xengine performance on RS6000

[David Fricker]

In article <1991Apr17.190136.17974 at nrcnet0.nrc.ca> ng at cfd.di.nrc.ca writes:
>We have tried out the xengine on our RS6000 model 530 running 3003.
>The system is configured with 64meg of main memory and a 24-bit, 3D
>and Z-buffer graphics subsystem.
>
>The outcome, which ranges from 41 to 49, is extremely disappointing.
>We also have a Silicon Graphics 4D/25G and a SUN 3/60. Both perform
>about twice as good as the IBM.
>
>However somebody has claimed that on a 550, the xengine runs at 333.
>Has anybody tried it on their RS6000 ?
>
>I suspect it could due to something not configured right in our system.
>Any opinion welcome.
>


In a related vein, we've done some 'benchmarking' using local programs
on an RS6000 model 530 & various SGI Irises.  The relative performance
figures are _very_ strange.  On one CFD program doing a 2D grid, an
SGI Personal Iris (4D/25) outperforms my RS6000 model 530, while a 
3D grid run has my model 530 outperforming a 4D/340VGX Iris! (The 340
means the iris has 4 processors.)  Both programs were written by the
same person & solve roughly the same problem--vortex sheet rollup--but
one program solves it in 2D while the other solves it in 3D.

Any ideas as to why these relative performance figures are so strange?

david fricker
--
-----------------------------------------------------------------------------
David Fricker			|     phone: 216-433-5960
NASA Lewis Research Center     	|     M.S. 5-11
Cleveland, Ohio  44135         	|     email: fsfrick at bones.lerc.nasa.gov