Date: Thu, 10 Dec 1998 11:06:15 -0800
Subject: overclocking article from Intel emloyee
Let me preface this message with, I too am an Intel employee and this
message IS NOT representative of Intel or their polocies. This statement
is merely some of my opinions.
I would like to address your comment about "We don't care what you do with
the CPU in the privacy of your own home" and isn't there a way to test for
previous instances of overclocking.
It is true that manufacturers can not really dictate what people do with
products in their own homes but then again at the same time you have to
look at what the legal responsibilites of that comany are in terms of
service and support for that product. The majority of customers are good
honest people but there isn't much preventing someone for pulling out their
CPU playing, good round of badmitten with it then putting it back and then
being pissed off because their computer doesn't work anymore. As was
stated in the last email, who eats the cost of these replacements?
Typically not the people that sold you the computer in the first place but
the component manufacturer, ie. INTEL.
As for the second point ... I'm sure that you are aware that one of the
primary dangers of overclocking is electomigration. To steal blatently
from some research in stead of retyping:
"Electromigration (EM) remains a serious problem for the reliability of
VLSI integrated circuits, and will become an even more serious issue as
future IC's employ linewidths below 0.2 micron, since current densities
will not scale as quickly as interconnect cross-sectional area shrinks.
Future scaling of circuit geometries will increase the fraction of lines
susceptible to EM. It has been observed experimentally that as linewidths
shrink, the median-time-to-failure (MTF) actually increases for "bamboo"
lines whose linewidths are equal to or smaller than the grain size of the
constituent polycrystalline metal film, but early failures become a severe
concern. "
And check out the following pictures:
[link lost in import: /~micro/fig20.html]
[link lost in import: /~micro/fig21.html]
If you don't know already, the polycrystalline line widths directly effect
the speed of the processor, the smaller the line width the faster it goes,
but at the same time the more prone it is to EM failure. So the smaller
and smaller device linewidths go the more prevalent EM issues are going to
be. This may be one of the reasons that Intel is deciding to lock down the
clock speed on processors instead of letting people experiment.
In the past line widths have been large enough that EM MTF was far enough
out there that even overclockers weren't in danger of crashing their CPUs
for quite a while. But now the game has changed and Intel being the
customer concious company that it is has decided that instead of taking a
chance they'd rather guarentee that a CPU will work at it's rated speed.
Intel's built in speed buffer (the chip can actually go faster but rating
it at a slightly slower speed guarentees operation) is also a big part of
the stability and reliability of intel CPUs. Not to sound like I'm company
bashing or anything but have you ever noticed that you can't overclock AMD
chips very much (if at all).
Point? Oh yeah, what was my point? Point is that EM failure is VERY VERY
difficult to detect until it actually happens. Given certain
polycrystalline line widths, you have certain probabilities of EM failure
(increasing as line width shrinks). But that's all they are is
probabilities. So it is nearly impossible to "test" the chip to find out
if it has an increasing amount of potential for EM failure (indicative of
overclocking or not, overclocking just speeds up the process EM failure, it
does not have a tell tale signature), especially since every chip is unique
and the characteristics of the chip initially are unknown.
Besides, the next batch of processors due out should be fast enough for
just about anybody to do anything they want with them anyways! <grin>
Remember when you thought 500 MHz was fast?!?!?! <instert evil laugh here>
I hope my babbling helped some, if nothing else the pics of
electromigration are great!
Lawrence
Lawrence,
Unless you've miss-quoted, I think the statement actually says that with
smaller linewidths, the MTF actually *increases*. This would mean the time
to failure is longer! Hopefully the early failures can be weeded out by
burn-in, because surely Intel would not want to ship those to customers,
over-clockers or not.
The argument against OC escapes me. If it is such a big warranty issue,
surely Intel could find a way to record the maximum clock speed the chip
ever ran at, rather than trying to look for electro-migration effects which
happen to all CPU's OC or not.
Kevin Caldwell
Calgary, Canada
There is one simple reason why electromigration will not be an issue,
and it is simply the flip side of the coin that Intel are now seeing.
The Celeron is selling bucketloads because it gives the most bang for
the buck - why? Thanks to overclocking. In the future if thew design
of Intel chips is such that they are more prone to EM failure than AMD
in timescales that are of concern to the marketplace, it won't be
Intel chips that are getting sold. Intel are eschewing copper
technology for the moment, following a different path to the one
licensed from Motorola by A.M.D. - that in itself could be critical if
AMD chips stand up better in future.
Intel have close to a monopoly, but dwindling all the time. In the
past they sold on name AND by being the best, but now they are just
the best - AMD's name is as well known as Intel's. If AMD's product
should become equal then that market share which has fallen from 95%
to 75% over the last couple of years will tumble still further.
Ain't a capitalist ethic a ***? :-)
Cheers!
John
Very useful to read your comments on overclocking - but since Intel
thenselves have in the past produced chips that they in effect UNDERCLOCKED
in order to maximise sales over a range of CPU speeds, they are hardly the
people to lecture anyone else on the clocking issue.
Most people accept their own responsibility for the consequences of
overclocking for private use, and though I know it goes on, having bought
about ten different systems over the years - all Intel based - not a single
supplier has ripped me off (at least to my knowledge!). Furthermore I have
never had a single CPU failure despite overclocking (by a small margin) many
of the CPU's I have owned. Thus it seems reasonable to assume that Intel
build a quite significant headroom into their chips.
Though I take the points which you make on the dangers, especially with new
chips, I feel the overclocking issue fundamentally is much less to do with
consumer protection and much more to do with Intel and others protecting
their commercial interests - otherwise why would they have produced, for
instance, several Pentium chips in the past which were actually capable of
much more than they were clocked to, not including headroom, and were in all
respects (except for the label) the same chips as "faster" ones Intel
marketed at (in some cases) nearly double the price.
Finally, you can understand the jaundiced views of many consumers who buy
the "latest" chip for $750 in July, only to find that it is sold for $350 in
August (and still at a huge profit) - that's the kind of blatent cynical
marketing which would be a international scandel in nearly any other
industry. In those circumstances it is no wonder that the average user has
nothing but contempt for Intel and their hypocritical stance on
overclocking.
Robert Young hit it on the head! I remember purchasing 2 dual Pentium Pro
systems and seeing the value of said systems go into the round file a couple
months later with the intro of the PII!!!! Even though a few lame ones might
seek replacement of OCed CPU's I think the majority would simply take thier
lumps not bothering Mother Intel.
I have a P Pro also... Is there any way to speed the thing up?
> Robert Young hit it on the head! I remember purchasing 2 dual Pentium Pro
> systems and seeing the value of said systems go into the round file a couple
> months later with the intro of the PII!!!! Even though a few lame ones might
> seek replacement of OCed CPU's I think the majority would simply take thier
> lumps not bothering Mother Intel.
