Kuro5hin.org: technology and culture, from the trenches
create account | help/FAQ | contact | links | search | IRC | site news
[ Everything | Diaries | Technology | Science | Culture | Politics | Media | News | Internet | Op-Ed | Fiction | Meta | MLP ]
We need your support: buy an ad | premium membership

[P]
Massively Parallel Ray Tracer

By danimal in Technology
Thu Aug 16, 2001 at 02:34:17 AM EST
Tags: Technology (all tags)
Technology

Square USA, the folks that brought you Final Fantasy the movie have poured a lot of effort in to the technology behind their art. One prominent example of that is the massively parallel ray tracer, Kilauea.


This renderer (described in the SIGGRAPH 2001 course notes PDF on the Kilauea web page) is an incredible piece of work. Anyone interested in seeing some very nice global illumination renders should visit the Kilauea page.

Basically using the MPI standard the developers have build a rendering system that can scale across multiple processors on different machines. They have taken a very thoughtful approach to the design of Kilauea. Breaking a scene's geometry apart and using those pieces across machines to make it fit in memory and then passing ray data around just screams forethought and speed.

The sad part to this story is that Kilauea may never get used. To date Square has lost about US$100 Million on Final Fantasy. I just hope that the work of these developers, to push the limits of reality in computer animation and graphics doesn't get lost.

Sponsors

Voxel dot net
o Managed Hosting
o VoxCAST Content Delivery
o Raw Infrastructure

Login

Poll
I liked Final Fantasy: The Spirits Within
o Yes 57%
o No 28%
o Huh? 14%

Votes: 50
Results | Other Polls

Related Links
o Square USA
o Final Fantasy
o Kilauea
o Also by danimal


Display: Sort:
Massively Parallel Ray Tracer | 18 comments (17 topical, 1 editorial, 0 hidden)
I don`t think square breaks on this... (3.00 / 1) (#1)
by tomte on Wed Aug 15, 2001 at 04:40:39 PM EST

The technologie developed in the past years to make the Movie possible will pay back in the near future.
They wouldn´t have made it without this in mind.
prepare for new SquareSoft-titles with amazingly realistic cut-scenes and graphics...
--
Funny. There's a brightness dial on the monitor, but the users don't get any smarter.
Possibly... (4.00 / 1) (#5)
by nurglich on Thu Aug 16, 2001 at 10:23:37 AM EST

They wouldn´t have made it without this in mind. prepare for new SquareSoft-titles with amazingly realistic cut-scenes and graphics...
Or perhaps for this...

------------------------------------------
"There are no bad guys or innocent guys. There's just a bunch of guys!" --Ben Stiller, Zero Effect

[ Parent ]
My suggestion (4.00 / 3) (#2)
by trhurler on Wed Aug 15, 2001 at 04:41:32 PM EST

Look here for something a bit tastier. This guy really IS pushing the limits; Square, unless I'm misunderstanding, has done nothing more than a straightforward parallel version of the same old raytracing we're all familiar with.

--
'God dammit, your posts make me hard.' --LilDebbie

uh (none / 0) (#4)
by Ubiq on Thu Aug 16, 2001 at 08:50:41 AM EST

What causes the difference in realism between the pictures rendered by the Squaresoft renderer and the renderer you mention?



[ Parent ]
Well (none / 0) (#8)
by trhurler on Thu Aug 16, 2001 at 01:02:12 PM EST

For one, the tachyon project is all about speed. It isn't making mistakes, per se, but it also doesn't support a lot of the fancy "make it look nice" features that aren't really "raytracing" that many raytracers use. I can pretty much guarantee you that Square's renderer isn't going to generate any significant frame rate; in the not so distant future, tachyon will. Think about it. Raytraced animations:)

--
'God dammit, your posts make me hard.' --LilDebbie

[ Parent ]
Tachyon design details (none / 0) (#18)
by johns on Wed Aug 22, 2001 at 03:00:19 PM EST

