[Bioclusters] Itanium 2

[Jeff Layton]

Joseph Landman wrote:

> On Tue, 2002-10-01 at 09:23, Donald Becker wrote:
>
> > Our customers started receiving their first commercial HP Itanium 2
> > systems late last week.
> >
> > Prior to that, all of the previous Itanium 2 system were at a tiny number
> > high-profile, well-publicized sites with special agreements.  There were
> > no quiet users with unbiased opinions running a cluster that you just
> > didn't hear about.
>
> I am concerned about the performance issues on bioinformatics codes.
> Early versions of the chip (including the Itanium 1) had some ...
> problems ... with performance on common informatics applications.  This
> certainly may have been due to compiler quality, as the Itanium series
> performance is tremendously sensitive to the ability of the compiler to
> optimize for the chip.

I think an aspect of this is the specint performance of the CPUs. The
Itanium 1 did not have stellar specint performance at all (rather pathetic
actually). The Itanium2 has better specint performance but it's not all
that great as well.

Intel gives a specint base number of 810 (I didn't see anything about
compilers, OS, MB, etc.). Looking around on spec.org we can see the
following numbers:

Athlon XP2600+: 813
Intel P4/2.26: 818
Intel P4/2.4: 852
Intel P4/2.8: 976
Intel P3/1.0: 442
Intel P3/1.13: 461
Intel P3/1.26: 611
Intel P3/1.4: 648

One can see that consumer CPUs can easily beat the Itanium2 on
specint performance. Of course, specfpu is a different story, but I'll
wager the typical Bio codes are very heavily dominated by specint
performance :)

>
>
> I know that Intel released a new set of compilers.  I dont have an
> Itanium 2 to do some BLASTs these days, so I can't really test the newer
> systems.  Hopefully someone can.
>
> > As is usual in this type of product roll-out, the lead systems vendor
> > (HP) will have weeks or months of near-exclusive delivery.  The tier-2
> > vendors will get only a few sample systems.
>
> True.  HP is pushing these pretty hard everywhere.  Unfortunately I
> don't know how good a fit they are for string comparison codes (lots of
> bioinfo) with small amounts of statistics.

HP is being VERY aggressive. I have first hand knowlege of this.


>
>
> --
> Joseph Landman, Ph.D
> Scalable Informatics LLC
> email: landman@scalableinformatics.com
>   web: http://scalableinformatics.com
> phone: +1 734 612 4615
>
> _______________________________________________
> Bioclusters maillist  -  Bioclusters@bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters

--

Jeff Layton
Senior Engineer
Lockheed-Martin Aeronautical Company - Marietta
email: jeffrey.b.layton@lmco.com

"Is it possible to overclock a cattle prod?" - Irv Mullins