🚀 go-pugleaf

In article <1992Nov4.084645.16364@brtph560.bnr.ca> kimmyd@bnr.ca (Kimmy Posey P920) writes:
>In article <1992Nov4.004352.15077@fasttech.com>, zeke@fasttech.com (Bohdan Tashchuk) writes:
>>In <1992Nov02.215953.23041@kksys.com> gk@kksys.com (Greg Kemnitz) writes:
>> 
>>>One -- NONE of the manufacturers whose products we handle recommends the use
>>>of such a device on any of their 386, 486SX or 486DX products, excepting the
>>>486DX/50 localbus products.
>
>Imagine that!

Your point?

>>>Five -- If you will look closely at what I wrote, I said that we have found
>>>that MOST 486 cpus do not require this type of device.  I never said ALL,
>>>and in fact did describe one case where we have found them to be desirable.
>> 
>
>I would be interested in seeing your data and a description of how you
>gathered it.

I explained what we based this on in what I posted.  As always, I am open to
reasoned discussion as to why the information may be flawed.

>Sadly many don't have any idea what a datasheet is. 

Unfortunately these days data sheets are not as widely circulated, and they
are increasingly more difficult to correlate with a particular design
without equipment that the average individual or smaller shop cannot afford.
Most computer shops these days do not have the financial resources to staff
their own engineering department.  Even the largest stores usually rely on
the advice of the manufacturers.

>>Intel has provided a chart for the maximum allowed temperature for the 486,
>>with and without a "typical" heatsink. I've studied this chart. Have any of
>>you guys???
>
>Unlikely that it would do any good even if they had ...

Again, one relies on the manufacturers to have done their homework.  If AST,
Compaq, Northgate and others do not have the resources to correctly design
their products, which it appears you are asserting, can you reasonably
expect the average man in the streets to second guess them with greater
accuracy?

>>>Two -- We have NEVER seen a system fail in service due to a processor chip
>>>failure.  This is despite having sold many many hundreds of systems.  Our
>>>customers are in general long term customers, and most of our business is
>>>repeat business, so we have the opportunity to track the history of the
>>>systems we have sold.  We also service a large number of systems sold by
>>>others, many with substandard cooling, and have never seen a failed cpu chip
>>>in them either.
>
>Oh really!  And you test each failed system for a bad CPU?  If so I applaud
>you.  My guess is though that you flip the power on and then sound the alert:
>
>  "Man your Battle Stations!  New motherboard required!  All hands to your
>   cash registers"

Actually we replace very few motherboards, and of the ones we do, we are
generally able to determine the cause of the failure quite precisely.
Before a motherboard is replaced, the cpu is swapped, and we test it at
elevated temperature.  We have yet to need to replace a 486 motherboard that
failed in the field, and have had only one fail in burn in.  Its failure was
related to operation with bus mastering DMA controllers, and was not cpu
related.

Of the 386 motherboards replaced, again, none were due to cpu failure.  If
the cpu was bad it would have been replaced instead of the motherboard.  And
if the motherboard was replaced, we would have moved its cpu chip to the
replacement board and re-run the test cycle.  I took a look at the failure
records, and so far all of the motherboard replacements I have reviewed were
diagnosed to a specific cause.  And virtually all were covered under
warranty, so our clients paid nothing for the replacement.  The other boards
replaced were done due to memory expansion requirements (old designs only
held two or four meg, and the clinet wanted more), bios incompatabilities
where the new bios was either not available or priced so close to the cost
of the new board that going with the new board made economic sense, or
compatability issues due to problems with functionality with option cards.

>BTW I don't 'hundreds' of systems is enough of a sample to call your data stat-
>istically meaningful.

It is certainly meaningful to us.  I agree, we are not doing the same volume
of systems as IBM, but if we had increased the cost of each system by $25 to
cover the cost of the fan (which we have seen no demonstratable need for
based on failure data) and charged 500 customers for it, people would have
spent $12,500 for unnecessary fans.  I should also point out that the
systems we have sold have been selected based on a review process which
includes considering cabinet design and airflow issues.  We have rejected
many designs from consideration for reason of inadequate airflow, and had we
been inclined to sell them, we would have certainly looked to some sort of
device to deal with the need for additional cooling.  We have done this with
the localbus systems, as even with good airflow they run hot and need the
extra cooling capacity.

>>This is an incredibly pompous statement, and displays gross ignorance of the
>>temperature related failure modes of these chips.
>
>I could not agree more!

What I wrote was intended as a basic opinion of one person.  I did not
choose to spell out all the details of what goes into arriving at it, since
I do not feel inclined to write a textbook on system design and
configuration.  I leave that to the experts.  But do not assume that because
I do not provide a sixty page paper on the topic that I have not had
experience with the subtleties of chip failure.

>>First, failures typically are "soft". The part fails at high temperature and
>>when it is cooled starts working again. This sort of flakiness is often seen
>>in computer systems that are returned for service but "no problem found".

Agreed.  And one can set up conditions to test for them.  We have done so
many times when the circumstances have warranted.

>>Second, failures typically are in "worst case" paths. These paths depend on
>>both voltage and temperature. 99+% of the chip may be working, but one or two
>>instructions or cache locations may be flaky. You think your computer works,
>>but it's not QUITE working.

In which case it will exhibit flaky operation which can be isolated and
resolved.

>>Third, much of the reason for the conservative temperature ratings Intel places
>>on their chips has to do with LONG TERM reliability. There are phenomena like
>>electromigration that cause cumulative damage long term, and not immediate
>>failures.
>>But then again, I guess that long term failure doesn't matter much anymore.
>>The PC business is moving so fast that a five-year-old computer that was
>>state-of-the-art then is almost worthless now.

Agreed, long term reliability can be affected by overheating.  Long term is
relative.  For a computer these days, five years is very long term.  For
many other things five years is short term.

>Very well stated. Don't get too upset with Greg though. Like most salesmen he 
>thinks he knows more about computers than the people who design & build them.
>Alas Romper-Room never goes away. 

Cute.  No, I do not know more about them than the folks who correctly design
and build them.  That is why I consult with the suppliers of the equipment,
and rely on their information unless it is shown to be incorrect.  If one of
the manufacturers says their product requires extra cooling, we use extra
cooling.  If they say it does not, we still do some checking during the
initial evaluation to be sure the one we are evaluating is ok, and may
reject the product on the basis that we are not comfortable with the cooling
as delivered. 

In any case, so far the only evidence that I have seen presented by those
who say these devices are required is that the Intel data sheet shows some
particular temperatures that should not be exceeded.  I certainly accept
that, but no evidence has been presented that the systems on the market
today categorically do not have sufficient cooling.  I would very much like
to see the detailed information that supports this contention, and if it
were provided would like to see the response of the various vendors to it
(for example AST, Compaq, Northgate, and others...).

I really am open to considering the alternative viewpoint.  But please back
it with specifics that show that the problem exists, not merely that it
COULD exist.
-- 
Greg Kemnitz / K and K Systems / PO Box 47804 / Plymouth, MN 55447-0804
Internet: gk@kksys.com  / Voice: +1 612 475 1527 / Fax: +1 612 449 0488
Alternate Internet: gk@kksys.mn.org   /   ICBM:  44 59 20 N, 93 28 22 W