Hi, I'm the author of Tachyon. One of my colleagues saw this discussion and suggested that I comment briefly. Here are a few quick answers to comments and questions relating to Tachyon. Neither one of them have invented any new fundamental ray tracing techniques at the highest level of the ray tracing algorithm. Both do unique things in their implementations of parts of the standard ray tracing techniques in order to make these techniques parallelize well, and perform well for the purposes for which they were originally developed. It is difficult to make any direct comparison between Square's renderer and Tachyon. Square's renderer is a production tool used to render feature length movies, and caters to the needs of movie production. Tachyon was originally just a research tool for me when I was a graduate student, and even to this day it still lacks many bells and whistles that more fully-featured rendering programs implement. Adding features tends to decrease readability, maintainability and performance, so I've chosen to keep Tachyon as simple as possible. There are already quite a number of ray tracing programs available which do a great job, so I've chosen to aim Tachyon at different targets. Tachyon's main development goal has always been speed. If I had the free time, I could implement all of the features found in other ray tracers, but I'm not sure that the world needs another renderer for that purpose. We do however need more tools that do well in rendering large scientific data sets. Tachyon does very well at doing rendering of large scenes with millions of objects, and runs on a diverse group of platforms. People have used it to render very large planetary terrain data sets (Mars!) complex biological molecules, computational fluid dynamics simulations (volumetric rendering) and various other things. I guess its fair to characterize Tachyon as being more useful for various of these scientific rendering needs than for "photorealistic" rendering, though it could certainly be applied to rendering photorealistic images as well. Someone asked why Tachyon's example images look so simple. That's actually a very easy question to answer: I'm much more of a programmer than an artist, so my test scenes are quite lame!! I've recently been somewhat interested in writing file translators for packages like Blender, which would provide a much more convenient mechanism for trying out photorealistic rendering with Tachyon. Who knows, I may even eventually implement photon mapping using the existing Tachyon as the basis for a more sophisticated program, but right now I don't have time, and there are lots of fun projects for me to work on in the interactive ray tracing realm, so something like that is probably quite a ways off. If people are curious about Tachyon, and the things I mentioned here don't clear it all up for you, please feel free to send me an email.

[ Parent ]
Re: My suggestion (none / 0) (#6)
by danimal on Thu Aug 16, 2001 at 10:32:22 AM EST

Look here for something a bit tastier. This guy really IS pushing the limits; Square, unless I'm misunderstanding, has done nothing more than a straightforward parallel version of the same old raytracing we're all familiar with.
Actualy, the Tachyon linked by you and Squares Kilauea are very similar. Both use MPI and shared memory when possible to utilize multiple cpus on the same or different hosts.

From what I see there isn't anything really new in Tachyon as far as raytracing goes.

-danimal

[ Parent ]

Apparently, (none / 0) (#7)
by beergut on Thu Aug 16, 2001 at 12:13:06 PM EST

from listening to this guy speak, he is looking at ray-tracing from a different angle. Where Square, et al, project rays from an object (and from each object in the scene,) toward a "view", he is essentially working the process in reverse.

Or something.

But the speed of his results are not simply due to his (damned near Godlike) abilities to program, but due to the algorithms he's developed.

i don't see any nanorobots or jet engines or laser holography or orbiting death satellites.
i just see some orangutan throwing code-feces at a computer screen.

-- indubitable
[ Parent ]

Um. . . (none / 0) (#10)
by abdera on Thu Aug 16, 2001 at 01:33:44 PM EST

not simply due to his (damned near Godlike) abilities to program, but due to the algorithms he's developed

Isn't developing algorithms a pretty fundamental part of what programming is? Seems to me that it's a better guage of a programmers abilities than the number of languages used, lines of code written, amount of money made, pr0f1c13ncy 1n 31337 h4x0r 5p33k, etc.

#224 [deft-:deft@98A9C369.ipt.aol.com] at least i don't go on aol
[ Parent ]

Can you... (4.00 / 1) (#13)
by beergut on Thu Aug 16, 2001 at 02:35:05 PM EST

be more specific?

I can program a red-black tree, and make a few niggling optimizations that can cause the resulting algorithm to run a bit better, perhaps be a bit more flexible in the ways it's used, or what have you. But, I didn't invent the red-black algorithm. I just implemented it.

"Programmers", per se, are just that. It takes a little more upstairs to be really good.

i don't see any nanorobots or jet engines or laser holography or orbiting death satellites.
i just see some orangutan throwing code-feces at a computer screen.

-- indubitable
[ Parent ]

Perhaps I overstated (none / 0) (#16)
by abdera on Fri Aug 17, 2001 at 12:08:29 PM EST

the importance in algorithm development as a factor of programming. Certainly implementation and optimization of existing algorithms, among other aspects, are important as well. I would maintain, however, that developing algorithms is an aspect of programming, and a very high form at that.

#224 [deft-:deft@98A9C369.ipt.aol.com] at least i don't go on aol
[ Parent ]

ray tracing order (none / 0) (#11)
by danimal on Thu Aug 16, 2001 at 01:44:31 PM EST

almost every ray tracer fires rays from the view toward objects. when they intersect with a body then the shading algorithms are run.

the tachyon library will provide speed, but without shading that makes it realistic why are you even ray tracing? the reason to ray trace is to get real soft shadows (not shadow maps), reflections, and refractions.

speed without realism will be just that, a fast renderer that looks like butt.

