#OpenGL Questions
29 messages in this thread
Just thought I would drop a note concerning all this tech stuff since I've
doing a lot of research lately for the same reason: MAX and World Builder
1. Did a demo of the new SGI Indigo2 Extreme last week with the brand new video
subsystem. It knocked my socks off. SGI has another video subsystem in the
works due out November and its 2-3 times faster than the one I saw. The basic
box starts at $35,000.
2. Talked to Intergraph also and was inpressed with the GLZ video subsystem.
They are going to do a demo for me later.
3. The 3D Demon GLint board has some good spec for those on a tighter budget
plus you can buy the board separately.
4. Got info on the AccelGraphics board. They advertise it as faster than the
GLint stuff at 250K triangles/sec versus 150K for the GLint.
Basically I came to this conclusion:
If you want to go to 3D MAX the Mips processor is probably the ultimate since
it has better floating point performance. Also the 10000 series processor will
be out next year. Cost is astronomical.
Windows NT on SGI? You could run 3D MAX and Alias on one machine. Hmm…..
The Alpha is next. 300Mhz due shortly and I can get a 275Mhz basic box right
now for around $5000. Probably the best overall solution but the video (OpenGL)
card is the unknown. What going to be available? GLint only? We will see…
Another solution is Intergraph with their workstations. They have the GREAT
OpenGL cards ready now. Again, cost is high. Dual pentiums and P6s,
Alphas…$24000 for dual Pentium 100s
Finally, do the Build-It-Yourself thing which is what I do. Pick and choose
conponents you want, etc… Dec Alpha 300Mhz, Accel OpenGL card, Micropolis
SCSI-3 9 GIG, 128 MB RAM, … Around $12000. Much cheaper this way.
Again, the OpenGL card is the unknown and that is the key. The new dual
pipeline video streams for the SGI really makes it scream now. Too bad
Intergraph won't sell their OpenGL cards separately since that would be the
answer. Build-It-Yourself with an Alpha and an Intergraph card. Oh well…
>>They advertise it as faster than the GLint stuff at 250K triangles/sec versus
150K for the GLint.
Have you looked into/heard anything about Lockheed Martin's Real3D board? I
want to say I remember hearing a quote of 750k textured/shaded polys per sec.,
but I can't remember for sure.
John Stetzer
JWS
Haven't seen that card. Got a number? I'll call and get info for everyone.
BTW, do you have an idea how much this is going to speed up rendering?
I tired of waiting 3-30 hours for scenes to render at SVGA/PAR resolutions.
I thought the SGI was suppose to be fast until some of the people here told me
no.
I've got a P100/32MB/ATI 4MB. Some frames take 15 minutes to render.
Thanks
>> Got a number?
Not off the top of my head; maybe someone will pop up w/it. I believe I
remember seeing their ads for their boards in some isssues of CGW and for
employment oppertunities in some game magazines like Next Generation.
>>I've got a P100/32MB/ATI 4MB. Some frames take 15 minutes to render.
My understanding is that none of these options will help in the actual
generation of the final image; they are tools ment to speed up the development
of the 3d scene used to generate the final image/animation. These boards allow
you to see your scene textured/lighted/etc., which would aid in speeding up the
creation of your scene, with final rendering being done via the processor. The
output of these cards would not be of good enough quality to be used as final
output. Just my understanding of these things, I could very well be wrong.
John Stetzer
JWS
Your right in that none of those option affect the rendering… except the
processor.
In the case of the OpenGL cards, the rendering is offloaded to the card itself
somewhat. So its a case of the processor and the video subsystem under Win NT.
3DSr4 no, its processor only.
