Optimal polycount for rF1

Rantam

Well, first of all, consider this just as a suggestion if you're interested on creating a car for rF1. Today we have more powerful PCs than those available when rF1 was originally released, so you may increase these numbers without noticing a big impact on performance.

Anyway, the optimal polycount is around 15000-25000 (including wheels), for a high res version of the car, divided like follows:

  • 8000-10000 Body
  • 1000 cockpit outside view (and that's even too much)
  • 1000-1500 x 4 for the wheels. But it's better to use less polys if possible and use them in the body
  • Apart we have the body details and other elements like: wipers, mirrors, driver, etc.

Additionally the exterior of the car will need:

  • 200-500. Shadows model to generate the shadows
  • 50. Collision model

(You have an example of how these look in the following screenshot of a Mercedes 300SLR Coupe WIP)



And we have also the Hi-Res cockpit (using on cockpit view). You may use 4000-5000 polys on this one, but this number is not so important. In some of the cars we're creating we have very high detailed cockpits with around 20000 polys, and rF can move them smoothly without problem.

The golden rule in any case is trying to optimize the polycount as much as you can. Keep in mind that a lot of players still having old computers running rF1.

Hope this will be helpful (Topic originally discussed here)

;)
 
Last edited:
Use the vehicle viewing utility to show how many polys are being rendered. Use it to generate LOD meshes so that poly count ranges from 35000 at extreme close up down to ~ 2000 for long distance ( > 250m ). The viewer will allow you to work out where the LOD meshes should switch to give seamless viewing as the distance changes.
 
Something to keep in mind is that, in general, with today's latest cards, it's materials, not geo, that can cause the biggest bottleneck in rendering. That's not to say you can model million polygon cars (the numbers above sound reasonable). However, in general you can push, say, a million polygons with one material faster than you can push a million with, say, ten materials.

Let's say you had a car with 20 materials--each material had a unique texture. If you pack, or "atlas" those 20 textures into fewer, larger textures (2048x2048's for example) so that you only need 5 materials, your geometry will render faster.

Of course it's a good idea to keep your polygon usage under control, but it's a REALLY good idea to keep your material counts under control, too :)

Edit: Are you listening, track builders? :p
 
Problem for track builders is that the .tdf is linked to materials, so for detailed track bumps you need to 'waste' materials for the purpose of creating those bumps. (note: if you like your bump's with the tdf, i don't :p)

Edit from Scott -- well, not even bumps, but varying grip levels (like painted concrete for example). And this is correct, but you can still be modest in your material counts if you plan ahead :)

Edit from GT: True point, I assume a simular system will be used with rF2. Also, you can just reply to threads :p
 
Last edited:
Scott Juliano said:
Something to keep in mind is that, in general, with today's latest cards, it's materials, not geo, that can cause the biggest bottleneck in rendering. That's not to say you can model million polygon cars (the numbers above sound reasonable). However, in general you can push, say, a million polygons with one material faster than you can push a million with, say, ten materials.

Let's say you had a car with 20 materials--each material had a unique texture. If you pack, or "atlas" those 20 textures into fewer, larger textures (2048x2048's for example) so that you only need 5 materials, your geometry will render faster.

Of course it's a good idea to keep your polygon usage under control, but it's a REALLY good idea to keep your material counts under control, too :)

Edit: Are you listening, track builders? :p
Click to expand...

Thats interesting, I have always wondered if it was better to have lots of small textures or one large one.

What if you use a low number of textures but have the same textures in a large number of materials (to get different shaders on different components) rather than keep the materials numbers low (share shaders across different components) does that reduce or speed up render time, is it the material count or the texture count that is most costly?
 
Packing all smaller textures into one giant texture with different materials using that large texture, you'll still end up having all the various materials, so it makes no difference.
This means you'll need to think about materials as soon as you start mapping, because you'll want to group the polys with the same material properties on the least amount of textures, as much as you can, in order to keep the material count low.
 
Great thread.:cool:


I try to keep my poly count as low as possible and Rantam's example is a pretty good guide (although as of late I have been increasing the poly count on our models).

I do our cockpits a bit different because I only use one model. I put all the detail into the cockpit but use a draw distance of about 3 or 5 so all of the high poly parts like gauges, switches and shifters turn off at a very short distance.

I also try to only use 3 or 4 large textures per car.

Body 2048X2048
Cockpit 2048X2048
Wheels 512X512 or 1024X1024
Suspension 1024X1024


We have always gotten compliments on how FPS friendly our mods are and I attribute it to using large textures.

So I can attest to Luc Van Camp's post about using fewer but large textures.

Bob
 
Well, a lot of interesting things to read have been written here. Despite i was going to leave the texture thing for another thread (just to have a clear topic for this one), thanks guys:

@Scott Juliano for the valuable information about keeping number of materials as low as possible (i always wondered about that).
@Bob for that summary of how Team Players organize the textures. I'm glad to see we at HistorX try to do it that way with our new cars:)
@Major_catastrophe. Believe it or not I've never used the car viewer. I always check the cars on 3DSimEd or in game.

Thanks indeed for sharing such valuable information! ;)

Best regards
 
Once I tried a F1 Mod which each car has 40.000 polys. With a overclock i7@4.0Ghz + GTX 480 I got slow downs below 20 FPS. I was wondering what a Mod like that is about. No use at all.
 
Being a great modeler does not mean you will be great at doing mods. Creating game content is an art that goes beyond just making high poly models.

Being a good modeler for game content is knowing how to conserve polys where you can and using textures to make it appear as though you are creating high poly models.

Knowing how to balance 3d with 2d is really an art that a lot of very good 3d modelers don't understand.

That is the reason Rantam created this thread. If more guys understand what makes a mod achieve good FPS the more people will enjoy their mods.
 
True... no fun by just look a amazing detailed 3d model at showroom in be unable to racing with it.
 
Bob (Punko353) said:
...Being a good modeler for game content is knowing how to conserve polys where you can and using textures to make it appear as though you are creating high poly models...
Click to expand...

That's the Rosetta stone of game content creation, thanks for summarizing it so well Bob! Keep the polycount as low as possible, and cheat the eye with proper textures to make it look more detailed! ;)

Regards
 
How about Bump map v's Seams?
To get a bump map to look good without any pixelation or jaggered edges, it needs to be pretty big. I have tried and failed many ways to make door seams look good with bump mapping, even a 2048x2048 bump map still shows up pixelation on the seams.

If I add seams to the model instead, I will get a much better outcome without having to use a 1-5mb bump map. The current car I'm building is about 15-17k polys for lod A (including everything, wheels cockpit etc), by adding seams to the doors, hood and trunk it will gain about 1000 polys.

My question is, what would have more impact, a 1-5mb bump map or 1000 polys?
 
Well, there's no doubt you're going to get better results using seams in the 3D model. And gaining 1000 polys is not that bad. I also tried the bump map approach and reached the same conclusion: is not that good for that.

Personally i prefer to use bump maps only on these objects which need a good amount of detail (for example some parts of the cockpit), but for the door seams i'll always prefer the option of doing them on 3D (in fact many of our scratch made cars have 3D door seams).

Another option would be simply to paint them (as ISI and Simbin do, for example). The result is very good in most of cases and you'll save polys to be spent in more important places :)

Regards
 
You must log in or register to reply here.

Back
Top