[BioDarwin] [Fwd: Re: [Bioclusters] G5]

[Michael Maibaum]

On Wednesday, June 25, 2003, at 03:05  pm, J.W. Bizzaro wrote:

>
> From: "landman" <landman at scalableinformatics.com>
> Date: Wed Jun 25, 2003  2:50:13  pm Europe/London
> To: bioclusters at bioinformatics.org, bioclusters at bioinformatics.org
> Subject: Re: [Bioclusters] G5
> Reply-To: bioclusters at bioinformatics.org
>
>
> There has been serious discussion on other sites
> (http://www.haxial.com/spls-soapbox/apple-powermac-G5/) discounting 
> some of
> the marketing speak associated with it.  That said, using the 
> specialized SIMD
> versions of BLAST/HMMer, you can indeed achieve significantly better 
> results
> on a SIMD processor than running the non-SIMD on a single processor.

Except this is a known critic of the Mac who's arguments are more 
flawed than apples on this topic.
Especially as what Apple did with Hyperthreading was the same as Dell 
did in there tests. The 'which' compiler to use argument is endless, 
gcc is likely to be superior for the x86 instruction set for code-gen 
than ppc and especially the 'new' ppc970.

>
> However, noting what the folks noted on that website above about 
> benchmark
> "gaming" (e.g. not really doing an "apples" to "apples" comparison, if 
> you
> could pardon the pun, by purposely overoptimizing the machine far 
> beyond what
> an ordinary user would do, tuning the prefetch cache, using a better 
> compiler
> for one and a known worse compiler for the one you wish to show the 
> poorer
> performance for ... ) makes me suspicious if any of that occured for 
> the BLAST
> and HMMer benchmarks.


The Mathematica test was interesting, a real world application and a 2 
fold difference in speed.

>
> That said, the A/G BLAST does look good w/o the gaming, and so does 
> HMMer.
> Hopefully those of us working on the SIMD versions for the IA32/x86-64
> platforms will be able to finally help out with a real "apples" to 
> "apples"
> comparison.  Basic results (using the poor compiler, and no SIMD) on 
> Opteron
> are quite interesting.

I'd certainly like to see G5 vs Opteron comparisons.

Michael


-- 
Michael Maibaum
internet: mike at maibaum.org    |    http://mike.maibaum.org
voice:     [m] 07958 604025        |

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 186 bytes
Desc: not available
Url : http://bioinformatics.org/pipermail/biodarwin/attachments/20030625/502d302b/attachment.bin