[ydl-gen] Benchmarks of code on G5, G5/altivec, x86, PS3/CELL

[Sanders, Rob M.]

Thanks Nathan.  I'll have to take a look there.  I'm sure I'm
not the only one in a catch-22 of not being able to get funding
to investigate a new technology because no one is willing to spend
money on an uninvestigated (hence risky) technology.....<sigh>

				Rob


-----Original Message-----
From: yellowdog-general-bounces at lists.terrasoftsolutions.com
[mailto:yellowdog-general-bounces at lists.terrasoftsolutions.com] On
Behalf Of Nathan Moore
Sent: Thursday, December 21, 2006 10:53 PM
To: Discussion List for General Yellow Dog Linux User Topics
Subject: Re: [ydl-gen] Benchmarks of code on G5, G5/altivec, x86,
PS3/CELL


Hi Rob,

I attacked a similar problem a few summers ago while an intern at  
IBM.  The application I spent the most time porting was GROMACS, a  
computational physics/chemistry/biology code that is used to simulate  
substances on the molecular scale. GROMACS code is heavily used in  
the popular Folding at Home screensaver/drug design application, and has  
been HIGHLY tuned (down to the assembler level) for ppc 970 and i686  
architectures.  It is also pretty easy to compile and has a fantastic  
developer mailing list (cross compiling for Blue Gene was another  
story though.).

There are serial and parallel benchmark systems that are pretty well  
understood, see specifically,

	http://www.gromacs.org/

and

	http://www.gromacs.org/gromacs/benchmark/single-processor- 
benchmarks.html

All this said, there is likely someone at IBM (at Watson, Almaden, or  
Rochester labs) who has already or will be soon be porting gromacs to  
the cell architecture.

regards,

Nathan Moore

On Dec 21, 2006, at 9:21 PM, Rob Sanders wrote:

> Hey folks,
>    I was talking with Kai a couple of days ago and he recommended
> taking
> this question
> to the YDL list, so here it is...
>    Has anyone done (or know a good link for) some performance
> comparisons of the PS3/Cell
> to "current" top performers in the PPC and x86 arenas?  Even something
> relatively "trivial"
> such as multiplying 2 "large" arrays of floating point numbers.
> Ideally processors like
> say a G5 without using altivec, a G5 with altivec, a recent Xeon or  
> AMD
> with and without SSE[234], and
>   the Cell?
>    I know just enough to be considered dangerous about vector
> programming, and I'm trying (in my
> "copious" spare time after family and daytime job) to boot strap  
> myself
> on how to work the CELL....
>
>
>
Rob
>
>
>
>
> _______________________________________________
> yellowdog-general mailing list
> yellowdog-general at lists.terrasoftsolutions.com
> http://lists.terrasoftsolutions.com/mailman/listinfo/yellowdog-general
> HINT: to Google archives, try  '<keywords>  
> site:terrasoftsolutions.com'




- - - - - - - - - - - - - - - - - - - - - - -
Nathan Moore
Physics
Winona State University
AIM:nmoorewsu


_______________________________________________
yellowdog-general mailing list
yellowdog-general at lists.terrasoftsolutions.com
http://lists.terrasoftsolutions.com/mailman/listinfo/yellowdog-general
HINT: to Google archives, try  '<keywords> site:terrasoftsolutions.com'