Slashdot Mirror
Final Fantasy At 2.5FPS
Rikardon writes: "Adding a little fuel to the ATi-vs-NVIDIA fire started earlier today on Slashdot, NVIDIA and Square are showing a demo at SIGGRAPH of Final Fantasy: The Spirits Within being rendered in 'real time' (four-tenths of a second per frame) on a Quadro-based workstation. Now that I think of it, this should also inject new life into this debate." Defender2000 points to the Yahoo article. Update: 08/14 09:30 PM by T : Original headline was wrong, said ".4FPS" but as cxreg pointed out, .4 frame per second isn't .4 seconds per frame. Sorry.
Yeah, well, it doesn't take long to string clichés together....
-jKarma: T-rexcellent.
Rendering a movie at 320 x 240 or 640 x 480 is much easier than rendering it at the resolution and size of a movie theater's screen. If the Quadro was rendering the movie at 100 x 75 pixels, all this doesn't mean much.
*This page intentionally left pointless*
0.4sec/Frame rendering is more powerful than Excel has EVER had. This is the first time I've a convincing argument in favor of Quattro based systems.
o/~ Join us now and share the software
And modern cards can't do blur, depth of field, etc. etc.???
I think the point was to show that real time animation of the quality presented in the FF Movie is almost at hand.
I had a friend in the early 90's in the computer animation field who was wowed when his first 486 with an astounding 8mb of RAM could render a full frame of a 640x480 scene in under an hour or so. So I can imagine that wherever he is now, he's happier than can be.
And yeah, if they wanted to demo some huge frame rate, they could dump the textures to a lower quality..but then it wouldnt be all that impressive now, would it?
So, yippee, it can render fast...too bad that has NO BEARING on the actual quality of the production (with the possible exception of the team gets to iterate on the work a little more).
will this improve the plot?
The FF render times sound about the same as numbers I heard from Pixar about Toy Story. What was that post a couple weeks ago, about the machine you want always costing $5000? Well, the frame you want to render will always take 90 minutes.
Yep, very true. Also, I think that a lot of people aren't considering that even though frames might have taken 90 min. on an SGI the entire frame is not rendered all at once. I don't know for sure if the 90 min. refers to the entire frame, but I doubt it. There are layers upon layer for backgrounds, main characters, the ghost alien phantom things, shadow passes, reflection passes, caustic passes (in rare cases) and on and on.
This Wiki Feeds You TV and Anime - vidwiki.org
Can be found at http://www.nvidia.com/view.asp?IO=final_fantasy
The article (on yahoo) is pretty exagerated and sensationalistic, but the images are still very impressive, even they are about what you would expect at 2.5 FPS with such a powerful card. I think it is a pretty good indication of what the next generation of console games (after gamecube and x-box) will look like.
This Wiki Feeds You TV and Anime - vidwiki.org
I see this a lot, and I feel a need to correct it in hopes that the correction meme will spread beyond this page and infect at least a few people. So here goes.
You say that Intel is 100% X86 and AMD is 99.9% compatible. You state it as if obvious, and it sure _sounds_ obvious, but any meaning that can be attached to that statement is either false or irrelevant.
The x86 architecture is documented in technical manuals published by Intel. Actually, I'm going to specify the "ia32 architecture" and thus ignore anything before the 386. Anyway, these manuals detail what the processor is going to do when fed certain instructions. Assemblers are written to these documentations.
The first thing you have to realize is that each Intel processor has a different technical manual, because there are instructions added with every major revision. So if we take any given Intel CPU model (say, an i80386DX), and compare it to any other processor model (Pentium II), we can say with certainty that they are not 100% compatible. The Pentium II will react differently if given certain instructions; for example, it will process MMX instructions instead of objecting to them. You could not say these processors are 100% compatible and retain any meaning in "100%".
Secondly, there are bugs. No Intel chip matches the specifications perfectly, and so every chip would be slightly under 100% compatible even with its own manual. (Yes, you hear about very few bugs, but there are many more that aren't really very important that you can read about on Intel's site if you want).
Now we can say that AMD chips are no different from different Intel chips. Some AMD chips have capabilities like 3dnow! that an i80386DX does not have (and this is admitted by the chip in its processor flags). But that's no different from a Pentium II having MMX. And AMD chips have bugs also, but there's no evidence to show that AMD chips have more bugs or anything.
Intel chips are not "100% X86 compatible" just because Intel makes them. That's like saying that Windows NT is 100% MS-DOS compatible just because Microsoft made it and can define what MS-DOS is. Even Microsoft will admit that certain applications which will run under MS-DOS will not run on Windows NT.
