I'm posting to both NG's as there seems to be an awful lot of smart hardware
people here. I was reading one of my PC-related magazines on an airplane
the other day when I came across an "infomercial-article" from Gigabyte
touting their newest and greatest mobo's using all solid capacitors. I'll
include a link to the Gigabyte site below, but I'm just wondering what the
consensus is of you folks that are "in the know." Reading up on Gigabyte's
'Ultra Durable' article you would think this is the best invention since
sliced bread.

My motive of course for posting this is that I am close to making a decision
on upgrading my current 5-year old Athlon XP system to something dual-core
etc., and simply don't know at this point to go the Intel or AMD route. I'm
pretty agnostic along these lines, just want the best bang for my buck.
Anyway, here's the link:

http://www.gigabyte.com.tw/FileList/..._all_solid.htm

Thanks to you all in advance.

TLG

On Mon, 26 Mar 2007 18:37:12 +0200, "The Lone Gunman" <lone (AT) invalid (DOT) net>
wrote:

This is essentially a high-availability server design practice trickling down
to the desktop. Every five-nines+ server motherboard I've designed from late
1999 on used aluminum capacitors sporting solid polymer material (see Sanyo
OS-CON, for example) wherever high capacitance/low ESR devices were required -
which was mostly on the processor and chipset VRD rails.

These aren't cheap, so I was restrained from using more than theory required
(though I always put a few extra, strategically located footprints down and
left them off the assembly boms). The generally reflects a trade-off between
load-lines and how complex (which pretty much means how many phases) a
switching regulator one wants to construct. 6-phase switchers have been fairly
standard in the Xeon space, but 8-phasers are gaining, trading fets for fat
caps. With fewer caps comes less resistance to putting down premium devices
and burying the life-span issue.

Does it make any difference to a typical desktop user? Plainly, the likelihood
of failure of electrolytic caps as the desktop ages is going to be a major
factor, with those who stick with a system for 5 years more likely to succumb
to sudden system death, and those who keep on the cutting edge (if not the
bleeding edge) most likely to beat the reaper...

Cheers

/daytripper

daytripper <day_trippr (AT) REMOVEyahoo (DOT) com> wrote:
: On Mon, 26 Mar 2007 18:37:12 +0200, "The Lone Gunman"
: <lone (AT) invalid (DOT) net> wrote:
:
<snip for brevity>

: This is essentially a high-availability server design practice
: trickling down to the desktop. Every five-nines+ server
: motherboard I've designed from late 1999 on used aluminum
: capacitors sporting solid polymer material (see Sanyo OS-CON,
: for example) wherever high capacitance/low ESR devices were
: required - which was mostly on the processor and chipset VRD
: rails.
:
: These aren't cheap, so I was restrained from using more than
: theory required (though I always put a few extra,
: strategically located footprints down and left them off the
: assembly boms). The generally reflects a trade-off between
: load-lines and how complex (which pretty much means how many
: phases) a switching regulator one wants to construct. 6-phase
: switchers have been fairly standard in the Xeon space, but
: 8-phasers are gaining, trading fets for fat caps. With fewer
: caps comes less resistance to putting down premium devices and
: burying the life-span issue.
:
: Does it make any difference to a typical desktop user?
: Plainly, the likelihood of failure of electrolytic caps as the
: desktop ages is going to be a major factor, with those who
: stick with a system for 5 years more likely to succumb to
: sudden system death, and those who keep on the cutting edge
: (if not the bleeding edge) most likely to beat the reaper...
:
: Cheers
:
: /daytripper

Thanks for the great post, Daytripper. I **think** I understand most of
what you said. Yeah, my mobo is approaching five years now and I'm starting
to see problems that I'm almost sure are h/w related. Won't go into that
here, just wanted some feedback on the OP. You're the MAN, Daytripper.
Glad to see you're still alive! :-)

/TLG (still wondering why there are so many MIA's from the .chips NG)

daytripper <day_trippr (AT) REMOVEyahoo (DOT) com> writes:
Sure it is more likely that a sudden system death will occur in 5
years than in 3 years, as long as the probability of sudden system
death in the latter two years is non-zero.

What is more interesting is whether the rate of system death increases
with age, and by how much.

