Date: Mon, 26 Jul 1999 19:47:33 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: Poul-Henning Kamp <phk@critter.freebsd.dk> Cc: "Eric J. Schwertfeger" <ejs@bfd.com>, Cosmic 665 <the_hermit665@hotmail.com>, freebsd-smp@FreeBSD.ORG Subject: Re: Overclocking Message-ID: <199907270247.TAA49515@apollo.backplane.com> References: <55719.933019101@critter.freebsd.dk>
index | next in thread | previous in thread | raw e-mail
This is how fabs generally work in regards to chip speeds:
* The fab makes a chip
* The chip is tested. Due to doping inconsistancies and other issues
the chips that come off the fab will only test to certain speeds.
* The chips are placed into buckets based on how fast they test to.
But then marketing and distribution come into play. In recent years fabs
have gotten a lot more reliable and this has resulted in higher and better
yields.
So lets say in a 10000 lot Intel winds up with 8000 chips that test to
500MHz and 2000 that test to 400MHz. But lets also say that Intel needs
to ship 5000 500MHz chips and 5000 400MHz chips. What Intel (and all
chip manufacturers) will do is throw some of the higher-testing chips
into the lower-MHz rating in order to cover their orders.
When you purchase a 400Mhz chip all you are guarenteed is that it can run
at 400MHz. It is possible that your chip was tested to 500MHz but you
can't tell for sure. Intel's FAB has gotten good enough that they
physically changed the celeron design in order to be able to laser or
PROM a limit to the frequency multiplier to prevent people from
overclocking the chips after people found out that they could do it
reliably.
Now, this is why you *DON'T* want to overclock:
When a chip is tested the frequency limit is determined by a number of
factors both internal and external. For example, a single line within
the register file or a single bit in the cache might be just a tad too
slow and limit the overall frequency the chip can sustain. Or it could be
an external signal that doesn't quite meet spec above a certain frequency.
The problem is that usually the failure is something relatively minor in
the chip, but which can have major consequences to the execution of
instructions. Worse, the failure condition is not necessarily reliably
reproduceable. You may believe that your overclocked chip is working
correctly, but there is a good chance that it isn't, quite. If you
overclock a chip you can wind up with wierd failures that take days
or even weeks to show up, or failures that show up as arithmatic
miscalculations. For example, the FP unit might begin to produce
slightly incorrect results, or certain L1 cache situations might fail
to the produce the correct data (this was a serious problem with a lot
of 486's that Intel inadequately tested. They would work fine w/ Windows,
but would fail under UNIX because UNIX utilized chip features that would
tickle the cache bugs).
In regards to heat dissipation: Heat dissipation is not usually a
problem when you are overclocking a chip but still leave it under the
maximum clock rating that the manufacturer makes, at least as long as
the package type remains the same as the package type the manufacturer
uses in their high-end products. I think for most Intel cpu's the package
type is the same for nearly all speed grades of a particular chip class.
If you overclock a chip beyond the maximum rating sold by the manufacturer
in the package type in question, you *can* melt it. Overheating also
introduces additional noise on the die and, even worse, may cause the
gates making up the die to slow down. Thus some people see chips "fail"
after being on for a period of time, and then work again after the
computer's been off for a while.
-Matt
Matthew Dillon
<dillon@backplane.com>
:In message <Pine.BSF.4.05.9907261205510.17960-100000@harlie.bfd.com>, "Eric J.
:Schwertfeger" writes:
:>
:>Much of the practicality of overclocking comes from the family of chips.
:>The same fabrication process is used for PPGA Celerons from 333mhz up to
:>the latest 500mhz, so I wouldn't expect you to damage CPU or motherboard
:>overclocking a PPGA 333 by 50%, as long as you didn't tweak the voltage in
:>order to make it run.
:
:It can be said as simple as this: "You Are Wrong". Running the chip
:at higher clock will lead to increased heat generation, which isn't a
:good thing for your silicon.
:
:
:Rule #1:
: Do not Overclock.
:
:Rule #2:
: If you overclock, do not complain that things don't work.
:
:
:--
:Poul-Henning Kamp FreeBSD coreteam member
:phk@FreeBSD.ORG "Real hackers run -current on their laptop."
:FreeBSD -- It will take a long time before progress goes too far!
:
:
:To Unsubscribe: send mail to majordomo@FreeBSD.org
:with "unsubscribe freebsd-smp" in the body of the message
:
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-smp" in the body of the message
help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199907270247.TAA49515>