The tangent of this string is going to help get a lot of issues out on the
table as we all face system choices. Mark is certainly right about SGIs not
being that fast, unless harnessed to a Reality Engine. I remember the REngine
as being over $200G. However, the other day I saw advertized a new SGI maching
and Reality Engine2 for 92G. I don't know the specs and realize that is a lot
of money, but if that system delivers the kind of performance the previous
REngine did then it clearly demonstratres SGI being very destablilized by
NT/multi-chip configs running 3D software. It really does seem that
NT/multi-chips are coming on strong. I understand that an upgrade for NT is due
1st Qtr 96 and is addressing (so to speak) driver issues. Regarding the 3-D
accell hardware – I understood that they can not only assist in preview speed
but also lend a hand in final render. Someone out there have the info on that?
Does anyone know if P6 is going to require software to be specifically
re-written, to some extent, to take *FULL* advantage of it's new architecture?
The QUAD-P6 might be the winning ticket. I will really be interested in seeing
Benchmarks of the Quad-P6 against ALPHA multi-325. For *not all THAT much
money* we will be sitting on some serious compute power. When I think of all
the time spent previewing anim sequences, effects and scene/model edits, let
alone all the added flexibility of features/ routines that can be incorporated
under NT – oh, lordy… if all this works I will be a happy camper <|;-)
Jeffrey
Jeff, I caught a lot of flak over the SGI thing. Let me give you the story:
I went to a SGI demo of the Indigo2 Extreme with the new video subsystem SGI
developed. It knocked my socks off. They did 30fps in realtime with shading and
rendering. An additional new video system is due in November that promises 2-3
times the speed of the system I saw. Egad! By using dual video pipelines they
increased the speed dramatically. The power of an Onyx on an Indigo. What that
trully represented was a tremendous drop in price. From a $100000+ Onyx to a
$35000 Indigo. Its still too high!
Now the $64000 question? Will it render any faster to Beta or SVHS? I don't
know. The powers in the know like Metrolight, say no. I will be doing a demo
here in Dallas at Paradigm next week so I will let you know.
On to the OpenGL Thing. As I understand it, rendering is offloaded somewhat to
the OpenGL system on the videocard, however the processor also does make a
difference. The SGI people were talking that their limitation now was the
processor and not the video subsystem. They are waiting for the Mips 10000
series chip next year.
Mearwhile back at the MAX, I have talked to a variety of hardware people about
Win NT stuff. I am able to buy a DEC Alpha bare box for $5000 and figure I can
build a whopper Alpha 275 Mhz setup for around $12000. Much better than the
$23000 Intergraph quoted me for dual Pentium 100s.
Right now the great unknown is the OpenGL cards. Which one is going to be the
fastest? Probably the Intergraph card set but they won't sell it separate.
Stay tuned…
Mark,
>> The powers in the know like Metrolight, say no. I will be doing a demo here
in Dallas at Paradigm next week so I will let you know. <<
depends on what quality you need. I asked others at work and they had heard of
the boardset and said yes, it's faster than the reality engine, but no, it's
NOT suitable for final renders and will only help your UI for testing up to,
but not including, the final colour and lighting stages if you need Renderman
or raytraced type quality.
Like, I said, if you're not doing broadcast stuff it might work for you. If you
want quality like the reality engine flight simulator then you might be in
business.
Hardware accelarators for use with MAX might provide some of this
functionality, but probably aren't quite as fast for a small fraction of the
cost.
Anyway, only you can decide at the demo if the quality will serve your needs.
Thanks for the help as I can see that I still have a lot to learn. At least I
can talk about this stuff without putting my foot (both feet?) in my mouth.
Question: We know that 3DS cannot do raytracing (except for shadows) in its
present form without a add-on module yet it produces good results. However it
must not be THAT good since only recently has it been used in films. Its been a
video tool. No raytracer.
Given that I have no knowledge of the SGI hardware architecture (Duh! does it
show?) I assume from your letters and my reading that the OpenGL hardware is
designed to produce fast realtime rendering by offloading the rendering to the
OpenGL hardware. This allows the processor to devote clock cycles to "normal"
duties such as disk i/o and such. Now while the OpenGL rendering is great for
games and flight simulation, its not good enought for final rendering at film
quality (which you must be doing) or even D1 or Beta video since you need the
raytracer.
