sharieh@bind.cs.fsu.edu (a sharieh) (04/11/91)
Dear thinkers,
I am running a Fortran code on the Intel iPSC/2.
If I define T_com to be the time to send a data unit of type
floating point from one processor to another, and
T_stu to be the time to start up a message( sending a message of
length 0), and
T_cal to be the time to do one floating point operations such as
a=b*c or a=b+c,
then what are the values of the following ratios:
T_com/T_cal, T_stu/T_cal?
Thank you in advance.
A. Sharieh
--
=========================== MODERATOR ==============================
Steve Stevenson {steve,fpst}@hubcap.clemson.edu
Department of Computer Science, comp.parallel
Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabellpriol@irisa.fr (Thierry Priol) (04/13/91)
>From article <1991Apr11.190317.23509@hubcap.clemson.edu>, by sharieh@bind.cs.fsu.edu (a sharieh): > > Dear thinkers, > > I am running a Fortran code on the Intel iPSC/2. > If I define T_com to be the time to send a data unit of type > floating point from one processor to another, and > T_stu to be the time to start up a message( sending a message of > length 0), and > T_cal to be the time to do one floating point operations such as > a=b*c or a=b+c, > then what are the values of the following ratios: > T_com/T_cal, T_stu/T_cal? > Benchmarks on the last release (3.2) of the NX/2 give these results: T_msg = T_stu + k * T_com 1) message length <= 100 bytes T_stu ~ 360 micro-seconds / bytes T_com ~ 0.359 micro-seconds / bytes 2) message length > 100 bytes T_stu ~ 704 micro-seconds / bytes T_com ~ 0.359 micro-seconds / byte 3) Concerning T_cal, I have no result with FORTRAN) -- Thierry PRIOL Phone: 99 36 20 00 IRISA / INRIA U.R. Rennes Fax: 99 38 38 32 Campus Universitaire de Beaulieu Telex: UNIRISA 950 473F 35042 RENNES CEDEX - FRANCE E-mail: priol@irisa.fr -- =========================== MODERATOR ============================== Steve Stevenson {steve,fpst}@hubcap.clemson.edu Department of Computer Science, comp.parallel Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell
jet@karazm.math.uh.edu ("J. Eric Townsend") (04/19/91)
>From article <1991Apr11.190317.23509@hubcap.clemson.edu>, by sharieh@bind.cs.fsu.edu (a sharieh): > T_cal to be the time to do one floating point operations such as > a=b*c or a=b+c, The i860 can do a single precision operation per cycle, and operates at 40Mhz. In double precision, it can do an add in one cycle and a multiply in two. If you have twice as many adds as mults, then you could see a peak of 60MFLOPS. Intel claims to have hand coded some stuff up (LOOCS) to 32MFLOPS or so. -- J. Eric Townsend - jet@uh.edu - bitnet: jet@UHOU - vox: (713) 749-2120 Skate UNIX or bleed, boyo... (UNIX is a trademark of Unix Systems Laboratories). -- =========================== MODERATOR ============================== Steve Stevenson {steve,fpst}@hubcap.clemson.edu Department of Computer Science, comp.parallel Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell