 | | | Story about Rendertime - RE: OGL acceleration question | Story about Rendertime - RE: OGL acceleration question 2004-06-08 - By Michael Klein
Back
A magical number in 1989 at a company which used the hardware PIXAR
renderman metal cube with 2 proc boards was max. 5 minutes per frame. It
wasn't allowed to render in higher times per frame because of the big amount
of costs they had to control. 4 seconds of a complex animation we would
laugh about today cost up to EUR 120.000. There was no raytracing with
shadows or refraction possible within the time and budget. Unlocking drop
shadows with Alias 2.5 was forbidden to use for production. A stupid picture
on my AMIGA 1000 rendered 16 hours. Crap. Inefficent.
In 1992 I started to work with a Personal Iris. Thanks, Big Bang! Because
trained by extreme limitation rules of my AMIGA 1000 I made a big jump
forward but found out very fast that even this dream machine had its limits
... reached very fast. So a complex scene was something with 150.000
pologons and a damn tiny thumbnail flipbook (nice icon on a OS X desktop) of
something (I thought it was) complex took days to render. I remember my
diploma working time. 12 days day and night for a 8 seconds trailer. I took
the hardware at home to have the chance to finish the trailer in time under
nerd conditions. So the final rendering was 45 minutes per frame. I had to
find other Softimage user which could help me out.
The next step was the nice Indigo Extreme. A big step forward but I learned
the main 3D rule: once you have the faster machine you will play around with
all those little forbidden road show features (and turn them off later one
day before the deadline).
A big multiprocessor Onyx with some Indigo's in a network destroyed all my
sensibility for AA settings and other stuff in 1995. Especially if you
didn't bought the stuff. So it was getting normal to turn on everything
before starting the rendering. O.K. ... nice luxury ... but what will happen
if you are forced to go back from Porsche to Beetle. Freeze!
And even how fast your hardware is ... you think: if I only had more CPU
power it would take minutes to finish this job ... but that's an illusion.
Today our VR ammo which is just a bunch of 9 2,8 GHz CPU's take hours to
render a 12 pass 500.000 polygon mobile phone animation. Sometimes I have
the feeling: are we doing something wrong here? Is is probably getting to
much overdone to split every project into layers instead of just rendering a
single one? Is this the opposite C64-Syndrome (I mean: the more advanced the
hardware is, the more inefficient it will be used or programmed ... wasting
RAM and Power).
I'm not sure 100% anymore ... in my point of view the way of different
working makes sense and it's some kind of solving the problem professional
using special knowledge, skills and a very special tool. But does the client
really understand why a 12 layer mobile phone will render many hours on a
fast little farm to get a multilayer high quality post-controlable result?
The most bugging question is: why so complicated ... why not rendering just
one simple, clean and perfect layer ... (I'm tired about explaining it again
and again ...). It's not so easy anymore to impose those Stone Age guys out
of the Valley of the Unknowing.
The good thing with faster technology is: we set AA to 1 3 if something is
still flickering and we don't care ... but the negative thing is the
feeling: wasn't that too easy? Is 10 minutes per frame probably waste of CPU
time? Why does this little thing fill up my 2GB of RAM?
I don't know ... probably routine-blinded by the high speed growth of
technology.
(just my thoughts during a little break)
President
-- --Original Message-- --
From: owner-xsi@(protected) [mailto:owner-xsi@(protected)
<mailto:owner-xsi@(protected)> ] On Behalf Of kim aldis
Posted At: Tuesday, June 08, 2004 5:44 PM
Posted To: XSI
Conversation: OGL acceleration question - RE: +++ Slower Rendering in XSI
4.0 +++
Subject: RE: OGL acceleration question - RE: +++ Slower Rendering in XSI 4.0
+++
in the past 25 years, I've seen my frame times go from around 20 minutes per
frame to, well, around 20 minutes per frame. We just want more.
Something Loren Carpenter said a good few years ago was that he felt render
times were limited rather more by any given person's attention span than by
the power of the hardware and I think that still stands up. Maybe the time
will come but I rather doubt it.
Carpenter also has an interesting story about his first presentation of the
ray tracing work he was doing way back when - he was one of the original
developers of ray tracing for image synthesis. He was presenting to
management at, I think, IBM, who were wowed. So much so that one of them
asked what he'd need to do it in real time - remember, this was some 20
years ago. Carpenter thought for a moment, then said, "I'd like a helicopter
and 768*576 Crays in a field with a red, green and blue lightbulb on the top
of each one".
Respect.
---
Unsubscribe? Mail Majordomo@(protected) with the following text in body:
unsubscribe xsi
|
|
|