And just as an aside, there are no undocumented instructions of even the most remote practical significance in Intel chips. Undocumented instructions are ignored by assembler and compiler developers.
I have seen the future, and it is inconvenient.
While it is impressive to see the movie rendered in real-time (with adjustable lighting sources and shadows and reflections) it really doesn't look as good as the movie did. I'm at siggraph now (just saw the demo five minutes ago) and the interactive polygon rendering techniques just can't match the radiosity/raytracing used for professional moves - its getting close though!
60 frames per second divided by .4 (frames per second) = 150. If we oversimplify and apply Moore's law to the speed of 3D processors, we will halve this every 18 months.
As I see it, we are about 7 - 8 years away from this kind of rendering in real time.
Thoughts? Comments? Complaints?
Life is the leading cause of death in America.
- Size of rendered frames. What resolution was NVIDIA rendering out, maybe 640x480? 1024x768? FF was probably rendered out at 1880x1024 (about 2-3 times the number of pixels as compared to 1024x768) if not more.
- How did they have to massage the data before passing it to the rendering pipeline? I hear FF was rendered with Renderman
... are they claiming they can render RIB files through the Quadra chipset? If not, how much time does it take to convert/cook the data? If so, then ... wow
- How good did it look in the end? Were all the elements rendered properly, and does it really look anywhere near as good as the movie we saw in the theatre?
Don't get me wrong, I'm excited to see this kind of technology coming, I can totally see this replacing, or at least complementing, our Linux render farm at some point in the future. But it sure would be nice if we had some usefull technical details to qualify this 90 mins verses--M
It doesn't actually take 90 minutes per frame to render. That is if the rendering was done on a single CPU. Square used a massive renderfarm so each frame took a variable amount of time based on the complexity of the rendered image and the fraction of the farm dedicated to that particular render operation. That's why you see things like "Final Fantasy took 1 million years to render" or whatever when you know it isn't exactly true. Look at ArsTechnica where they did an interview with some people from Square about the rendering process. I think there was even a slashdot article about it.
And yes, it's a little rediculous for NVidia to suggest that their card is 100k times faster than Square's rendering hardware for FF-TSW. But what's more rediculous is that yahoo took that statement and printed it in its article with no explanation of exactly what NVidia means when they say that.
The user could interact with the movie and affect the animation in real time. Or, to put that in perspective, imagine fragging your office mates in a photo-realistic Quake VIII. :)
.4FPS is NOT the same as "Four-tenths of a second per frame" which is it?
That's pure film. Anything using an optical printer(used for special effects, titles, CG stuff) or other method for combining film and computer effects needs to be at least twice as fast.
The following sentence is TRUE. The previous sentence is FALSE.
Not only that, but holy bejeezus, there isn't a single link to the pertinent information in the submitter's italicized text. Timothy had to pull the story link out of some other submission. Come on people, I don't care about your freaking thread on Slashdot in the last 8 articles that mentioned Nvidia or SIGGRAPH or Squaresoft, I want to see the story.
This space for rent. Call 1-800-STEAK4U
As I pointed out previously, NVidia's "Quadro" and "GeForce" lines are actually the same hardware. GeForce 2 boards can be "converted" to Quadro 2 boards with a jumper.
The GeForce 3 and "Quadro DCC" boards both use the NVidia NV20 chip, have the same driver, and appear to be very similar if not identical. It's hard to find differences in the feature set. Only ELSA (which is basically a unit of NVidia) sells the Quadro DCC, and apparently only through 3DS Max dealers, along with a special 3DS MAX driver. It's more of a private label than a real product line at this point.
From the: Who-needs-a-woman-when-you-have-a-video-card dept.
.4 (dipstick)
Ok, quick Geek Test: If, upon reading this news post (despite the ditzy title), you did not instantly gasp, shiver, or become aroused, you are NOT a geek. Period.
Which sort of answers my question to my friend after we watched FF for the first time. Is this the top of our abilities in CG? Or was it a matter of the producers saying, "Um...no. We can do a LOT better, but we'd have to wait 100 years for it to build/animate/render instead of 2, so we cut it down to size."
If that is the case, then it's just a matter of BBF (Bigger, Better, Faster (tm)) in terms of hardware before we see something twice as good as FF. Otherwise, if this is the height of skill we have, then we're talking development of new technologies and methods of doing this sort of detail before we see something else come out.
I'm no graphics expert, so maybe someone can answer that question for me. At any rate, the movie still made me shiver. Now I can watch it on my desktop...at 2.5fps, not
Blog,Twitter
were they just rendering it on a 21 inch screen or rendering it at what must be the fantastic resolution needed to get it to look right on a giant movie screen?
Photos.