-danimal

[ Parent ]

Heh (none / 0) (#12)
by trhurler on Thu Aug 16, 2001 at 01:58:39 PM EST

The thing is, raytracing times are dominated by the ray projection. If you're doing research into high speed raytracing, obviously you should concentrate on that aspect. Granted, anyone who wants a "practical" application is going to have to do further work, but what we were talking about was parallelizing raytracing, and the fact is, as far as I'm aware, nobody on the planet has yet equalled the tachyon implementation in that regard.

Face it - if not for the parallel aspect, this article would have been "just another raytracer by some company." The author claimed this company is "pushing the envelope" or some similar phrase. Well, they're not. They've got a very good but still very orthodox implementation of very ordinary algorithms. If you want "pushing the envelope," you're going to have to settle for research projects with very specific goals - and you should evaluate them based on those goals.

--
'God dammit, your posts make me hard.' --LilDebbie

[ Parent ]
Same old raytracing? (4.50 / 2) (#14)
by hamsterboy on Thu Aug 16, 2001 at 02:42:13 PM EST

I don't think so. From the lecture notes, it sounds like Square's renderer is more than plain-vanilla raytracing.

Raytracing is a technique that involves shooting one ray for each pixel (or more, for smoother output) from the camera into the universe. If the ray hits an object, the renderer determines, from surface properties and locations of lights and other objects, what color that pixel should be.

Radiosity is another beast entirely, and takes reflection into account. If you have a white tabletop, you place a bright green box on it, and you shine a bright light on them, the tabletop has a "green" patch on it where the light reflected off of the box falls. Raytracing won't do this. Most big commercial applications allow radiosity, often in combination with raytracing algorithms, and some are dedicated to the task.

Now Square's renderer works with photon mapping, which is essentially raytracing with an extra rendering pass, which computes how much each surface is illiminated by reflection from other surfaces. This method is parallel-friendly, is much more time-efficient than radiosity, and the results are comparable.

Kilauea is NOT, however, just plain old raytracing. :)

-- Hamsterboy

[ Parent ]

Not get used? (none / 0) (#9)
by Wormwood on Thu Aug 16, 2001 at 01:12:45 PM EST

I'd imagine that Sqaure, apart from using this for their own internal projects, might lease out its use to other companies...

any thoughts on... (none / 0) (#15)
by poltroon on Thu Aug 16, 2001 at 07:00:18 PM EST

other recently unveiled raytracers, like Entropy?

I still wonder if the future of rendered movies necessarily obviously lies in raytracing. The Final Fantasy movie certainly appears to be pushing the bar in terms of what we've seen of photorealism in CG movies, and it was rendered without raytracing, for the most part; I'm guessing raytracing may have been used in certain shots.

capital costs vs. fluid costs (none / 0) (#17)
by toddg on Fri Aug 17, 2001 at 03:28:42 PM EST

"To date Square has lost about US$100 million on Final Fantasy."
I assume you're talking about the movie -- the series as a whole is still majorly in the black. :)

<rumor> I would like to see the actual accounting numbers for the FF Movie, how the ~$150 million was spent. I was under the impression that $80 - $100 million of it was for capital costs of constructing Square Hawaii (where the movie was produced) and hiring and training all the new people, new this, new that, etc. That would be right in line with estimates of what a new world-class facility would cost. Furthermore, initial projections (a few years back) assumed that several films would be produced and the capital costs would be spread out over those several. Although that may not happen, already some of the character capture and modeling work for the video games has been done there, so that facility is not so far in the red as you might think. </rumor>

I'd like to point out how FFVII cost $70 million. Most of that was constructing the new facilities for Square USA in Santa Monica; their reports said they would amortize it over the next 5 games.

Anyway, accounting is tricky. Things might not be as bad as you think.

Massively Parallel Ray Tracer | 18 comments (17 topical, 1 editorial, 0 hidden)
Display: Sort:

kuro5hin.org

[XML]
All trademarks and copyrights on this page are owned by their respective companies. The Rest © 2000 - Present Kuro5hin.org Inc.
See our legalese page for copyright policies. Please also read our Privacy Policy.
Kuro5hin.org is powered by Free Software, including Apache, Perl, and Linux, The Scoop Engine that runs this site is freely available, under the terms of the GPL.
Need some help? Email help@kuro5hin.org.
My heart's the long stairs.

Powered by Scoop create account | help/FAQ | mission | links | search | IRC | YOU choose the stories!