Hi,
Original question:
> Hi people,
> In the process of evaluating the performence boost due to the installation of
> Weitec's SPARC POWER uP, I observed the following. I ran a tiny c program
> which goes through a loop of floating point multiplications.
> I got the following results:
>
> Sparc2 Sparc2+Weitec Sparc10/41 Sparc10/51
> ------- ------------- ---------- ----------
> 26.7 10.2 10.1 8.1
>
> Compiling with 'cc -O' resulted in these results:
> 16.3 8.0 7.7 6.0
>
> These are results (for user time) reported by '/usr/bin/time' when running the
> following program:
>
> main()
> {
> register int i;
> register double c = 1.0;
>
> for (i=1; i< 100000000; i++)
> c=c*1.000005;
>
> }
>
> Sun rates these machines as follows:
> Sparc2 Sparc2+Weitec Sparc10/41 Sparc10/51
> ------- ------------- ---------- ----------
> MFLOPS 4.2 (?) 17.2 21 (?)
> SPECfp 22.8 31 67.8 83.0
>
> Finally comes my question:
> How come the difference between the Weitec and Sparc10/41 and Sparc10/51
> are so minute ??? (will summarize)
Basically the concensus was that such benchmarks are quite meaningless and
that application specific benchmarks should be run.
Thanks to:
don@mars.dgrc.doc.ca
Piete.Brooks@cl.cam.ac.uk
casper@fwi.uva.nl
jon@alpha.ee.ufl.edu
eckhard@ts.go.dlr.de
Craig.Warner@Ceram.COM
lct@ai.mit.edu
jdd@db.toronto.edu
szh@zcon.com
peter@jrc.nl
stern@sunrise.East.Sun.COM
-- ========================================================================== || Eli Lopez, Computer Center Manager, National Semiconductor, Israel || \\ Phone: 972-9-594-236, Fax: 972-9-558-322 // ====================================================================
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:08:57 CDT