---In message <20020305081834.F26327@dr-evil.shagadelic.org>

On Tue, Mar 05, 2002 at 05:00:28AM -0800, Andy R wrote:
>
> > The current RS/6000s are PCI bus, and haven't been MCA
> > for quite a while now.
> > 
> > They have a Motorola PowerPC processor, nothing like
> > the old M68k.


Jason R Thorpe wrote:

>You're both wrong :-)
>
>RS/6000s have:
>
>	POWER
>  -- or --
>	POWER2
>  -- or --
>	PowerPC
>
>processors, depending on model.


One can find older CPU performance data at SPEC ( like
http://www.spec.org/osg/cpu95/results/ ).  Of course, benchmarks are often
not a good indication of real world performance, especially with the
manufacturers performing the benchmarks.

 If you look at the CPU95 benchmark results, and examine the IBM RS/6000
machines in detail, you can see which machines had which CPU.

 The RISC "POWER" cpu line did precede the use of the PowerPC in RS/6000s.
The decision to switch to PowerPC may have been political, or maybe cost
-- perhaps IBM didn't want to spend money on two high-end RISC chip
development lines.

 I worked more than i would have liked with RS/6000s from about 1992-1998.
(I didn't mind the hardware so much but AIX will seem terribly wrong to
any UNIX sysadmin.  Perhaps you have heard, "AIX" stands for "AIn't uniX.")

My recollection is that in 1996-1998, IBM was having trouble getting the
PowerPC 60x to work well in Multiprocessor configs. Machines that would
support more than 4 CPUs were a long time in coming, and i don't recall
every seeing a PowerPC-based machine with more than 8-CPUs.

 The confirmation of this seemed to be when IBM switched "back" to the
POWER3 line around 1998-9. 

 If i remember correctly, the POWER line (or some evolution of it) was in
use by the AS/400 machines throughout this period.  (Which still enjoy a
niche market.)

 Floating point in the various incarnations of the POWER CPUs was pretty
good, generally better than FP in the PowerPC.  Being RISC, PowerPC
greatly outperforms a comparably-priced CISC chip like the Intel x86 or
Motorola 680x0.  But in my experience the UltraSPARC and US-II were quite
a bit better in FP than a comparable PowerPC. Even better in FP than Sun's
US-II are comparable the MIPS cpu (SGIs) and as someone else has
mentioned, the Compaq (nee DEC) Alpha.

 If i had a problem than needed a lot of UNIX floating point cycles, i'd
try to find a used Alpha. If your problem is parallelizable, you might get
more bang for your buck with a bargain Sun with a couple or four
UltraSPARC-IIs (but you'd almost certainly have to run Solaris on it, at
least for the time being.  It's the only OS with solid SMP support). For
the Alpha, whether the OS should be Tru64, NetBSD or FreeBSD would
probably best be answered by your specific needs.

 One of the difficulties with the RS/6000 is its lack of a non-proprietary
PROM.  There's some brains in there but, it doesn't activate the console
for long after you would see on a Sun.  Until the AIX kernel loads, your
pretty much stuck with a 3-digit red LED as your window into what the hell
the machine is doing. (Which means you have to look up the codes if the
boot hangs, etc.) And it makes it a major pain in the ass to boot off of a
specific disk or type of media.


 --johan