I should have gone into greater depth. There are two major methods of flipping the standard 3:2 35mm frame to a wide aspect ratio.
The first is called a matte whereby the top and bottom of the film are simply not used.
The second method involves using an anamorphic lens which stretches the image vertically to fill the full frame. In the theaters, another lense is used to do the reverse.
There is actually a third method which is used by a few high profile directors involving a lens which projects onto the sound strip area of the frame in conjunction with a matte, but I won't go into that.
Anyhow, a great site which explains most of this stuff is http://www.hometheaterforum.com/home/wsfaq.html.
The world is neither black nor white nor good nor evil, only many shades of CowboyNeal.
- Square has tie-ins to Sony (exclusivity clause of Final Fantasies on the PS1, rights to publish the movie).
- Microsoft has tie-ins to nVidia (nVidia makes some of the chips for the XBox).
-Square now has tie-ins to nVidia with this demonstration.
Does this mean that more Square games will get ported to the nVidia chipsets, most notably Final Fantasy for the XBox? If I had a choice between the relative hardwares (rather than my PC, which would come first) I'd love to see what Square could do with an nVidia chipset.
Not as far as you might think. Actually, it's about 100 feet down the hall from where I'm sitting right now.
Consider the resolution. Images rendered for film are typically done at about 3000 x 2000
The final product is about half that. Resolutions up to 4k x 3k are used for intermediate special-effects editing. Digital cinema will be operating about HDTV resolutions.
we're left with another problem: you can't record VGA signals on film.
With digital cinema, it's (roughly speaking) VGA in - so you could sit in the theater watching what's being generated at that moment. For film, it's not that big a deal to replace the VGA out with a connector to a real-time film recorder (remember the hall I mentioned above? go to the other side of the hall).
Can we get a "-1 Wrong" moderation option?
Lucasfilm's Sony camera, on which they have filmed Episode II, and which was considered to completely supercede analog film, picks up 1920x1080 resolution. You don't really need that much resolution to look fantastically better than what passes for film these days.
No, he was absolutly correct. Toy story is not close to being rendered in real time yet and this isn't the same. There are many details to the REYES architecture that is used in PRman and likewise used to render toy story. One is the subdivision of NURBS patches and subdivision surfaces to the pixel level. Another is the surface, light, and volume shaders used. There are many many things that people are missing when they say 'Movie X rendered in real time'. What they really mean is 'Movie X rendered in near realtime, at a MUCH lower resolution, with a bajillion hacks to make it look as close as possible to the original.'
This Wiki Feeds You TV and Anime - vidwiki.org
What res? Film is usually somewhere well above anything a GForce can touch. 640x480 != 2048x1152 (or higher for Super 70 mm).
Also, 2.5 FPS isn't "real time". 24 fps film is "real time". 30 fps on video is "real time".
HOWEVER, this would be incredibly useful for generating dalies; spot render checks; web-based trailers and streaming video; Television-quality animation; etc.
Now you can PROVE to a director that a plot sucks, even in final form, and no all the whiz-bang graphics don't help!
Learning HOW to think is more important than learning WHAT to think.
"4/10s of a frame per second means you can do just over 2 frames per second."
Wouldn't it be 4/10s of a second per frame means you can do just over 2 frames per second?
It's kinda like, "One small step for man, one giant leap for mankind." :)
--
Mod up a post Rob doesn't like and you'll never mod again
Heh, if you think I'm going to the beach with .4 SPF sunscreen on, you're out of your mind!
Disconnect your television. Do your own research. Draw your own conclusions. They're probably lying. Don't be a sheep.
Maybe I am mistaken, and the decimal point was there the whole time... I thought I cut n' pasted it directly though. Oh well... I won't lose any sleep over it. :)
"Mind, as manifested by the capacity to make choices, is to some extent present in every electron." -Freeman Dyson
Notice how the official "press release" doesn't state the resolution it was rendered at? What's the movie resolution? Several thousand by Several thousand I imagine. Does doing it a 640x480 or LOWER mean the same thing? I have a hard time believing that a Quadro Setup can render something in .4 of a sec that their SGI setup takes 90mins to do. If Nvidia WAS INDEED 100,000 times faster at this using a Quadro setup, we might have heard of this before?
Something's missing from this methinks.
3-Server OC-3 Linux Counter-Strike Cluster
www.rnp.ca
and believe me, there's no way nvidia's chip came anywhere remotely close to that of the movie.
Until their chip can produce a single frame that matches the image quality, they're still just making toys for quake fiends. Diffraction, interference, antialiasing...just a few of the photorealistic rendering staples, and nvidia has only recently been able to do antialiasing. They've got a long, long way to go before we're going to see actual movies rendered using their hardware.
Jesus! We do math good at slashdot.
