Re: New Model: TP755CX

["Robert George" <george@holmes.cs.nps.navy.mil>]

On May 10,  8:50am, Randal Whittle wrote:
> Subject: Re: New Model: TP755CX
> > Can anyone in the know comment on the relative speeds of a P100 vs a
> > good PowerPC chip?
>
> 	I don't know that I'm "in the know", but from what I've read,
> the Pentium 90's and 100's, in raw terms, aren't as fast as the PPC 601
> chip.

Randy,

     I'm currently working with the MIT Computational Structures Group on the
StarT-NG supercomputer which is a parallel processor comprised of PPC 620's.
 I'm also on the Power Consortium's non-disclosure agreement.

     The 601 is has marginally faster (~10%) integer performance than a P5
running at the same clock rate.  Floating point performance is significantly
(~35%) faster.

> Before the end of this year
> however, you will see Macintosh machines with the next generation PowerPC
> chip--the 604--in them.

     We have 120Mhz 604's right now, and they are not samples.

> From what I can see, the 604 will blow the doors off the Pentium

     The 604 integer unit is a 64-bit version of the 601's integer unit, so you
won't see much performance improvement there.  They've expanded the on-chip
cache, and improved the floating point unit.  The big difference is the vastly
increased clock rate.  By the end of the year AIM is expecting to ship 166Mhz
versions.

     If it makes any difference, the PowerPC is a much more elegant
architecture IMHO than the P5.  In all fairness, the PowerPC design was not
crippled by an installed user base.  The P6 chip, btw, is a beautiful design.
 A true RISC processor with 4 execution units.  They provide backward
compatability by executing x86 instructions as macros (remarkably similar to
how the Alpha can execute VAX instructions as macros).

> and quite likely by the time Intel catches up
> with their next chip, Motorola will have the PPC 620 out, keeping one
> step ahead of them.

     Unfortunately the 620 is probably not going to reach the mass market. The
620 has four execution units, compared to three on the 601 and 604, a better
floating point unit and an on-chip L2 cache interface.  AIM feels that the only
market for this chip will be the scientific/research community.  They believe
that they will make more money selling enhanced versions of the 604, which is
much less expensive to make.

     The next big PowerPC chip will be the 630, which will probably be a VLIW
(very long instruction word) processor, with 18 execution units or so.

> 	In terms of market share, I doubt anything huge will happen, but
> it appears to be a safe bet that the PowerPC chip will make a decent dent,
> given its huge performance advantages and its relatively low price(s).

     Agreed.  In the PC market anything can and will happen.

Robert

-- 
___________________________________________________________________________
Robert George      Computer Engineer     (408) 656-3316 Fax: (408) 656-2814
Army Research Laboratory/AMSRL-SS-IC/2800 Powder Mill Rd/Adelphi, MD 20783
            URL: http://cs.nps.navy.mil/people/phd/george
"I am Homer of Borg, prepare to be assim... Ooooooooh donuts!"