We have two consumer-type boards running all the time since seven
years, so I think that any fear that the reaper is coming soon to
visit desktops as soon as they reach five years is exaggerated,
especially since many desktops are turned off much of the time.

Sudden deaths that we see occur more in power supplies, RAM, fans, and
disks.

- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton (AT) mips (DOT) complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html

On Fri, 30 Mar 2007 12:05:55 GMT, anton (AT) mips (DOT) complang.tuwien.ac.at (Anton
Ertl) wrote:

Your methodology is intrinsically unsound. Two measurements using boards of
unknown construction does not support your conclusion.

Liquid electrolyte caps have a very low mtbf.
That's the beginning and end of the question...

Cheers

/daytripper

On 2007-03-30 16:09:46 +0100, daytripper <day_trippr (AT) REMOVEyahoo (DOT) com> said:

Actually, no it's not. The issue is whether, empirically, electrolytic
failure *on motherborads* (not in power supplies) is a significant
cause of motherboard death, and in turn whether motherboard death is a
significant contributor to the death of desktops. My guess is the
answers to these are "yes" and "no" respectively.

The kinds of design adopted for high-reliability machines are
interesting here, but not for the reason you might think: every machine
which has any pretension to high-reliability I've ever seen has
redundant hot-swappable PSUs and fans (as well as disks, obviously).
Assuming the designers were competent that tells you what tends to
fail: power supplies and fans (and disks). Further, since those things
can now be swapped without taking the machine down, the reliability of
other components becomes the thing that controls the reliability of the
whole system. For machines where the system board can't be swapped
with the machine up, which is most of them, that means you might need
to pay serious attention to that (for machines where system boards
*can* be swapped with the machine up you're probably paying so much for
the machine that you expect serious attention to be paid anyway).

It would be interesting to see statistics as to what kills desktops.
My guess (which should be taken for what it's worth, namely nothing
is):

1. most of them are thrown away, working;
2. disk failure
3. PSU failure
4. fan failure
5. everything else, trailing a long way behind.

--tim

On 2007-03-30 23:34:00 +0100, daytripper <day_trippr (AT) REMOVEyahoo (DOT) com> said:

Do these machines (or any other HA systems) have much to do with
desktops (which was the original question, remember)? No.

Indeed you do not. And again: what exactly does this have to do with
the reliability of desktop motherboards? I suggest: nothing.

On Fri, 30 Mar 2007 19:05:47 +0100, Tim Bradshaw <tfb (AT) tfeb (DOT) org> wrote:

No pretensions here: http://www.stratus.com/products/index.htm

An entire "side" - power, cooling, logic - is field replaceable without any
user or process ever realizing it's happening - from fault, through automagic
call-home-to-mama, through FRU replacement, through re synchronization.

Yet, and, as I said earlier, the motherboards use capacitors that have an mtbf
an order of magnitude higher than that found on desktops (until recently).

Because you don't get to 6 nines on the cheap, son.
No matter what is replaceable once the system's been fired up...

Cheers

/daytripper

Tim Bradshaw wrote:

....

That (with the exception of the top point) was indeed the approximate
ordering I read recently in a paper by credible people (may have been
one of the recent ones discussing real-world vs. specced disk MTBFs - or
not). But it's not clear what the service life of motherboards was
considered to be, so it's possible that inclusion of 5+ year old MBs
would have changed the conclusions.

- bill

On Fri, 30 Mar 2007 23:12:14 +0100, Tim Bradshaw <tfb (AT) tfeb (DOT) org> wrote:

Yes, it has to do with the thread, specifically the migration of components
used in highly-available systems down to the desktop.

Also, it refutes your concept that something easily replaceable in the field
without disruption could afford to use cheaper, lower-mtbf components.

Ok, so you got confused along the way. Mostly your own doing, as the point has
always been: use cheap, low-mtbf caps and you run a higher risk of a system
failure, compared to the use of high-mtbf components.

For what ever misguided reason, you got caught up in the phrase "sudden system
death", and tried to make some kind of point countering it. I can't imagine
why, a failure that takes down the system is usually sudden, and the passage
of time is no friend to a low-mtbf component used in fairly high quantity...

Cheers

/daytripper