4/10s of a frame per second means you can do just over 2 frames per second.
God damn. People go to college and come out knowing this much about math?
"Remember when the U.S. had a drug problem, and then we declared a War On Drugs, and now you can't buy drugs anymore?"
So it renders to a computer screen at 2.5 FPS. That's nice and all, but a long way from film-making.
Consider the resolution. Images rendered for film are typically done at about 3000 x 2000 (give or take depending on aspect ratio, etc). Now, even assuming we could gang up 16 or 25 or whatever of these nvidia boards, we're left with another problem: you can't record VGA signals on film. All the hardware shortcuts and special-purpose circuitry in the latest video card are useless when it comes to final render for film, because they're not built into the gadget (and there are several different sorts) that's actually bombarding the emulsion with photons. (Typically some sort of three-pass (R,G,B) laser scanner).
Yes, it'll make for wonderful computer games (if you like that sort of thing) and maybe even some interesting experiments in real-time porno animation, but it doesn't do much for the film industry, nor would it at 10 times the speed (24 FPS is typical movie framerate). It'd have to be about 250 times faster for full framerate, full resolution images. About 12 years at Moore's Law rates. (Althogh I suspect at that resolution the flaws in the rendering and physics would become very distracting.)
(Actually, it helps the film production process, where animators can preview their work that much quicker. Faster graphics is always good, just let's not get carried away with the hype.)
-- Alastair
Generally considered to supercede analog film? Please. 2K resolution is really just marginally passable. High quality work is generally done at 4K with some even at 8K, which is what it really takes to match 35mm at its best.
-- It only takes 20 minutes for a liberal to become a conservative thanks to our new outpatient surgical procedure!
by... who... where is your math....
Has the whole world gone mad?
a new class of people emerged... the innumerates...
.4 FPS != 4/10 second per frame.
"Prefiero morir de pie que vivir siempre arrodillado!"
I may be very, very wrong, however it was my impression that the "rendered" graphics of modern video cards are shortcut 3D images that are very, very unlike raytraced images: i.e. Quake3 looks nice, but it looks absolutely nothing like the stunning beauty of a Truespace or 3dsmax image (i.e. one is averaging surface point lighting, whereas the other is actually tracing the rays of light throwing shadows, umbras, etc). I though the Quadra cards were only really relevant for modeling (i.e. moving stuff around and such), but they still used an FPU for the real rendering.
From IMDB:
Final Fantasy was shot at 1.85:1.
Anyway, movie aspect ratios have varied ever since the advent of TV. Movies were originally all shot at 1.33:1, and when TV was popularized, it used that aspect ratio. The movie industry was panicked that TV would steal all its customers, so it came up with all sorts of names for new and exciting aspect ratios like "panavision" and "cinemascope". It had nothing to do with technical matters like shooting on 35 millimeter film (and not all films are shot on 35 millimeter, BTW), though, and everything to do with marketting. Because different companies used different systems, the aspect ratios varied wildly by film. Today, the aspect ratio is a choice of the director. 1.85:1 is the most common, but not the only one by any means, and is mostly used for movies where the look of the film is secondary. Special effects movies usually use something bigger. here is some more info.
The cake is a pie
I just saw the demo. It looks NOWHERE as good as the real movie, geometry and lighting/textures are all greatly simplified. The action is definately watchable, which implies > 1 fps. (Looks like around 5-10 fps to me). Aki's hair is made up of WAY less strands (but thick, so she's not balding) and the skin textures aren't as detailed. Still, very impressive. If what I saw was a video game, it'd be GREAT!!!
what does realtime rendering give us?
1. "What now, master?"
2. "Now turn around, bend down and touch your toes!"
If you have a Geforce3, go find the Zoltar demo. It's on the web if you look hard enough. Something like 220 megabytes worth of crap, and all it does is model and animate a human head. But HOLY SHIT does it look incredible! Also, find the Chameleon demo. Again, Google is your friend.
Vintage computer games and RPG books available. Email me if you're interested.
Final Fantasy was anime. Since when have we expected anime films to have good scripts? OK, The Matrix had a pretty good script. Apart from that.
sub f{($f)=@_;print"$f(q{$f});";}f(q{sub f{($f)=@_;print"$f(q{$f});";}f});
1.5 million verticies/2.5 fps, gives about 600,000 verticies per frame. Since the average human has less than 100K hairs, it works out. So what exactly is wrong with their math? Of course the Geforce3 doesn't render at 2000x1600 at full quality (it can't since it doesn't do a lot of the rendering techniques) but that level of geometry is still pretty damn impressive for a sub-500 dollar vid card.
A deep unwavering belief is a sure sign you're missing something...