[REL] S3D - Silent 3ditor

[DarkFish]

I tried restarting the mission, but it doesn't work.
Looking at the aircraft glass didn't get me further either, it's not too different from some other transparent examples in game (FD, trees)
Judging from these it must be somehow possible to eliminate the rainbow effect, but I just can't figure out how

In the meantime I did find out that there doesn't have to be an alpha channel in the texture, just enabling alpha+disabling z-buffer write in S3D is enough to get this strange effect.

You don't have the unifiedrender controller active do you?
I notice the Glass does not use that controller.

[DarkFish]

I didn't, I just set up a test mission with the model using a UnifiedRender controller and IT WORKS

Apparently models need this controller to counter the rainbow effect

all thx for your help

Very cool.


Maybe the Glass stuff has a hard coded controller then.


Anyway,
Glad you got it sorted.

[skwasjer]

Originally Posted by keltos01

tried... gave the same screen...

first time ever this happened to me !

keltos

I've already replied to Keltos in pm a week ago, but the game is limited to 65K (65536) uv indices per model. Depending on how good the uv-map can be compressed, this equates to a similar amount or more number of texture coordinates, but there's no rule here.

The solution is to break up the model into two or more parts... This is recommended anyway, because the bigger a model, the more CPU/GPU power is needed to be rendered, as opposed to having the same model in 10 submodels. Clipping a model, computing a bounding box, lighting computations, hitzone/intersect calculations, running it through a shader, etc, all can usually be done faster on several smaller models...

Good information.

Also,
It's easy to get a bit carried away with doing 3D models.

I studied some tutorials on low poly modeling.
And that helped alot.

[skwasjer]

Very true. This is actually the hardest part of modelling, making a high poly model look good at half or even less the number of polygons (as far as 10x smaller) from the original.

I forgot to add, any modern game uses a form of lazy loading, meaning it won't load an object unless is has to (S3D does so too partially). Translate this to a huge single model vs 10 smaller ones, anyone can understand why breaking up models is faster. If sections of the model are not in view, it shouldn't need to load them. It saves precious system memory, doesn't need to access your disk, doesn't need to upload the model from system mem to GPU mem, and so on...

The opposite is true for materials/textures, 4 1024x1024 textures is usually worse than one 2048x2048, but it depends a bit on the situation.

Do you think or know that the models in SH series are broken up because of this?
As an example,
detaching some faces even though a part of a single obj file?

[skwasjer]

Detaching objects in 3D program as you've indicated in the past is usefull for making unique vertices (with different normals) for a plane, allowing for hard edges. But if you import it as a single object, it's still one object, just with a little bit more (duplicate) vertices. So in context of what I wrote above it won't make alot difference as far as performance is concerned.

But if you import the subobjects from an OBJ-file into seperate 3D chunks, then they are considered seperate objects that can each be loaded (or unloaded for that matter) whenever needed. When an object is loaded into system mem but can be clipped because it's outside of the view frustrum, then it won't pass through the entire render phase, leading to better FPS. Big objects that are only visible for 10% still need to pass through the render phase completely, which hurts performance, especially when you run low on GPU mem...

So yea, in some way the developers definately considered this, although it's not apparent (or not needed) in all game files.

Very cool.
That answers a few things.
Thank you.

[skwasjer]

One thing I'd further like to add is, don't go overboard by dividing a model into too many submodels either. In the end, more models have a little bit more overhead in terms of system memory, matrix computations and device state changes (swapping a material, setting renderflags), so you have to balance it out a little. The best rule of thumb probably is to stay away from large models with big dimensions that will always be in view when they are not cut into pieces. In interiors you can probably get the best improvements, for instance:

- don't use one big model for the hull because then it will always need to be drawn no matter where you look.
- Don't combine two desks in an interior that are spaced apart 40 meters...
- etc.

For the very same reason, ship hulls (even though big dimension), are usually one big model, because you are likely to see it completely anyway (you are not supposed to be only 50 meters away from it right?), so breaking it apart doesn't give you a performance boost, on the contrary...

Confusing?

[keltos01]

clear as day !

making the guns decks a separate obj in the Narwhal sub allows me to work on it without having to go through the hassle of rebuilding the whole hull and hull AO map everytime -> faster too

Thanks Skwasjer !

keltos

[skwasjer]

Patience... I can't access my home computer right now because I had to move out temporarily (I recently seperated), but I will post it as soon as I can.

Regards...

[Rubini]

Hi guys,

I already asked skwasjer but he recall me that he don´t have his files for a while.
So, can someone send me a link (by PM ) of the last S3d beta with zones edition feature?

Thanks!

Rubini.

[iambecomelife]

Originally Posted by Rubini

Hi guys,

I already ask skwasjer but he recall me that he don´t have his files for a while.
So, can someone send me a link (by PM ) of the last S3d beta with zones edition feature?

Thanks!

Rubini.

Same here, please.