[rescue] RS/6000??

[Francisco Javier Mesa-Martinez]

On Mon, 15 Oct 2001, Dave McGuire wrote:

> On October 15, ssandau at bath.tmac.com wrote:
> > This 59H says it has all PCI slots free (?). I just wrote to ask about a
> > network connection, too. We'll see. Did some RS/6000 machines have PCi
> > architecture, like new Suns?
> 
>   The current RS/6000 machines are PCI, and are built around
> PowerPC-architecture processors.  That goes back a couple of years.

Some of them were, the high end (64bits) were POWER3 based. Now when they
release the POWER4 on the desktop, I guess they will be converging both
architectures. Now a POWER4 would be nice!!!!!! Aaaaaggggghhh (drool like
Homer).

The specific machine for the original post, a 59h is Microchannel based. I
have one of these at home (but it is sick...). 

There were 2 processor lines and 2 bus architectures:

Original: POWER1/2 + Microchannel
Cheapo Original: PowerPC + Microchannel
Spicy: POWER3 + PCI(64bit)
Crispy: PowerPC + PCI

 
>   I second James' assertion about the floating point performance, by the
> way.  I have a PowerStation 350 here...it's got a 41MHz Power1
> processor.  It cranks about three times the floating point performance
> of a Sun SS2, which is clocked at 40MHz.

The POWER was one of the early superscalar architectures. They get ALOT
done per clock cycle. The POWER and POWER2 architectures were multichip
implementations, which hindered the ramping of their clock cycle. They
were so complex, that the POWER architecture has almost as many (if not
more) instructions than x86. So much for RISC, but then IBM does things
its own way. Also POWER implementation used almost an order of magnitude
more transistors than its contemporary RISC counterparts.

POWER was a braniac architecture (i.e. get lots done per clock cycle, in
fact get a lot of instructions done per cycle). The Alpha was the opposite
(at least at its beginning)... fast clock cycle but don't do as much per
clock cycle (i.e. KPSS, Keep the Pipe Simple Stupid!). I guess Sparc falls
somewhere in between... 

POWER4 could be considered the negative of the Itanium (or whatever that
monstrosity is called). It is funny to see the PWR4 beating the crap out of 
Merced.... after intel claimed that there was no more life for RISC...
 
>   Lots of bang for the clock cycle (and nowadays, for the buck!) in
> those RS/6000 machines.  I've not been messing with them for very
> long, but I like what I see so far.
> 

Yup some of the POWER3 processors run at a "pathetic" 375MHz, yet they
kick the living shit out of a 1.4Ghz P4. And I am talking about real FP
code, not synthetic market-o-benchmarks. I am sick of people comparing
32bit processors performance with 64bit processors (it is all fun and
games until someone asks them intel reps if those MFLOPS are DP!!!)