PHIL: Performance Baseline for VRML-capable Machines

Mark D. Pesce (
Tue, 14 Jun 1994 18:11:28 -0700 (PDT)

Some thoughts about speed and platform independence:

Virtual Reality is computationally expensive. There are no two ways around
that. It must be understood that, just as some machines can not play MPEG
video or QuickTime movies, some machines can not render geometry in real
time. The goal is NOT to make VRML compatible with every machine, but to
establish a baseline, and work from that.

I believe that a 486/25 represents a "low-end" baseline, for the following

A) Such a machine costs just about $1000.00, which makes it an entry point

B) It will be equipped with an SVGA-compatible card, so that 256-color images
can be rendered in relatively high (640x480) resolution.

C) It can handle high-speed communications well (28.8 Kbps modem or Ethernet).

D) Such a machine can sustain a rendering speed of approximately 25,000
Gouraud-shaded or texture-mapped polygons, per second.

It is my strong affirmation that designing a system which can handle old
386's and Macintoshes before the MC68030 (25 Mhz) is just going to hamstring
VRML capabilities while providing essentially NO win. Machines that are
multimedia-capable fit the characteristics outlined above; no one would
think of running "Myst" on a 386/16, even though there are plenty of
machines of that caliber in the world.

Planning for change:

The processor world is changing rapidly. The widespread acceptance of 100
Mhz Pentium and PowerPC technologies, at extremely inexpensive price points
means that the average desktop, in 18 months (which is about how long it
will take to get a real VRML base, with clients and editors, up and running)
will have over twice the performance, perhaps as much as ten times the
performance, of the machines most of us are using now.

We must plan for the future, and unfortunately, ignore the past. If we
don't, then we'll end up with something underpowered, which is the kiss of
death in computing. I don't plan on having our work be obsolete before we
even get out of the starting gate.

There are some innovations which "lower the bar" on which kinds of computers
are capable of rendering interactive virtual worlds; I will be discussing
these innovations in future postings.

Planning for the future:

In the future, they say, we'll have all the MIPS and bits we need to do
anything, anywhere. I remain more pragmatic, and would like to focus on
what kinds of machines will be commonplace in early 1996.

They'll average 16 MB of RAM, and 1 GB of hard disk
They'll have a Pentium/P6/PowerPC at > 100 MHz
They'll have high speed (28.8 Kbps or greater) connectivity to Internet

Such a machine will have a sustained performance of roughly 150,000 polygons
per second, perhaps more; hardware vendors will be introducing a new
generation of video cards, this fall, which provide > 500,000 polygons per
second on a PC-compatible, for $300-$500. Expect these cards to become
commonplace, once enough 3D applications fill the market, much in the same
way SoundBlaster and clone cards are now common.

This is the ideal target platform for VRML, and the one to which
specifications must be written. To sacrifice strength and flexibility on
the altar of backwards-compatibility is like asking the human being to
express the DNA for cyanobacteria just because he's evolved from them.
We must "raise the bar" on what our applications are capable of; without
that goal, we won't develop anything innovative or novel.

Mark Pesce

|| * That's AL * WorldWideWeb