[sdiy] IN your mind, what is ....

[Magnus Danielson]

From: Rainer Buchty <buchty at cs.tum.edu>
Subject: Re: [sdiy] IN your mind, what is ....
Date: Sun, 1 Feb 2004 15:55:09 +0100 (CET)
Message-ID: <Pine.LNX.4.56.0402011546080.15581 at atbode100.informatik.tu-muenchen.de>

> > This is indeed interesting news... there are a number of bad
> > design-choices behind IA-64 which makes me doubt it in the long run.
> > Let's say it hasn't been a big seller.
> 
> They had forecasts that Itanium servers would gain $25B revenues by 2004.
> They now corrected their forecasts to $8.7B by 2007.
> 
> Or the other way round: In the second quarter after its official
> introduction, already >10.000 Opteron servers were sold but only <5000
> Itanium servers.

I've heard similar numbers, but I didn't have the refs at hand. It's time to
swing the axe. Maybe Intel should resurrect the Alpha! ;O)

> > > > Admittedly, the 6502 was the first girl I ever kissed -- umm, I mean the
> > > > first CPU that I programmed in machine language! -- so I've got a soft spot
> > > > for it. Though it was no sophisticate, the '02 was hardworking and friendly,
> > > > and we had many good times together. :-)
> >
> > You need to get our more often!
> 
> Uhm, that's not me who you were citing.

No, I know...

> But OTOH, I fully agree that the 6502 is just sweet (although the first girl
> I ever kissed was a Sharp MZ80K :)

But NOW you have joined the group in need of that comment! ;O)

> > > The thing which I still like most about the 6502 is that you can build
> > > 2-CPU systems from it so easily. Just clock them off 180° phase different
> > > clocks and they will happily share the same system resources. The same
> > > works for Motorola's 6800/9; that kind of multiprocessing was IIRC used in
> > > the Fairlight CMI series I and II (to bring this somewhat on-topic again :)
> >
> > Interesting. I haven't even thought about that! ;O)
> > You're excussed for bringing it a little more ontopic! ;O)
> 
> :)
> 
> It's something I really like to have with e.g. Atmel AVRs to set up a nice
> "processor/controller" architecture with one controller spreading around
> parameters via shared memory to a set of sub-processors which do nothing
> but computing -- uninterrupted by the data distribution.
> 
> But the 6502 or 6800/9 approach with the inverted clocks and a simple bus
> gate (aka 74244/245) can't be adopted. There needs to be a more
> sophisticated bus controller which then leads to BUSRQ/BUSACK-kind of
> protocols interrupting the sub-processors.
> 
> I don't want to have the voice processors interrupted by the MIDI/GUI
> processor...

The best thing you can do is to *really* separate the processing. It's not even
given that a *Generic* CPU does any particular job best, it just do a set of
jobs fairly well. For real-time continous sample processing, you can do alot
better, and that's why early digital processing like those of Eventide H949 and
the Lexicon PCM-70 and 480L didn't use generic processors in their sound
processing but used more dedicated engines instead. It's rock-solid,
(relatively) cheap to implement and can when tuned well make a heck of a lot of
processing per sample.

The generic DSPs is really the bastard of generic CPUs and needs of Digital
Signal Processing, and ugly kids they are... now on both steriods, Hemohes and
what have you... probably a bit of speed too.

Cheers,
Magnus - needs to get out more too when I come to think about it...