For production quality work then the raytracying engine must be involked and
rendering time then goes through the roof. This new SGI video card set uses
dual data pipelines which has helped the data throughput substantially. And the
next video card set is going to be better blah blah blah. But they have
improved the OpenGL system which doesn't address the raytracing rendering which
is a function of the CPU only. This is why the SGI people are waiting for the
Mips 10000 series processor I assume.
So what your saying is even though the realtime output is good, we're not
viewing final output in realtime. The raytracing engine therefore is accessing
only the processor and not the OpenGL subsystem. So for the moment OpenGL is a
Phong or Gouraud shading engine only.
This must be the reason people are more concerned with packing 4 or even 6
Pentiums or P6s, or even multiple Alpha chips, in a box and not so much on the
video subsystem.
But since people are using 3DS with the Phong shader to produce commercials,
isn't the OpenGL engine good enough for Phong shaded commercials also?
The output I need is only S-video, not even Beta. I'm just producing VHS tapes
and CD-ROMs with Authorware. I trying to get that First Job (outside the
office) and will be buying Beta SP equipment in the near future (hitting the
auctions right now).
Waiting for your opinions…
>>it must not be THAT good since only recently has it been used in films.
>>Its been a video tool.
I'd say that 3DS's lack of presence in films to date, is due much more (if not
exclusively) to antiPC-prejudices in the film industry, rather than to a
perceived lack of (rendering) quality.
Dave
David,
>> I'd say that 3DS's lack of presence in films to date, is due much more (if
not exclusively) to antiPC-prejudices in the film industry, rather than to a
perceived lack of (rendering) quality. <<
I think that's just part of it – there are other issues. I know Frank Foster at
Sony Imageworks has had a long hard battle against the status quo. I've worked
with him on pre-visualisation (2 films) and seen him go from end titles
(monochrome – a very embarrassing use of 3ds, IMHO), to actual full screen
depictions of VR (the very cool scenes in "Johnny Mnemonic") . Sony's next
hurdle is to be allowed to do CGI composites with live film plates.
We've had it a lot easier at MetroLight where there's less politics to deal
with and actually been allowed to do the aforementioned type of CGI/live
composite shot. A lot of the credit for making this happen needs to go to Brett
Leonard who seems to be the king of stretching his CGI-dollars. He's a big fan
of 3ds. Despite this there's now a perception that 3ds is only good for VR-like
sequences in film and for photo-real work you have to use Renderman. This one's
a little more difficult to argue since Renderman has that cool shader language
and the talent base to use it.
There's a couple of other problems which may be helped with the coming of
NT/MAX and was partially alleviated with the intro of the 3ds SGI renderer.
Film CGI requires a ton of hardware generally and 3ds requires tons of RAM in
particular. I'm talking more RAM than you can put on many PC motherboards –
let's say 512mb+ would be a good starting point for complex scenes at 2k
resolution. For example, having 16 Pentiums with 128mb RAM did us no good when
layers of scenes even at D1-rez required 400mb to render without paging – we
were saved by the fact we could use the SGI ChallengeXL which shares 1gb RAM
between 8 cpu's. With NT and MAX we should be able to get multi-proc render
servers which share a large pool of RAM, say 1gb for starters. Furthermore,
they should be equal or faster than the SGI multi-proc systems for a fraction
of the price. Great for high-end video and film work.
The final, and biggest, problem is one I've mentioned before. It's the talent
pool situation. There just aren't that many people with the experience and
training in 3ds to do the kind of CGI that film audiences expect, whereas on
the SGI side there are quite a lot of talented artists with 5+ years
experience. Hey, I'm barely keeping my head above water with the challenges
that come my way and I've been using 3ds full-time since early r1. It's the
artists that drive the industry as much as anything. Encouragingly, I've heard
a lot of positive interest in MAX from the SGI artists. Maybe some will defect
….<g>
Thanks for the info, Martin. Yikes! Those RAM requirements are pretty
heady! Don't worry about keeping your head above water — we've all got faith
in you!
