New IBM Graphics Workstations

[Dave Ciemiewicz]

In article <1990Jul29.165033.22289 at portia.Stanford.EDU>,
dhinds at portia.Stanford.EDU (David Hinds) writes:
> In article <9007261139.AA05802 at aero4.larc.nasa.gov>
blbates at AERO4.LARC.NASA.GOV ("Brent L. Bates AAD/TAB MS361 x42854") writes:
> >
> >    The last I heard was that the IBM's were comparable to the bottom
> >of the SGI line, 4D/20 with minimal graphics.
> >--
>     The IBM RS6000's have several levels of graphics support.  The 8-bit
> color 3D graphics level is quoted as doing 90K 3D vectors/sec, and 10K 3D
> polygons/sec.  The 24-bit color 3D graphics system is quoted as doing
> 990K 3D vectors/sec and 120K 3D polygons/sec.

Has it been released yet or is it still vaporware?  Something else to
consider is that IBM's high-end graphics board is an IBM proprietary
and does not run the GL.

>     For comparison, SGI says a base IRIS 4D/50 with 8 bit planes does
> 140K vectors/sec and 5.5K polygons/sec.  An IRIS with GTX graphics is
> supposed to do 475K vectors/sec and 100K polygons/sec.

Don't forget the VGX system's 1M vps and 1M pps.  This board uses the GL
unlike IBM's top-o'-the-line.

>     It is not obvious how to compare the figures, however.  The IBM
> report quotes the lengths of vectors and sizes of polygons used.  I looked
> through all our SGI stuff and couldn't find the corresponding information.
> These were the same SGI tables that quote 100 MIPS and 50 MFLOPS for a
> 4D/240, which are the theoretical limits, rather than performance on any
> standard benchmark.

I'm not sure what tables you are looking at but the tables I've seen SGI quote
for the 4D/240 are 4 cpus x 20 VAX Dhrystone MIPS = 80 MIPS and a floating
point performance 16MFLOPS double-precision Linpack.  I've never heard they
numbers you quote.

The graphics performance numbers for the Personal Iris and GT graphics systems
vps are based on 10 pixel, connected, full 24-bit color, arbitrary orientation
vectors.  For pps, 10x10 (100 pixel), full 24-bit color, unlighted, Gouraud
shaded, Z-buffered, arbitrary orientation, polygons.  The numbers for the VGX
system are based on anti-aliased vectors and polygons in triangle meshes.

> 
>  -David Hinds
>   dhinds at popserver.stanford.edu

My own personal opinion is that most quoted benchmarks are useful only as a
guideline, actually mileage may vary due to driving conditions.  The really
interesting performance comparisons come for independendant vendors who run
application scenario comparisons between various platforms.  These aren't
benchmarks in the traditional sense of having a very specific test of floating
point performance.  These comparisons are based on tests that similate
trying to do actual work and take into account full system performance
(CPU, disk subsystem, graphics, et cetera.)  These, to me, represent the
really interesting comparison points.  Some of the vendors use these numbers
to determine the pricing of their software on various platforms so it would
seem these kinds of comparisons are interesting to them too.

Of course, the bottom line is what is the real performance of your personal
tasks.  If you can port your application to make the comparison.

						--- Ciemo