How many polygons do you think PGR3 is pushing?

Hardknock

Veteran
Okay, this is nothing but baseless speculation on my part. But I find it pretty interesting.

Recently Bizarre revealed that their cars are made out of 80k polygons. And that they are shooting for 8(or more players).

Here's a pic of a modeled McLaren F1 that was released toda:
mf18006001fg.jpg


So taking this information into account, at the bare minimum:

80k * 8 = 640k

640k * 60fps(rumored) = 38 million 400k Polygons per second for just the cars!

Now we have to factor in the environments, which I'm guessing could push this over 60 million per second(I'm guessing environments would use a slightly lower number of polys than the cars.)

Am I calculating this right? Does that sound feasible? Compared to racing games of today, is that a pretty substantial leap? :)
 
One thing of note:

The cars won't have 80k visible polygons per car per frame -- the models may be made of ~80k polys but if the game is drawing that many polys that aren't seen then something is wrong :?

Say you got 20-30k visible polys per car (that's probably being a bit generous as most of the time you're just seeing the back/top of the car) and the calculations change a bit.
 
Guess that's when the xenos early z-pass will come in handy when you have alot of overlapping polys in a single frame.
 
Even with 80k polygons for the car models that is still impressive. There should be some type of LOD system put into place so the farther away a car is the less polygons for that model.
 
As long as Bizarre doesn't communicate the LOD method used in the game, it's pretty uselless to try to do any calculations, I think.
 
Half of that 80,000 number is for the car interior, and most likely the only time you will be seeing the interior in full detail is when the camera is inside the car, and that can only be for one car at a time obviously. When you are outside the car, the interior will probably be greatly reduced in detail so you only see things like the driver and a very basic (and darkened) representation of seats, the dash, etc.

Outside the car, the 40,000 exterior polygons will only be at maximum detail. As the car gets further away, I'm sure there will be some smart LOD system that will greatly reduce the number of polys in use.

And further, only about half of the current LOD polys will be facing the camera, and the other half are simply back-culled.
 
Sonic said:
Even with 80k polygons for the car models that is still impressive. There should be some type of LOD system put into place so the farther away a car is the less polygons for that model.

Especially for the interior, I can't see there being a whole lot of need for modeling the speedometer/odometer when you're in the heli-cam 100feet up.
 
Will a game like PGR3 really use sophisticated LOD?
I do not think so.

I am pretty sure there will be only one type of car modell + interior and maybe 1 or 2 low poly versions for helicopter shots or when you see your competitors very far away.

True LOD tends to create really ugly artefacts in realtime applications. Some people may know Sega Extreme Sports which had really cool Terrain LOD, which allowed huge/realisitic terrains, but let to ugly artefacts everytime detail increased/decreased.

Basicly you can´t use different LODs for the same camara shoot. You will get artefacts when you get near the object, unless your LOD is extremly smooth, which is (imo) unrealistic for current games like PGR3.

You need some sort of geometry mip mapping. For heightmaps (or their generalization: geometry maps) this is in principle very easy, but still leads to artefacts in praxis.
 
Recently Bizarre revealed that their cars are made out of 80k polygons. And that they are shooting for 8(or more players).

I'm pretty sure 12 car races has already been announced.
 
DotProduct said:
Will a game like PGR3 really use sophisticated LOD?
I do not think so.

I am pretty sure there will be only one type of car modell + interior and maybe 1 or 2 low poly versions for helicopter shots or when you see your competitors very far away.

True LOD tends to create really ugly artefacts in realtime applications. Some people may know Sega Extreme Sports which had really cool Terrain LOD, which allowed huge/realisitic terrains, but let to ugly artefacts everytime detail increased/decreased.

Basicly you can´t use different LODs for the same camara shoot. You will get artefacts when you get near the object, unless your LOD is extremly smooth, which is (imo) unrealistic for current games like PGR3.

You need some sort of geometry mip mapping. For heightmaps (or their generalization: geometry maps) this is in principle very easy, but still leads to artefacts in praxis.

I don't know why they wouldn't. You're talking about the ability to bump up the number of vehicles in the game by doing so. You could save at least 2 cars' worth of verts by using LOD. It's not like you could ever make out all the detail in all 8 cars at once, so I don't see why not. As you say, make the LOD smooth. This is a case for all games. But for racing games, so little of the car animates that it only makes sense. With he polygon power of these systems, I think it would be even easier than last gen since you wouldn't have to hack off as much for a lower-detail model. They could also use HOS models and dynamically tesselate, but that's probably not reasonable to expect in a launch game with a tight schedule. PEACE.
 
Well, let's say at the start of the race where all the cars are together... would there even be LOD used at this point? We could take this as the worst case scenario....
 
Well, let's say at the start of the race where all the cars are together... would there even be LOD used at this point? We could take this as the worst case scenario....

That is a very good argument against LOD in games, because it is always about steady 30 Frames in the worst case scenario.

And when there is the possibilty of 8/12 cars all together in front of the player, than the engine has to be able to handle it.

The same argument is valid for sophisticated occlusion culling, too. What is the benefit of sparing GPU power 90% of the time, when the player can stand on a hill and see everything?

Both LOD and ocusion culling only make sense, when you design the game/level itself so that such worst case scenarios can´t happen.

But from experience I would say that most exisiting outdoor games have neither sophisticated non trivial occlusion culling nor continues LOD.

One reason is that you often have to deal with worst cases where occlusion culling and LOD does not help.
 
Operation Flashpoint has a very tweakable terrain and model LOD system, which also includes whether trees or vehicles cast shadows, max distance for shadow rendering, desired frame-rate range, overall view distance (upto 5 kilometres)...

Jawed
 
on a related note, Bizarre mentioned that the Brooklyn Bridge structure in PGR3 contains as much polygons as an entire city in PGR2/1...
 
I still think PGR3 has a very "aggressive" LOD management engine in place.
Iron Tiger said:
Why don't they put some of that unified shader power to use on normal mapping the details so we won't see as much of the polygonal edges? It should also make the LOD engine more efficient, or at least produce better looking results.
http://media.xboxyde.com/gallery/public/1533/726_0007.jpg
That Viper isn't much farther from the "camera" than the Saleen, but the LOD is much lower on it.
 
There is an article on PGR3 in this months EDGE.

The cars are composed of 40k polygons for the interior and 40k for the exterior. Brooklyn Bridge is 600k polygons and Manhattan Bridge 1 million polygons.

Bizzare are aiming for 60fps. They also mention the draw distance for one of the unnamed cities is 3km.

For reference one of the Tokyo tracks in MSR took up just 90k polygons in its entirety.
 
Did'nt they say 40k for exterior and another 40k for the interior.
But do bear in mind that not all of those will be visable, i say 35k will be visable, and wont the enviroment be normal mapped???
thus saving polys.
 
Well let's see... this is an interesting question...

All figures are PER FRAME and not PER SECOND and figures do not include occusion culling.

*Vehicle Interiors > ~40,000 polygons (+- 5,000) per frame
*Vehicle Exteriors > ~40,000 polygons (+- 5,000) per frame
*Vehicle Driver > ~5,000 polygons (+- 1,000) per frame
*Average Spectator - ~2,000 polygons (+- 1,000) per frame
*Enviroment Average (including ambient objects) > ~2,000,000 polygons (+- 500,000) per frame
*Average number of spectators on screen > 250 (x2000 polygons each) = 500,000 polygons (+- 250,000) per frame
*Average number of vehicles on screen > 8 (times the total polygon count of vehicle interior/exterior/driver which is 85,000 (+- 11,000) X8) = 680,000 polygons (+- 88,000) per frame

TOTAL: 3,180,000 polygons PER FRAME(+- 838,000) or a total range of 2,342,000 to 4,018,000 polygons PER FRAME.
Extrapolated to POLYGONS PER SECOND: At 30FPS= PPS range of 70,260,000 to 120,540,000 polygons/second and at 60FPS= PPS range of 140,520,000 to 241,080,000 polygons/second.

Not bad for a first generation game!

Think of it like this... each CAR pushes over 4.8 million polygons per second (again assuming 60fps)... almost as many polygons as an entire GAME from this generation (for reference an average high end XBox game pushes roughly 150K-200K polygons/frame at 30fps or basically 4.5-6 million polygons/second and your average high end PS2 game pushes 100K-150K polygons/frame at 30fps or 3-4.5 million polygons/second). I would like to stress that this *IS* within what ATI/Microsoft indicated the polygon performance of the XBox360 system would attain in a actual game enviroment. Microsoft/ATI stated roughly 500,000,000 polygons/second of actual game performance, but based on stated efficiency claims I would estimate 90% of stated performance so a maximum of roughly 450,000,000 polygons/second of actual performance.
 
Last edited by a moderator:
Back
Top