Re: Processor types


Subject: Re: Processor types
From: Christopher E. Brown (cbrown@woods.net)
Date: Sat Jun 22 2002 - 15:53:32 AKDT


On Fri, 21 Jun 2002, civileme wrote:

> Not much remains to be done in terms of making processors smaller and faster
> with silicon. Memory is at the point where an alpha particle is energetic in
> comparison to stored memory charge--in fact this was a problem with come
> memory chips where a deramic case had some thalium in it. An alpha particle
> from radioactive decay was enough energy to flip a bit, even though alpha
> particles can almost always be stopped by a sheet of paper.

        Umm, this was a known issue since shortly after the
introduction of what has become conventional RAM. They started
watching their materials batches back in the early 80s to prevent this
kind of thing.. The paper comparison really does not come in, the
radioactive decay and resulting alpha particle were *within* the cell
arrays of the effected RAM chip. Yes, memory uses much less charge
these days, but is made with better tech. YES, a 512MB chunk of GRADE
A ram in a low background area with see at least 16 bit flips per
month due to alpha particles, plus whatever gamma exposure and power
fluctuations cause, this is what ECC is for. The general market gets
away with it, after all in (512 * 1024 * 1024 * 8) bits of data, if
you flip 16 of them in a month, what are the chances of it hitting a
critical bit and crashing the machine. Not much if your software
won't run for more than a week, and you power the machine down every
night anyway.

BTW: The standard def for Grade A memory used to be, *ONE* bit per
month per 32MB. This spec included alpha particles due to radioactive
decay within the substrate, flaws, etc. It did *NOT* include gamma
radiation errors, and assumed perfect temp/perfect power stability.

Besides, Seagate did worse with a *large* batch of drives. 20 and 40
MB RLL units (remember the old 5.25 HH units, golden casing?). Those
drives used cobalt in the platter coatings... Well, it seems that
Seagate got a large (several tons) batch of cobalt alloy from some
supplier in Mexico. It seems that the cobalt had originally come from
the US, and got misrouted somewhere during its disposal, from a
research facility where it had been contaminated. Not majorly bad,
but sitting near several tons of it would be a very unhealthy thing to
do. Kept getting failures on this batch of several thousand drives...
Finally someone in a research lab that works with mild radioactive
materials was doing a contam sweep and noticed their PC was HOT...
Seagate soon found out about this. Fairly minor (the radiation level
from the drive were not unhealthy per say, unless you removed the
platter and slept with it for a year) thing, but lost alot of data.

> So we are reaching some theoretical limits for the materials in use. There
> are other schemes, which could pack one of today's biggest into a single
> cubic inch, but they require expensive materials and special conditions to
> work, (like liquid helium temperatures).

We have been reaching limits every new generation of gear, and they
keep sneeking around them. Yes, we will hit something we cannot get
around, but don't count on it being now. When it happens alot of
folks will say "I was right", 99% of em will only be right because
they have been saying the same thing for the last 5 generations of
hardware. They are running into problems with vertical expansion
(large clock rate increases), and are working with horizontal
expansion (SMP within the CHIP). Also, CPUs are again easily
outrunning their system busses, so PCI-64 (64bit PCI at 66Mhz has been
around just as long as the 32bit 33Mhz type, but PC companies didn't
bother with it) and PCI-X (pci-64/66Mhz doubleclocked) are coming into
play.

> The next move up is likely to be to 64 bit processors. Intel has had its
> IA-64 out for a while, but none of the software that runs on a pentium will
> run on an IA-64. the instructions are different.

We are already running on 64bit CPUs, it is only the x86 world that is
over a decade behind the times. *FULL* 64bit CPUs with a 64 to
256bit memory path were on the market in the early 90s. SGI R10K, DEC
Alpha 21064/21066/21164/21264, etc are all true 64bit.

-- 
I route, therefore you are.

--------- To unsubscribe, send email to <aklug-request@aklug.org> with 'unsubscribe' in the message body.



This archive was generated by hypermail 2a23 : Sat Jun 22 2002 - 16:01:49 AKDT