Dave
David,
>> Those RAM requirements are pretty heady! <<
That's right. That's why it becomes prohibitive to have single processor
systems when multi-processors can share a RAM pool efficiently with the right
architecture (large, fast caches for each CPU and a fast 1.2gb SEC buses etc).
Mark, I'll gaurantee you that the fact that any film did not use 3DS for it's
effects had NOTHING to do with it's lack of raytracer. 99%+ of 3D feature film
work is NOT raytraced.
Unless you are talking about something like a flight simulator, or UI previews,
hardware-assisted rendering typically has NOTHING to do with final output,
period.
You really need to get an understanding of what is used for what before letting
these SGI hype-guys dazzle you with a lot of bells and whistles. If they were
responsible people, they'd offer this information to you instaed of letting you
find out later. All they are interested in is moving boxes, not helping you get
the most for your dollar.
-Alan
Like I've said before, I'm the rookie here and just trying to learn (A year ago
I didn't know whut a computer grafix artist wuz, now I are one…) type of
thing.
Plus I don't have access (yet) to the high-end stuff nor the experience of all
of you have so please be patient and keep correcting me.
Meanwhile, what I understand is:
1. Renderman is used for final output, but its not a raytracer.
2. 3DS is good but Renderman is better.
3. Renderman dosen't use OpenGL to help with the rendering.
4. MAX may or may not use OpenGL assist for final rendering. We don't know yet.
5. OpenGL probably is useful only for realtime applications such as games.
6. SGI is good, but the dual pipeline video streams may be a lot of hype.
7. RAM is your friend in 3DS. 64MB or 128 MB ain't gonna cut it.
Thanks for the help and for setting me straight Alan. I'm waiting for more
words of wisdom.
Bye.
Sounds like like you've got a great handle on this stuff, Mark.
Sheesh, you sounded so knowledgable, I thought you were a veteran of this
industry that was swallowing a little too much SGI hype, so sorry about
sounding like you should have known better.
Like MartinF has said, if the quality of the hardware rendering is acceptable
for an application, then so be it. But the overwhelming majority of finished
production 3D work is (at least) phong-rendered, and takes MUCH longer, no
matter what you are using. Renderman is the best, but other worthy renderers
such as Wavefront's and 3DS's are used, too.
Re RAM requirements, I'd suspect that 128mb will be needed for medium-sized
jobs with MAX, more for the larger ones.
Hey, we're all on the same learning curve, here, Mark. I'm learning every day
just like everyone else here. And this forum is a great place to keep learning!
Have fun,
-Alan
Thats all right, I probably should have known better. But I hadn't even seen an
SGI before the demo and got excited. My buddy about had an orgasm when I told
him about the SGI. So you can see where we come from… the wrong side of the
tracks. Plus there's all the hype about SGIs from the movie industry. Its like
"Elvis is in the building…"
I guess it showed me what is possible…maybe in the near future. If the Mips
10000 series processor is what Mips is saying its going to be then real time
surfaces could possibly be done… and as final output.
I think my trouble right now is knowing exactly what is acceptable as final
output. I'm happy with the Phong and metal shading I'm getting from 3DS. But
then I don't know better. I've got Truespace also but the 3DS stuff looks
better. I would like to see some Renderman output just to know.
In some ways I'm still at the elementary stage. Like I can't get my shiny gold
objects exactly the way I want them to look. On the other hand the rotoscoping
I've been doing with perspective matching has turned out better than expected.
I wowed the office.
BTW, I can get Sun Sparc stations for $2K. Model 20s I think. I'll find out
Sunday. My cousin is a big Unix person. I think those would be a good render
station on a network. She has the specs on all the Suns and will give then to
me on Sunday. What do you think? If they are 133Mhz machines I don't think they
are worth it. I can get 275Mhz Alphas for 5K.
Thanks for the support.
Hi Mark,
Hey, I've had some <Whoa!-now-I-need-a-cigarette> moments myself from checking
out the workstation-class stuff, too. All we are trying to point out is exactly
what all classes of machines are capable of in terms of quality and speed and
how accepted their output is in the industry. Once these issues become
apparent, one can look at the price levels and decide which is right for their
application.
It'll be quite interesting to see just how powerful the new Mips CPUs will be,
and how they compare power- and price-wise to the P6's and Alphas.
The term "acceptable output" can be defined multiple ways depending on who the
client is, and what they are expecting. There are many examples of different
levels of photorealism and detail already documented in movies, TV, and video
games. One only needs to do a little research to find an example of what you
are looking for. Then find out what you need to do it.
For some excellent examples of top-end Renderman output, rent some laserdiscs
like Jurassic Park, or watch Toy Story very closely this fall. The Listerine
commercials are another good example and there are dozens of others. Good Rman
animators can be spotted by the very worn copies of The Renderman Companion on
their desks and that certain pallor of a dedicated programmer. One of my wishes
for a MAX CoreComp is the ability to do a little object-oriented
shader-building to give us some of the flexibility of Rman, but more
user-freindly.
The bottom line is really the talent and dedication of the artist more than
anything else. I've seen wonderful stuff from very cheap equipment and horrible
stuff from workstations and everything in between. The best bet is to choose a
platform, learn as much as possible about it, and work your butt off to be the
best you can at it.
Best of luck to you,
-Alan
Mark,
I probably shouldn't have mentioned raytracing since that has caused a little
confusion. Renderman is the high-end renderer of choice in the high-end
video/film industry. IMHO, 3ds comes pretty close in quality (although it
doesn't have the shader language) and is a lot faster to boot.
>> However it must not be THAT good since only recently has it been used in
films. Its been a video tool. No raytracer. <<
In high-end broadcast and film raytracing is avoided completely _unless_
accurate reflections/refractions. It's better not to use it if you can because
of the speed – it's slower than cold molasses <g>. In fact, if a client wanted
raytracing MetroLight would charge a large premium for that. Much of the effect
of raytracing can be faked using a great quality scan-line renderer like
Renderman or 3ds. They have shadow-mapping and lots of different reflection
mapping types (flat, cubic static, cubic automatic, spherical) and I doubt all
of those features are supported in hardware. I suspect that renderman receives
no rendering help from OPENGL hardware accelerators.
To clarify: I was saying that the boardset you described won't help either
high-end renderers nor raytracing (which you should avoid anyway unless you
need accurate reflections/refractions).
So, other than what I've clarified about raytracing, I agree with your points.
So what we might be able to say at this point is:
1. Renderman is better than the 3DS render engine. (Is it available as a 3DS
plugin?)
2. Renderman doesn't use OpenGL to assist in rendering. So the processor become
paramount in this case.
3. MAX may or may not use OpenGL to assist in rendering. We don't know. (At
least I don't).
4. OpenGL probably is useful mostly for realtime items such as games.
5. You can never have too much RAM for 3DS.
6. And finally, someone will probably write a OpenGL render engine for 3DS and
make this whole conversation academic .
Thanks for the help. Later.
Mark,
this is where it gets complex <G>.
>> 1. Renderman is better than the 3DS render engine. (Is it available as a 3DS
plugin?) <<
that seems to be the concensus in the high-end video/film industry BUT only in
the right hands. Renderman is less user friendly and takes longer to master –
programming skills pretty much required to get the most of it.
No, it's not available for 3ds. There may be some standalone converters out
there but many things are not convertible. I'm hoping for RIB output in MAX but
I doubt it will happen.
>> 2. Renderman doesn't use OpenGL to assist in rendering. So the processor
become paramount in this case. <<
correct.
>> 3. MAX may or may not use OpenGL to assist in rendering. We don't know. (At
least I don't). <<
apparently not. It uses HDI (Hoops Device Interface?) which is supposed to be
faster and neater but less extensive than OPENGL according to MetroLight's
software developers.
>> 4. OpenGL probably is useful mostly for realtime items such as games. <<
possibly, and for hardware shading of models during early materials tests etc,
I think.
>> 5. You can never have too much RAM for 3DS. <<
well, you should have enough to keep the entire scene in RAM. More than that
will gain you nothing, less will slow rendering.
>> 6. And finally, someone will probably write a OpenGL render engine for 3DS
and make this whole conversation academic . <<
I don't know. I kind of doubt it. The board makers may chose to support HDI as
well as OPENGL.
So what did you catch FLAK for? So far, this has all looked like a good
exchange of very cogent info + opinions. Jeffrey
While I have been working with computers for 20 years back to the dinosaur days
of punch cards on a Univac, I have only been doing 3D Studio for a year. I
still got lots to learn.
Martin Foster of MetroLight has been kind enough to help me with some of the
technical aspects. He has been working with SGIs and other things so his
knowledge has proven invaluable. I'm a
network/database/Powerbuilder/conventional programming person who stumbled onto
this animation stuff by accident and now I can't put it down!
Anyway, he has explained the hardware architecture and the rendering engines to
me a lot better. At least I don't put my foot (both feet?) in my mouth now.
Later…
<<computers for 20 years back to the dinosaur days>> Funny! I was an artist on
a science team for a while back 1970. The science guys were looking at how big
pre-historic birds were able to fly. I was doing all kinds of drawings and they
were in-putting data into computers and I believe they were on FORTRAN. I kept
thinking to myself "god, this is crazy, I'll never go near computers." Oh well,
put my foot in my mouth on that one! Jeffrey
>>Real 3D board
Went hunting through some of my old CGW mags. and found a # for the Real 3D
board people (800-393-7730), and for the person on the other end of the line I
got a Web site (www.mmc.com/real3d). Haven't had the chance to check it out
yet. According to the person I spole with, the board they have right now is for
developers; she expects vendors to start shipping retail boards sometime in Q2
'96. Info FWIW.
John Stetzer
JWS
Mark,
Your data makes it look like the Alpha is still the most affordable
system for the bucks. How much would MAX suffer (windows NT speedwise) if you
don't run an accelerator board? Or do you have to use one to get MAX to
perform?
Is Gary Y. running an accelerator board?
Sanford
Thats a good question that I hope the Autodesk people will help us find out
(y'all jus jump here anytime).
The knowledgeable people tell me that Renderman doesn't use OpenGL to help
acceleration. So assuming (a dangerous thing) that the MAX render engine (Hoops
thing) also is not OpenGL dependent, then the processor only will determine
final render speed.
It appears the Renderman is the standard against what everything else is
compared to. I sense that MAX is aiming toward the high end to compete with
Alias and the other big packages. So would I think the MAX render engine would
be more like Renderman and not offload any rendering to the OpenGL engine.
Alan Iglesias and Martin Foster think the OpenGL subsystem is more geared
toward realtime applications such as simulations and games.
So while I got excited and the SGI demo I went to where they rendered in
realtime, it must not have been final output but only preview. However, SGI is
putting and lot of effort into improving the video subsystems. This must be for
the simulation customers.
I think the answer to your question is this: If you are running rendering on a
network, then the slave machines can use standard video cards. The modeling and
master machines need a OpenGL card to take advantage of the render preview
modes and the acceleration available under Win NT.
However the other thing to think about is what price are OpenGL cards gonna be?
If they are low then you might as well put then in all your machines that run
Win NT. I think this is going to happen. I'm already getting a lot of vendors
calling and sending me info on OpenGL and GLint cards.
Later…
Mark,
Thanks. Yes, Renderman is the high-end renderer of choice. I would
hope that MAX comes up with some "shader" type of renderer like Renderman so
you can write your own favorite shaders to get the look you want.
I expect to buy at least one of the OpenGL Glint cards for my main
workstation.
Sanford