Re: tlug: dual-pentium processors

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

> I see no point to paying extra cash for anything faster than 233 MHz, as
> the extra CPU speed[1] is ultimately limited by the bus speed[2].
Not necessarily. You forget the L1 cache. If you have highly
localized code most of the time ( and this is not so
unreasonable to assume ) then you do a lot of computation in
cache and you suck in new pages in a kind of burst mode. Using
programming tricks ( the "intentional cache miss" trick ) you
can even see to it that the required page is already in cache as
far as you need it. 

> [1] Does anyone else remember the dire warnings by the EE pundits around
>     86/87 about the dangers of emitting EMF in the high MHz range --

Pundits ????? Buhahahahaha....! In RF enginerering it is common
practice to run signals in the tens of GHz range over PCB
traces. You have to follow proper design rules ( characteristic
impedance, avoidance of crosstalk etc . ) though - but this has
been commonplace engineering in the radio field for decades.

>     which is why we'll never see a bus speed higher than 33 MHz? Why

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 ?

>     isn't that a concern anymore?

Probably because the digital guys got some teaching from the RF
gang, I guess....

> 
> [2] Yes, I know about the 83/100 MHz bus.  How much of a performance win
>     is that?  Enough to shave a couple of milliseconds off of my kernel
>     compiles?

It simply triples your I/O capabilities, thus reducing latencies
by the same factor. Depends on your I/O intensity. See above.

Actually, I'm always shocked by the knowledge ( or better lack
thereof ) in rf engineering issues common to digital hardware
designers. Just look at PCI and their great ( truly hair
raising ) scheme of dealing with line reflections. The inventors
of that ought to be nuked.

I just remember a friend designing a frame grabber using totally
inadequate wiring techniques ( actually, looking at it made
insulted my rf instincts enough to make me sick ). The thing
behaved erratically. After being unable to solve the problems he
followed my advice and did a new breadboard following sound rf
engineering practices - oh wonder, it worked without a hitch...!

There ain't anything like some years of solid rf design
experience.....

                            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