Playstation 2 and "Toy Story"
lmu...@my-dejanews.com
have a question to ask. It's about what Sony Corp. recently claimed in their
announcement about the next-generation Playstation. They claimed their next
generation console would do "Toy Story" quality animation in real-time. What
is your response to this claim? I'm a bit suprised since even Scott Sellers
of 3Dfx Interactive said that this a goal of 3Dfx (in other words, 3Dfx
hasn't achieved real-time "Toy Story" quality animation yet), and now Sony is
claiming that they've achieved it already. I just find it hard to believe.
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Gareth H-Jones
--
[][][][][ Gareth H-Jones ][][][][]
gjo...@3dmodel.freeserve.co.uk
http://www.3dmodel.freeserve.co.uk
BMRT Workspace for Rhino at: http://www.3dmodel.freeserve.co.uk/
lmu...@my-dejanews.com wrote in message <7brimj$15t$1...@nnrp1.dejanews.com>...
Guido Quaroni
The next generation render farms will be based on Playstation 2 and
animators will animate using a joistick.
Why spend several thousand dollars on a Sun Server with 4Gb RAM and 14
processors when a Playstation can do the same and in real time !!
There is only "one" problem. Most of the RIB files for a single frame
will fit on one CD or may need multiple CDs.
This means something close to 100.000 CDs for a movie. But this is a
detail...
Sorry for joking here but I couldn't resist. Please don't reply :-)
guido
lmu...@my-dejanews.com wrote:
>
> I'm not sure that this is the appropriate newsgroup to post, but anyway I
> have a question to ask. It's about what Sony Corp. recently claimed in their
> announcement about the next-generation Playstation. They claimed their next
> generation console would do "Toy Story" quality animation in real-time. What
> is your response to this claim? I'm a bit suprised since even Scott Sellers
> of 3Dfx Interactive said that this a goal of 3Dfx (in other words, 3Dfx
> hasn't achieved real-time "Toy Story" quality animation yet), and now Sony is
> claiming that they've achieved it already. I just find it hard to believe.
>
> -----------== Posted via Deja News, The Discussion Network ==----------
> http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
--
Guido Quaroni - Pixar Animation Studios
Email: gu...@pixar.com
Phone: (510)620-3561
Forkazoo
>
>guido
>
Okay, I won't.
<;) couldn't resist>
Kevin Bjorke
> They claimed their next
> generation console would do "Toy Story" quality animation in real-time.
If they have "Toy Story" quality animators, well sure :)
There's significant difference between "animation" and "rendering."
Personally, I'd LOOOOVE to be able to use a PSX2-based system for animation, if
there were enough user controls. The box is darned fast for desktop 3D, maybe
faster than an Octane. A "Net Yaroze"-styled PSX2 in tandem with a fast PC
might make a formidable challenger to existing 3D accelerator packages and
machines like the SGI. That's (probably) not Sony's goal, though.
Wonder which vendor will start taking pre-orders first?
Larry Gritz
>I'm not sure that this is the appropriate newsgroup to post, but anyway I
>have a question to ask. It's about what Sony Corp. recently claimed in their
>generation console would do "Toy Story" quality animation in real-time.
Presumably they mean Toy Story quality rendering?
That just says to me that they have no idea what went into rendering
Toy Story.
-- lg
--
Larry Gritz Pixar Animation Studios
l...@pixar.com Richmond, CA
Michael Herf
Story?
They obviously intend to compare pure geometry performance, which isn't half
the, uh, story, since shaders and lighting (I presume) take a majority of
the time. I think that they are close in geometry -- 75 million polys per
second translates to a little over 4/pixel at video res (and 60Hz), which is
certainly in a Renderman-acceptable range for that resolution.
Still, if Toy Story took several hours per frame on 1993 hardware, we're
still way off, even after scaling down to video (10x) and scaling up to
current CPUs (10x).
The Nintendo hardware gives (or claims to) as much as 50x memory bandwidth
to the CPU as 1993 SPARCs, and the video hardware has access to nearly 1000x
the bandwidth.
Some thoughts: real-time systems use many more static models, which can be a
quality tradeoff due to geometry aliasing. But they can be cached and dealt
with very quickly by hardware.
Right now, the lighting/shading isn't comparable, and I imagine that this
will be a target of future real-time architectures. However, this is
currently the major difference between real-time architectures and
RenderMan-based ones.
I think that a Renderman-exact implementation that leveraged the hardware in
the PSX2 would still be off by a factor of 1000 between Toy Story and
real-time on the PSX2. (That was really back-of-the-envelope.)
