chuck@amdahl.amdahl.com (Charles Simmons) (06/09/87)
Taking a trip in the way back machine...
A couple months ago, the subject of processing power came up. Someone asked
if we really needed all the processing power that should be at our disposal
in a few years. He suggested that software to use the processing power did
not exist and potentially would not exist. Someone else suggested that
graphics would consume all the processing power we were capable of producing.
This argument was countered by suggesting that processing power was a
small marketing niche.
I would like to present arguments that suggest graphics applications are
far more than a marketing niche. I recently had the opportunity to view
a number of video tapes of animation sequences produced on computers. Many
of these videos are produced on machines like the Cray, and, it is estimated,
take not hours, days, or even weeks, but whole months to produce. So the
software for the applications I will describe below exist today, and they
consume immense amounts of processor time. This suggests that faster
processors will remain a necessity for a very long time to come.
Graphics Applications:
1) Entertainment
A) Special effects sequences in movies.
B) Real time generated animation for video games. This would make
video games possible that reacted to your inputs as an individual and
created a world for you to explore as you explored it. The beginnings
of this idea can be seen in games that store lots of graphics on video
disk and every now and then present you with two or three options for
changing the sequence of graphics displayed on the screen. Real time
generation of the graphics would allow for far more than two or three
choices every few seconds.
2) Advertising
This field would use graphics in much the same way as the entertainment
industry would use graphics for special effects.
3) Military applications (video games for the Pentagon boys)
The obvious application that comes to mind is training pilots in
expensive simulators that display an image of the outside world, updating
and generating this image in real-time.
4) Research
Simulating physical phenonmena on a computer has proven to be
a good way of gaining an understanding about how the universe works.
Some sample simulations are simulating the evolution of a galaxy or
planetary system, simulating the movement of atoms in a molecule,
simulating chemical reactions, etc.
5) Education
Providing students with video images of complex physical processes
that cannot be directly photographed would be a good learning tool.
Allowing the student to interact with the generated image would be an
even better learning tool.
Notice that there is a high degree of overlapping between these various
applications. For example, the military applications and the research
applications are both forms of education. Education, of course, is a
form of entertainment.
The basic concept behind all of these applications is that humans have
high bandwidth visual information channels. It seems clear that currently
computers are not using the bandwidth of these information channels as
effectively as they could be used.
-- Cheers, Chuckbzs@bu-cs.BU.EDU (Barry Shein) (06/20/87)
Posting-Front-End: GNU Emacs 18.41.4 of Mon Mar 23 1987 on bu-cs (berkeley-unix) From: chuck@amdahl.amdahl.com (Charles Simmons) >A couple months ago, the subject of processing power came up. Someone asked >if we really needed all the processing power that should be at our disposal >in a few years. He suggested that software to use the processing power did >not exist and potentially would not exist. Someone else suggested that >graphics would consume all the processing power we were capable of producing. >This argument was countered by suggesting that processing power was a >small marketing niche. I was the contentious party and I think you've missed my views by about 10 degrees (in your own favor.) My argument is that too much attention seems to be placed upon raw processing power and not enough on what exactly the bulk of the users will do with a 100MIPs micro. Raw horsepower is an oversimplification and an easy to understand goal (hence, easy to sell to management.) >I would like to present arguments that suggest graphics applications are >far more than a marketing niche. I recently had the opportunity to view >a number of video tapes of animation sequences produced on computers. Many >of these videos are produced on machines like the Cray, and, it is estimated, >take not hours, days, or even weeks, but whole months to produce. So the >software for the applications I will describe below exist today, and they >consume immense amounts of processor time. This suggests that faster >processors will remain a necessity for a very long time to come. And thus you fall into the black hole. I have never contended that no one needs the extremes of horsepower people are talking about in the near future, just that most people won't if we don't start thinking deeply about what we are going to do with that horsepower right now. Of course the sort of folks who buy Crays (and even less) today will mostly sop up every bit of horsepower they are given, no argument. When Ken Wilson (as related by Bill Joy) says that he needs a 1000X improvement over the fastest currently available problems just to begin working on some problems he has defined, I don't doubt him for a second. I don't doubt he and people like him will need 1000X that when his current dream is realized. High end science is like that, many of their problems have n^3 or worse growth patterns and right now they can only work on tiny examples, they know exactly where they are going and what they need (well, they know enough.) I was referring more to the 100MIPs (or more, or less, pick your own level of skepticism) workstation which seems nearly realizable. I am simply worried about our priorities. Similarly, you fall into another black hole pre-supposing all this wonderful software that might/should someday exist to justify the iron, tho not seriously (actually, most of your examples were pretty good, but there's some basis for skepticism.) I'd be more comfortable if it were running slowly today, or soon, than believe that it will someday be available so we should just go ahead and spend our resources on the iron to put under it (Emperor's New Clothes? Chicken Little? One of these must be apt.) -Barry Shein, Boston University
jcz@sas.UUCP (John Carl Zeigler) (07/06/87)
Here are two pragmatic reasons why high powered processors ( mips * 10eX ) are indeed necessary, and do have a larger niche than weather forcasting: 1. Software performance tuning is expensive. If you can get 'acceptable' performance with a fixed cost processor, then you don't need those expensive performance gurus. "Who cares if we're only using 75% of the machine? My nroff is done by the time I hit return !!!!!" 2. User interface. This is an expanded version of the graphics argument. Many business and other 'volume' applications are becoming more complex internally to provide less complexity and better communications to the user. If want to present something visually to the user, or audially, or in a tabular form that allows them to make their next move very quickly, (I'm not talking 'bout hunt and peck menus here . . ) AND perform the complex sequence of calculations they specified in a decent amount of time, then you need more processing power than is currently available. As fast machines become smaller and cheaper, these two considerations will affect a lot of purchase decisions. -- --jcz John Carl Zeigler SAS Institute Inc. Cary, NC 27511 (919) 467-8000 ...!mcnc!rti!sas!jcz