Audi R18 2012 - Wip (no release)

mmm ... having to change cars detail just to play 1 mod is the best way to never play that mod, at least for lazy people as i am!

Or, you forget to lower the detail, you join a server, try 3 mins asking yourself why you have so low performances, you remenber you have to set lower details, exit from the server, change the setting, try to rejoin and the server is in race session!.

Seriously, an ideal situation, i think at least for the first 2 years of rF2 life, is a sort of standard LoDs polycount, taking as sample the ISI cars.

In short, my advice is to follow Luc guideline!

 

WOW!!!!!!!!!!!!!!!!!!



Fantastico!!!!!!!!!!!!!

 

Here's a little something to consider

Minimum number of times car needs to be rendered:

1. Into shadowmaps
There is a cascade of shadowmaps in rFactor. If well optimized - car will be rendered once or twice (if at edge between two shadowmaps)

2. Into reflection on the road, reflections in building windows, etc.
Once for each reflection.

3. Into reflection in other cars
Since reflection in other car is a cubemap, car will have to be rendered 2-4 times, depending on how many faces of the other car's cubemap it spans.
Multiply that by number of cars that need reflection of this car.

4. Into rearview mirrors
Once, or even 3 times if mirrors use separate views.

5. Onto screen
Once, or even multiple times, for multi monitor setups with separate views.


Now multiply all that by 200k triangles


Yes, if you disable reflections and shadows you can get smooth graphics with high-poly cars. So you can get high poly cars into rFactor engine, but not necessarily into rFactor 2 engine running at full details.


But polygon count is not all that matters. For every batch of polygons is rendered, there is a render command issued by game engine. Each such command also have it's cost and it's not a low cost. One of most important optimizations nowadays is to render as much polygons as possible with as little draw calls as possible. That's because GPU's are very powerfull and can't waste too much time waiting for CPU to give command (loosely speaking, but I don't want to go into details here).

For example - Crytek engine is capable of rendering one frame with just hundreds of commands passed from CPU to GPU.

Now think of how many GMT files you have for one car and how many objects with different materials each one of these GMT files has. You will need one command for each mesh. And of course issued more than once because of shadows and reflections.
You may end up with hundreds of draw commands just for one car.

And each of these draw calls can even take amout of time that GPU would take to render more than a thousand triangles. So you may have a car that is low-poly but still renders slow if you split it into too many little objects.

Defining good LOD's for car is not just about defining the same set of meshes with less triangles, but more importantly, about reducing number of meshes. You can remove those that become irrelevant (brake calipers, cockpit details), but you should also combine meshes with similar appearance and give them single material whenever possible.


And there are collisions. You need a good low-poly collision model, because collisions have pretty high per-tiangle cost. In worst case, time to detect a collision may be even dependent on the square of number of triangles. Costs of collision detection is something very often underestimated.


And finally - don't forget body deformations during collisions. These will be costly aswell, unless implemented on GPU.



That's a bit too much to describe in one post, but what I wrote above should give you a good idea of what you need to deal with.


That's why many game engines cheat nowadays on shadows and reflections. You either go for realism or you go for details.

 

That's a very informative post for a hobbyist like me K, thanks!

 

If I find the time, I'm planning on writting two documents.

One would be on GPU's in general - so people would know more about stuff like blending, z-buffering, transformation precision and performance hints (some of tracks I've seen in rFactor 1 were literally anti-performance designed )

The other would be about lighting (and shadows) - how it works in real world and how it's implemented in game engines like rFactor.

But of course, when I find the time...

 

I removed 50k tris only lowering wheels mesh

I've got a turbosmooth on all rims and a huge polys calipers eheh

Now 150k tris, going on with optimizing! Thanks all for replies and for such interesting data posted. I hope this can help all other modders, especially for those practising like me

 

Here is the optimizing result

I've exported max model to obj and then import in lightwave3d (i use it a lot at work and know it better than max) and cleanup the mesh. Now the count tells me 96k tris and 68k polys.

A lot better than it was before

View attachment 2348

 

This is looking very good, just the sort of content I'm hoping to see in rF2. Good luck with this, tommy86.

 

Ben fatto.well done!

 

Oh my God !!

Amazing, Work
Is what you would do on rFactor 1 an audi?
Sorry if the question was posed

 

O m g looks amazing, but how do you get the sound for the audi?

 

Here some spinner screens :

View attachment 2357
View attachment 2358

 

Amazing work!!!

 

Ultra!

 

Are textured headlights a good solution ?

View attachment 2360

 

Please NO!!


Use modeling stuff under the plexi cover...

 

Agree totally

 

ok ok I'll recover old modelled headlight, don't worry