[CentOS] CentOS 4U2 timing. -- Cell != PowerPC 64

Bryan J. Smith b.j.smith at ieee.org
Thu Sep 15 14:09:59 UTC 2005


Ed Clarke wrote:  
> The ppc64 arch is likely to become much more important
> once the Sony-Toshiba-IBM processor gets into consumer
> hands.

Understand that "Cell" is not a "generic" CPU.  It is a
configuration-fixed "Power-based core" with vector units
attached.  In a nutshell, it's like having a "moderate
performing" single CPU platform with specialized SIMD (Single
Instruction, Multiple Data) units.  You have to write
software to take advantage of those units, and compiler-based
optimizations are _un_likely to give you much without such
coding.

> The new Sony playstation is supposed to use it

Yes, because a console platform will have libraries written
specifically for the vector units, and titles written to take
advantage of those libraries.

Whether or not those libraries -- in part or in full -- are
released by Sony to the community is anyone's guess at this
point.  If Sony follows its history, it will release a
$20,000+ developer system with those libraries (no, the $200
Linux kit for the PS/2 was not the full development kit ;-).
 
> and I have seen Linux running on at least the simulator.

The PS/3 is supposed to run a release of Linux.  The PS/2
development kit was Linux-based, but the PS/3 is the first
that is a Linux target.  I'm sure GNU/Linux was chosen
because it allows them to build a quick target for the new
Instruction Set Architecture (ISA), while leveraging much of
the existing Power/PowerPC GCC target and OS.

There were rumors that the PS/3 might run MacOS X as a
set-top unit.  I could be to to leverage the existing
"desktop" MacOS X apps, using largely just the core (no
vector processing) as a "moderately performing" Mac system. 
But I don't know if the Cell's Power core is PowerPC (Power
is _not_ fully ISA compatible with PowerPC), no one has given
me a straight answer on that.  So I seriously doubt it.

Unless there is some "virtualization" mode to their Linux
codebase that lets them run a MacOS X instance to them run
those apps.  That might be more feasible.

> This is an extremely powerful chip - similar to an eight
> way SMP ppc.

Sorry, don't mean to cross you, but that's an
oversimplification that simply isn't remotely true.  It is a
single Power core with vector units -- nothing like a
multi-core solution.

Only Microsoft's XBox 360 is a 3-way multi-core Power CPU.

The PS/3 is a Power core with vector units attached,
radically different.

> http://www.blachford.info/computer/Cell/Cell3_v2.html

That's a good link, but even he doesn't know what the final
ISA will look like.  

I also don't think he realizes that the inprecise nature of
the vector units will limit the potential of the Cell for
many scientific/engineering applications.  19GFLOPS is great,
unless you actually need precision.

E.g., you could _never_ use the Cell for Computational Fluid
Dynamics (CFD), which was one of the first and most linearly
scaling applications for Linux clusters.


-- 
Bryan J. Smith                | Sent from Yahoo Mail
mailto:b.j.smith at ieee.org     |  (please excuse any
http://thebs413.blogspot.com/ |   missing headers)



More information about the CentOS mailing list