Re: tlug: dual-pentium processors

[Karl-Max Wagner <karlmax@example.com>]

> Right, that is what I was alluding to.  However ... how many motherboard

Not many for sure - if any.....

> designers are competent?  I've seen a fair amount of crosstalk on
> name-brand motherboards (ASUS) ... placing a probe on one of the address

I didn't do that so far but from the layout I'd expect exactly
that....

> lines, then strobing an adjacent line, and watching neat induced waveforms
> (yes, the probe and scope were properly grounded).

I've no doubt. Well, now you know that my rantings aren't so
idle....

> > Bullshit ! 33 MHz is almost DC. Even if you consider the harmonics (
> > don't consider anything beyond the 3rd... ) that's just middle VHF.
> > Where is the problem ?
> 
> Don't know enough here to comment, that's why I asked.

Wasn't directed against you, but against those "pundits"....

> > There ain't anything like some years of solid rf design
> > experience.....
> 
> Yep.  Which I wish that I had (things analog/RF have always been slightly
> to the right of black magic to me).

That's EXACTLY what they are - and unfortunately I've NEVER seen
a university that teaches students real hands on knowledge in rf
engineering. Well, maybe there I'm asking too much - the issue
is highly complex and it takes many years to gain mastery in it.
It reminds me heavily about things like tea ceremonies and the
like which take many years until you become a master in them.

Unfortunately, as computers become faster and faster rf
engineering problems crept silently in and started wreaking
havoc. At the end seventies things like line reflections etc
were no issue in digital technology - things were really slow
back then. However, now with processor clocks soon crossing the
gigahertz frontier we are rapidely entering the UHF frequency
range and that means plain hard rf engineering. Some farlooking
people already anticipated this - however, the crowd is now
slowly getting the facts - after they're established. The
results are that even standards are defined that are downright
hair raising from an rf point of view ( the only popular bus
well defined in that respect is VME ).

Another problem is delay times. Unfortunately the speed of light
is a pretty solid barrier and this ultimately has to be taken
into account when e.g. considering memory accesses over a bus
with given physical dimensions. This will means that "streaming"
data transfers and the like will become more and more
commonplace. It also means that in order to increase computing
power paralell processing is the way to go, even on a single
chip ( consider chips with multiple processors onboard as a fact
soon to be commonplace ). By 2020 your computer will house a 64
paralell processor chip and a total anmount of RAM in the 100
GByte range and your harddisk will have capacities in the
terabyte range. All interfaces will be fiber optical. This
machine will run Linux 10.0.32 or so, a massively paralell
kernel with extensive multithreading built in. Beowulf will be
history by then - it has been integrated into the kernel for
more than a decade by then. It will be a machine more powerful
than the best supercomputers today. Processor clocks will be in
the tens of gigahertz range and the motherboard material will be
some variant of RTDuroid.....

Prime time for Rogers Materials Division......
 
                                  Karl-Max Wagner
                                  karlmax@example.com
--------------------------------------------------------------
Next Nomikai: 18 September, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next Meeting: 10 October, Tokyo Station Yaesu central gate 12:30
--------------------------------------------------------------
Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp