How many Vertex and Polygon in current Real-Time 3D game?
Macbear Chen
Hi,
I am so interesting in RT3D programming...
I have a question about how many vertex and polygon used in current 3Dfx
game?
Like Quake II, TombRaiderII, MotoRacer, even some SimFlight game...
I am just curious about how many vertex and polygon used in current game?
Of course, it mean current Scenery it showing, not whole game.
--
麥克熊
E-mail: mac...@im.mgt.ncu.edu.tw
http://www.geocities.com/siliconvalley/vista/8177
TJ
Macbear Chen wrote:
> Hi,
> I am so interesting in RT3D programming...
> I have a question about how many vertex and polygon used in current 3Dfx
> game?
>
> Like Quake II, TombRaiderII, MotoRacer, even some SimFlight game...
Umm....
Best guess would have to be MILLIONS OF VERTICES and THOUSANDS AND THOUSANDS
OF POLYGONS!
I hope that's accurate! :^|
-TJ
Phil Frisbie, Jr.
Macbear Chen wrote:
>
> Hi,
> I am so interesting in RT3D programming...
> I have a question about how many vertex and polygon used in current 3Dfx
> game?
>
> Like Quake II, TombRaiderII, MotoRacer, even some SimFlight game...
>
> Of course, it mean current Scenery it showing, not whole game.
Ignore the other guy.
For current hardware, it would be good to keep the number of visible
triangles below 2000 or 3000. That includes surfaces, objects, and
effects (like multi-pass lighting, particles, sprites).
Phil Frisbie, Jr.
Hawk Software
http://www.hawksoft.com
Personal: http://www.geocities.com/SiliconValley/Lakes/7220/
Steve Baker
Phil Frisbie, Jr. wrote:
>
> Macbear Chen wrote:
> >
> > Hi,
> > I am so interesting in RT3D programming...
> > I have a question about how many vertex and polygon used in current 3Dfx
> > game?
> >
> > Like Quake II, TombRaiderII, MotoRacer, even some SimFlight game...
> >
> > I am just curious about how many vertex and polygon used in current game?
> > Of course, it mean current Scenery it showing, not whole game.
>
> Ignore the other guy.
Um, don't *totally* ignore him :-)
It kinda depends on what the questioner is interested in.
The *TOTAL* number of polygons and vertices in the entire game
could easily be in the 100,000 to 1,000,000 range. The number
that you see on the screen in any given frame is more like...
> For current hardware, it would be good to keep the number of visible
> triangles below 2000 or 3000. That includes surfaces, objects, and
> effects (like multi-pass lighting, particles, sprites).
...which one matters to you depends on whether you are someone
planning on designing the 3D models and levels for such a game,
or someone who has to tweak the rendering software to draw
such models in realtime.
--
Steve Baker 817-619-8776 (Vox/Vox-Mail)
Raytheon Systems Inc. 817-619-4028 (Fax)
2200 Arlington Downs Road SBa...@link.com (eMail)
Arlington, Texas. TX 76005-6171 SJBa...@airmail.net (Personal eMail)
http://www.hti.com http://web2.airmail.net/sjbaker1 (personal)
Ann Colby, New Orleans, LA:
"I sent out 100,000 emails for my product and received over 55
orders!"
(...and upset 99,945 other people)
TJ
Thanks for the insult Phil. *Appreciate it.*
Steve, thanks for replying realistically.
The last I heard, character models (I'm not sure about Quake II specifically)
in games like Turok, Unreal, Prey, and Forsaken (other 1st person shooters)
consist of thousands of polys individually. So for any given point in time
during that game, is it not correct to say that there will be anywhere from 2
to maybe a dozen enemies in the current view? Hmm, simple arithmetic says
that that could mean many thousands of polys, thus many times more vertices.
Am I right? If not, feel free to abuse me. :)
-TJ
Magnus "ShockMan" Blikstad
Quake/Quake2 model are at a max of 500-600 triangles, though the boss in quake2
is about 3000... donno about the other games. I dont really believe that the
current view only displays 2000-3000 triangles.... more like 15000-20000 maybe
10000. just a guess though. maybe you should mail the god in 3d games programing
(John Carmack - jo...@idsoftware.com)
-Magnus "ShockMan" Blikstad
Craig Garrett
This might give you an idea:
I had Quake2 (using a monster) record a log of gl commands. I isolated a
single frame in that log and took a look at it. It was in text format, one
line for each gl-call. There were 4000+ glcalls, most were glVertex and
binding textures to those vertecies. So there were roughly 2000 glVertex
calls in ONE FRAME of Quake2. Now, of course, it was running at about
30fps at the time, so you do the math, hehe, thats a lot of vertecies in
one second!
Craig Garrett
BTW, All of the polys in Quake are usually triangles or quads.
Michael K. Vance
TJ wrote:
> The last I heard, character models (I'm not sure about Quake II specifically)
> in games like Turok, Unreal, Prey, and Forsaken (other 1st person shooters)
> consist of thousands of polys individually. So for any given point in time
No. Character models usually consist of around 250 polygons each, sometimes as
high as 500. This is of course with games running on current consumer-level
hardware. Get a copy of jawmd2fx or md2view to look at individual character
models (or wireframes) in Quake2 to see what their construction is like.
> during that game, is it not correct to say that there will be anywhere from 2
> to maybe a dozen enemies in the current view? Hmm, simple arithmetic says
> that that could mean many thousands of polys, thus many times more vertices.
From various postings here and on OpenGL-Gamedev, it seems that most games right
now have about 2000 on screen, fully textured triangles at a time.
Games like Messiah promise run-time tesselation of surfaces to match your
hardware's capability.
--
| "Ours is a world of nuclear giants and ethical infants."
| - Gen. Omar Bradley
Jorn Gunster
> The last I heard, character models (I'm not sure about Quake II
specifically)
> in games like Turok, Unreal, Prey, and Forsaken (other 1st person
shooters)
> consist of thousands of polys individually.
To my knowledge a character model 'in game' contains about 500 - 1000 polys
(Q2 and Unreal).
The models for the cutscenes are much more detailed though.
Jorn Gunster
Phil Frisbie, Jr.
TJ wrote:
>
> Thanks for the insult Phil. *Appreciate it.*
I am sorry, I didnot mean to offend you. I should have made it clear I
was just needling you a bit because I interpreted the original post
differently than you and Steve did.
> Steve, thanks for replying realistically.
>
> The last I heard, character models (I'm not sure about Quake II specifically)
> in games like Turok, Unreal, Prey, and Forsaken (other 1st person shooters)
> consist of thousands of polys individually. So for any given point in time
> during that game, is it not correct to say that there will be anywhere from 2
> to maybe a dozen enemies in the current view? Hmm, simple arithmetic says
> that that could mean many thousands of polys, thus many times more vertices.
> Am I right? If not, feel free to abuse me. :)
My experience is with Quake and Quake2. Most Quake models are 200 to 400
triangles, but Quake2 does use more. I have not dumped the info for all
the Quake2 models yet, but the bosses may well be over 2,000 triangles.
Of course, backface culling removes about half the triangles from a
model. And bosses are usually by themselves, so the high triangle count
does not hurt so much.
Also, my recomendation about limiting a game to less than 3,000 polygons
is for AVERAGE 3D harware, and we all know most do not perform like even
a Voodoo1 card....
Andreas Jönsson
TJ wrote:
>
> Thanks for the insult Phil. *Appreciate it.*
>
> The last I heard, character models (I'm not sure about Quake II specifically)
> in games like Turok, Unreal, Prey, and Forsaken (other 1st person shooters)
> consist of thousands of polys individually.
This could be correct, but it is most likely that they don't render all
of these polys at all time. If an object is far away it would be quite
meaningless to render it with thousands of polys when it is sufficent
with hundreds.
> So for any given point in time
> during that game, is it not correct to say that there will be anywhere from 2
> to maybe a dozen enemies in the current view? Hmm, simple arithmetic says
> that that could mean many thousands of polys, thus many times more vertices.
> Am I right? If not, feel free to abuse me. :)
>
> -TJ
Actually the number of vertices is only slightly higher than the number
of polys, since most vertices are shared by several polys.
/
Andreas Jönsson
TJ
Ok. I'm truly sorry Phil. It's hard to interpret words on a screen as opposed
to actually conversing with someone. My fault.
Somehow I managed to poo-poo poly counts based on distant memories of maybe
not in-game character models, but more like pre-render/pre-import models from
the build/design phase (3D Studio MAX, Lightwave, etc.). Of course, they
usually lower the poly count to a hundred or so.
And I also forgot the shared-vertices, culled polys, etc. Basically I was
drunk when I wrote that post...
Just kidding!
A thousand apologies to all four of you who corrected me. I owe you free
web-hosting. Maybe. :^)
-TJ
Daniel Polli
For our current experience of game devellopper and for a target
consisting in a 166 with monster 3dfx.
we found since 9 month of dev that around 2000-3000 polygon (textured)
is a good choice (for actual hardware).
We found also that optimising the number of ply is not the harder job to do
also our current terrain contain 32767 ply but with use of bounding box
each frame see only 800 of it displayed (also after we add all ship and
other
stuff to reach the total number of about 3000 ply per frame)
The harder job seems to optimise the Pixel fill rate with a Z-buffer system
who fill even the ply behind another one (in the view ; outview ply are
"culled")
Macbear
Thanks all guys who give me the answer.
I thnik I have a basic concept to continue my RT3D research.
In fact, I have done some my 3D engine in sofeware-mode, but
perspective correction is just 1/3 speed than non-perspective correction
whcih I do, But depend on 3dfx voodoo, texture speed improve
so much. Even I can use linear-filter.
I think Virtual-Fighter 2 PC version is the suitable example for my
question.
Because the scene is fixed, not vary.
My question must divide in many part:
1: How many Vertex to compute for Matrix-Transform?
I guess 4000 is acceptable for current hardware.
2: How many Polygon to draw on screen?
Of course, it isn't include back-face polygon (culled).
But it should include screen-clip polygon.
It also include a polygon which overlap by other polygon.
In hardware Z-buffer, Should I use BSP tree to establish
my 3D environment?
Because in my software 3D engine, I just sort polygon Z value,
then draw the polygon. So I can't draw intersection-polygons correctly.
But In opengl Z buffer , I just need to draw polygon directly, even
ignore
the polygon sorting, it can draw the scene correctly.
I just want to get experience from current games...
In Heavy-Gear, I could see wire-frame directly....
In other games, may I see its wire-frame? A trick? A cheat?
Macbear
Steve Baker
TJ wrote:
>
> Thanks for the insult Phil. *Appreciate it.*
> Steve, thanks for replying realistically.
>
> The last I heard, character models (I'm not sure about Quake II specifically)
> in games like Turok, Unreal, Prey, and Forsaken (other 1st person shooters)
> during that game, is it not correct to say that there will be anywhere from 2
> to maybe a dozen enemies in the current view? Hmm, simple arithmetic says
> that that could mean many thousands of polys, thus many times more vertices.
Not so - those monsters are only going to be that detailed when you are
standing right next to them - in any decent game, they will be modelled
in
several different versions with differing polygon counts - then the most
appropriate version will be chosen based on range.
Hence, one of those monsters may be several hundred polygons (I doubt
'thousands')
if it's in the process of ripping you to shreds - but the others (which
are
presumably hanging around in the distance enjoying the show) will be
drawn with
*FAR* fewer polygons.
This is called "Range Based Level of Detail" (LOD for short) and is one
of the
most important algorithmic tricks for getting good frame rates in
complex
environments.
Covering up the switch from low to high detail is a problem. For very
detailed
models, the switch may be fairly invisible - but we serious flight
simulation
people use transparency to smoothly fade between the two versions of the
model to cover up the transition. This is called 'Fade LOD' (FLOD for
short).
It's also possible to 'morph' between the two levels of detail - but
there
are considerable algorithmic difficulties with doing this - and I don't
have
time to go into that here.
> Am I right? If not, feel free to abuse me. :)
Consider yourself abused (in the nicest possible way of course) :-)
Rick Weyrauch
TJ wrote:
>
> Thanks for the insult Phil. *Appreciate it.*
> Steve, thanks for replying realistically.
>
> The last I heard, character models (I'm not sure about Quake II specifically)
> in games like Turok, Unreal, Prey, and Forsaken (other 1st person shooters)
> consist of thousands of polys individually. So for any given point in time
> during that game, is it not correct to say that there will be anywhere from 2
> to maybe a dozen enemies in the current view? Hmm, simple arithmetic says
> that that could mean many thousands of polys, thus many times more vertices.
>
The monster models in Quake2 typically have between 600 - 1200
polygons. The models file (.md2) do not have any LOD switching built
into them and I don't believe the Quake2 engine creates LODs when
loading the models.
Certainly these types of games could benefit by using some of the tricks
that have been developed in the vis-sim world.
Rick
--
Rick Weyrauch voice: (972) 960-2301
Paradigm Simulation Inc. fax: (972) 960-2303
14900 Landmark Blvd., Suite 400 rwey...@paradigmsim.com
Dallas TX 75240 www.paradigmsim.com
Patrick Madden
Rick Weyrauch wrote:
> The monster models in Quake2 typically have between 600 - 1200
> polygons. The models file (.md2) do not have any LOD switching built
> into them and I don't believe the Quake2 engine creates LODs when
> loading the models.
>
Another message talked about having around 2-3k vertices per Q2
frame, so.... Do the monsters *really* have 600-1200 polys
per frame, or is this the total number per monster where there
are a number of different "frames" of animation.
Our animator built a somewhat stylized avatar that contains
many frames, but only 25 polys per frame. The Quake
monsters are more detailed, but they don't seem to be *that*
much more complex.
This matters, as you only need to xform, render, etc., the
current frame of animation--having lots of different poses
for the monster doesn't cut into your frame rate (ignoring
cache hit issues, of course).
Patrick
--
Patrick H. Madden, technology director, IKM Interactive
p...@ikm.com http://www.ikm.com
Michael K. Vance
Patrick Madden wrote:
> Another message talked about having around 2-3k vertices per Q2
> frame, so.... Do the monsters *really* have 600-1200 polys
> per frame, or is this the total number per monster where there
> are a number of different "frames" of animation.
There are approximately 200-500 vertices per model in Quake II. These vertices
are of course positioned differently according to whichever frame of animation
is current. Interpolation is used to smooth transitions.
> many frames, but only 25 polys per frame. The Quake
> monsters are more detailed, but they don't seem to be *that*
I can't imagine _anything_ looking very good with 25 polys, especially a human
figure.
Get a copy of md2view and you can look at each individual Quake2 model, with
different skins, turn off interpolation, change animation speed, etc. Jawmd2fx
works well for seeing the wireframes.
Rick Weyrauch
Patrick Madden wrote:
>
> Rick Weyrauch wrote:
> > The monster models in Quake2 typically have between 600 - 1200
> > polygons. The models file (.md2) do not have any LOD switching built
> > into them and I don't believe the Quake2 engine creates LODs when
> > loading the models.
> >
>
> frame, so.... Do the monsters *really* have 600-1200 polys
> per frame, or is this the total number per monster where there
> are a number of different "frames" of animation.
There are roughly 600 tris (and approx 900 verts) in _each_ animation
frame of the Quake2 monsters. The bosses have a bit more.
A few specifics:
gunner: 610 triangles and 978 verts
bitch: 652 tris and 998 verts
gladiator: 628 tris and 968 verts
>
> Our animator built a somewhat stylized avatar that contains
> many frames, but only 25 polys per frame. The Quake
> monsters are more detailed, but they don't seem to be *that*
> much more complex.
>
> This matters, as you only need to xform, render, etc., the
> current frame of animation--having lots of different poses
> for the monster doesn't cut into your frame rate (ignoring
> cache hit issues, of course).
>
The triangles in the models are all rendered as OpenGL triangle-strips
and triangle-fans, reducing the number of verts that must be xformed
each frame.
As a side-note, I've written a SGI Performer2.2 loader for Quake2 models
(md2 files) and levels (bsp files). If any you Performer/Quake2 geeks
are interested I'll post them.
> Patrick
> --
> Patrick H. Madden, technology director, IKM Interactive
> p...@ikm.com http://www.ikm.com
--
Rob Rodgers
Patrick Madden <p...@ikm.com> wrote:
>Another message talked about having around 2-3k vertices per Q2
>frame, so.... Do the monsters *really* have 600-1200 polys
God no.