that the CPU has no fan. The case fan
draws air in such a way that the predominant flow is through the many
horizontally stacked, spaced plates of the heat sink. It wouldn't seem
that it would matter so much where the air comes from, but apparently it
does. The heat sink stack is so tall it extends nearly to the edge of
the case, so that air entering through the side holes (highly turbulent
because it is a collection of small jets) encounters the top of the heat
sink almost immediately. I suspect that those holes behave more like
vorticity generators than like a duct. Take the cover off, and the
relatively laminar flow through the heat sink doesn't create enough heat
transfer.

Possible, but I think having the coolest outside air coming to the CPU first is
probably the key.

My problem has been running high ambient temperatures. With a 90F building temp
keeping CPU and disk cool is an issue. I looked for a Peltier cooler, but didn't
come up with one I really liked. And they draw a ton of power.

 

[Robert Myers]

Robert Myers wrote:
Possible, but I think having the coolest outside air coming to the CPU
first is probably the key.

The short circuit to the air flow with the cover off is just a few
inches between the heat sink stack and the exhausting case fan, which
exhausts much greater heat than the power supply. A piece of cardboard
or plastic that blocked that short circuit would be an interesting test.
There are actually holes upstream of the CPU to cool the disk drive,
and that air has to get through/around the CPU heat sink to exit the case.

My problem has been running high ambient temperatures. With a 90F
building temp keeping CPU and disk cool is an issue. I looked for a
Peltier cooler, but didn't come up with one I really liked. And they
draw a ton of power.

These boxes have run without air conditioning in summer weather. I
don't think 90F ambient should be a problem.

Robert.

 

[Bill Davidsen]

The short circuit to the air flow with the cover off is just a few
inches between the heat sink stack and the exhausting case fan, which
exhausts much greater heat than the power supply. A piece of cardboard
or plastic that blocked that short circuit would be an interesting test.
There are actually holes upstream of the CPU to cool the disk drive,
and that air has to get through/around the CPU heat sink to exit the case.


These boxes have run without air conditioning in summer weather. I
don't think 90F ambient should be a problem.

Now that I look, this system seems to have hit 52C air temp last August. That's
air, not components. That sound right for four servers in a room, 102F temp
outside, pulling in "cooling" air with a window fan over a hot black roof in
full sunlight. As the song says, "Somethin' Got to be Arranged."

Google says systems can run hotter, several have been up for 400+ days, so maybe
they're right. Hope so. :-(

 

[Robert Redelmeier]

Energy efficiency is the new wild card. It's the only consideration
I can imagine that would justify cutting it so close.

Older problems, but perhaps related/continuing:

http://www.nytimes.com/2010/06/29/technology/29dell.html

-- Robert R

 

[Robert Myers]

Older problems, but perhaps related/continuing:

http://www.nytimes.com/2010/06/29/technology/29dell.html

A long time before Dell's customer service problems and practices began
to get public attention, I had a long go-round with them that told me
everything I needed to know about the corporate culture there. I even
wrote about it in one of these forums, and Felger Carbon defended Dell
as not being the bottom-feeder I characterized it as being. I suspect
the (still unidentified) company that built the box causing the current
problem was the company that he would have claimed was the bottom feeder.

When I finally wrestled Dell to the mat, it turned out that there were
six hundred people ahead of me for the replacement part needed (so I had
to wait another six months for it), and the customer service rep had to
consult a manager before finally agreeing with me that there was
something wrong with the hardware, which manifested itself as a clear
data-corruption problem. The story reported in the New York Times
sounds very similar.

While it was still in the PC business, no IBM alum would ever comment on
the competitive landscape it faced in that market. As I infer the
corporate culture at IBM as it once was, they probably believed that
their corporate customers would wise up and stop buying the kind of junk
that was being sold at rock-bottom prices. History, of course, proved
otherwise.

As Yousuf pointed out earlier, it's quite a challenge to build a box at
a price that's competitive with what you can get from an OEM, and, even
then, although you know exactly who provided each part (at least in
theory), you can still wind up with a motherboard that becomes notorious
for having been built with bad capacitors.

If I had to finger a culprit here, I'd point at the business schools,
which seem to be so detached from reality that they actually think that
anything that looks good on a spreadsheet is a good business practice.
That anyone ever would have admired Dell just boggles my mind, just as
it boggles my mind that people *still* don't get why we are so much
poorer now than we were a few years ago. That is to say that, although
the PC business is cut-throat in a way that ultimately puts customers at
risk, it is not a problem that is peculiar to PC OEM's.

Robert.

 

[Yousuf Khan]

Older problems, but perhaps related/continuing:

http://www.nytimes.com/2010/06/29/technology/29dell.html

Quite the come down for a company which was a model taught at the
Harvard business school. But in reality, even back when it was a
business darling, people knew they were sitting on a slippery slope.
It's real business model lay in the taking of subsidies from bigger
companies that acted as its sugar daddy.

Yousuf Khan

 

[Robert Myers]

Glad you like it, I have been thinking of a 930 for a KVM server, drop in12GB
of RAM and 4TB of cheap disk and put all the boring little 512m servers on Earth
on it.

Everything now goes through this 64-bit Windows desktop, including a
virtual 64-bit Fedora 13 and a virtual 32-bit Windows XP Professional,
with a Cygwin X-server handling graphical output from other Linux
boxes. The virtualized machines, both Windows and Linux running
simultaneously, are at least as snappy as Windows and Linux running on
E8200 and E8400 Core 2 Duo. I wish someone made affordable 4Gb DDR3
non-ECC, since memory is the only thing that is ever remotely in short
supply. The virtualized XP Professional will allow me to decommission
a separate box running XP just to support a handful of legacy XP
programs.

Robert.

 

[Bill Davidsen]

Everything now goes through this 64-bit Windows desktop, including a
virtual 64-bit Fedora 13 and a virtual 32-bit Windows XP Professional,
with a Cygwin X-server handling graphical output from other Linux
boxes. The virtualized machines, both Windows and Linux running
simultaneously, are at least as snappy as Windows and Linux running on
E8200 and E8400 Core 2 Duo. I wish someone made affordable 4Gb DDR3
non-ECC, since memory is the only thing that is ever remotely in short
supply. The virtualized XP Professional will allow me to decommission
a separate box running XP just to support a handful of legacy XP
programs.

That's one of the reasons I'm looking at an i7-930 and Asus m/b for a hosting
system, I can get to 12GB with cheap memory. On the other hand, the i7-875
unlocked is cheap and allows o/c by use of multiplier. But no cheap memory
there, need 4GB parts. I'm tempted to build a host machine with Xeons and ECC
memory, slightly more reliable and all, but I think slower.

Lots of ways to go, each with a drawback. :-(

 

[Bill Davidsen]

Nonsense. Any PC must be expected to have memory and/or PCI cards
added.

Could you note the source of that opinion? Some where a major vendor said that?
For both cost and power efficiency reasons vendors seem to sell machines where
anything more than a memory upgrade puts it out of power. Add a disk, marginal,
hope you can power on spun down to avoid surge. And that 200w super gaming video
card? Not on the machines intended to let a casual user get on the net, or a
clerical worker do data entry or update a few things using the system as a terminal.

Seriously, I see stuff with 300w, even 250w power supplies, and as shipped they
have 50w of headroom if the voltage stays up. Honest, lots of vendorsmaking
them, and they work fine when used as intended.