However, Renderman is general, and this gap could be shortened dramatically
by making certain limitations in the real-time renderer. I believe that
something visually similar to Toy Story could be done in real-time or
near-real-time on next-generation hardware, if you believe the specs that
are flying around. Obviously, lighting would be weaker, and some parts
would not be as dynamic (and the system that animated this thing is assumed
to be magic), but I think it could happen.
mike
Larry Gritz <l...@pixar.com> wrote in message
news:7bsms8$apr$1...@sherman.pixar.com...
John Cairns
however, I think it is stretching the truth a bit. I'm almost certain
that this quote came from some marketeer in the following scenario.
Marketing person goes to engineering: "Okay we need some cool facts
about PSX2 to hype it up."
Engineering: "Okay, um, it has a 500 MIPS RISC processor, will render
75 million ..."
Note: To the marketing person this sounds like "*** $$$$ *** ... $$$$"
as do all facts.
Marketing: "What?"
Engineering: "Um? You could shade about the same number of objects on
average as were in Toy Story, real time."
Marketing: $$$$ *** $$$ *** $$$
Anyway, there is a grain of truth to it if you understand two facts.
PRMan uses a software based scanline polygon shader. PSX2 probably
uses a z-buffer-like hardware shader. Both algorithms are derivatives
of the z-buffer paradyne. PRMan uses a first or second order
reflection and illumination model (however it does not use a high order
model like raytracing, or radiosity), while PSX2 probably uses a first
order model. With these two facts, it would be easy to surmise that the
PSX2 could shade the same number of polygons that appeared per frame in
"Toy Story" in real time.
It is important to consider this in light of a few more facts,
however. The significant contribution to image quality is not in the
shading algorithm itself but in the reflection and illumination model
used, and higher order models tend to take exponentially longer to
shade, while producing scalar improvements in image quality. Also, the
biggest consideration in this whole scenario is taken for granted.
That game developers will produce games with scenes as complex as toy
story, fully animated. Don't hold your breath. Were still trying to
figure out good algorithms for bi-pedal motion.
Bottom line, take it with a grain of salt, but it is 'possible.'
Michael Herf wrote:
>
> What class of performance would a system have to achieve to do real-time Toy
> Story?
A hardware based, z-buffer shading algorithm that supports shadowing,
some higher order reflection, and a reflection and illumination model
similar to the Phong model. This is actually not much of a stretch in
today's hardware, however, most game designers still tend to use the
Gouraud model because of its inherent speed and ease of implementation.
Also, support for on the order of 10,000 - 100,000 polygons per frame
(at 60 fps that is around 600,000 - 60,000,000) which is in the realm of
the PSX2 claims.
>
> They obviously intend to compare pure geometry performance, which isn't half
> the, uh, story, since shaders and lighting (I presume) take a majority of
> the time. I think that they are close in geometry -- 75 million polys per
> second translates to a little over 4/pixel at video res (and 60Hz), which is
> certainly in a Renderman-acceptable range for that resolution.
>
> Still, if Toy Story took several hours per frame on 1993 hardware, we're
> still way off, even after scaling down to video (10x) and scaling up to
> current CPUs (10x).
Were talking about software versus hardware shading think 100 - 1000x.
And my understanding is that frames of Toy Story rendered between 7 and
20 minutes.
>
> However, Renderman is general, and this gap could be shortened dramatically
> by making certain limitations in the real-time renderer. I believe that
> something visually similar to Toy Story could be done in real-time or
> near-real-time on next-generation hardware, if you believe the specs that
> are flying around. Obviously, lighting would be weaker, and some parts
> would not be as dynamic (and the system that animated this thing is assumed
> to be magic), but I think it could happen.
>
exactly.
Larry Gritz
>
>A hardware based, z-buffer shading algorithm that supports shadowing,
>some higher order reflection, and a reflection and illumination model
>similar to the Phong model. This is actually not much of a stretch in
>today's hardware, however, most game designers still tend to use the
>Gouraud model because of its inherent speed and ease of implementation.
>Also, support for on the order of 10,000 - 100,000 polygons per frame
>(at 60 fps that is around 600,000 - 60,000,000) which is in the realm of
>the PSX2 claims.
Ah, but here's the rub: a TS image is in the range of 10,000 - 100,000
*primitives*, not polygons. Primitives are such things as quadrics,
bicubic patches or trimmed NURBS. The actual polygon count, if you were
to tessellate those primitives, would be another factor of 10-100 higher.
>> Still, if Toy Story took several hours per frame on 1993 hardware, we're
>> still way off, even after scaling down to video (10x) and scaling up to
>> current CPUs (10x).
>
>Were talking about software versus hardware shading think 100 - 1000x.
>And my understanding is that frames of Toy Story rendered between 7 and
>20 minutes.
No, they took several hours on average, and we would probably not
consider those images acceptable today (in terms of the quality and
complexity that we came to expect in ABL and the upcoming TS2).