Squad is optimizing colliders?

[Enceos]

Squad has mentioned twice in KSP Weekly that their 3D designers are working on optimizing colliders.

Does anyone have an idea of what they're doing specifically? Does it differ from what we usually do to optimize colliders?

[riocrokite]

1 hour ago, Enceos said:

Squad has mentioned twice in KSP Weekly that their 3D designers are working on optimizing colliders.

Does anyone have an idea of what they're doing specifically? Does it differ from what we usually do to optimize colliders?

No idea, imho colliders are currently quite optimized. what comes to my mind is adapters and cockpits since they usually have 2 mesh colliders so you can merge them at the cost of being less accurate. Some time ago I tested whether it would be beneficial performance wise to replace tanks mesh colliders with i.e. a box+capsule and it wasn't (not to mention mesh - collider inaccuracies).

[liquidhype]

They are basically reducing the complexity of the colliders and in some cases reducing the amount of colliders per part. Believe it or not, the collider polycount and the amount of colliders per part has more impact on performance than the complexity/polycount of the actual model itself, as it requires more draw-calls. People have tested this in the past by making 500k-1mill polycount parts and comparing their performance against a part which consists of multiple high-poly colliders and multiple parented meshes.

What Squad is doing will not boost the performance significantly, but it will certainly help those with less capable computers.

[Enceos]

@liquidhype Hmm, I have a model with 9 colliders >_<, 2 mesh colliders and 7 Unity primitives, is that bad?

I know that colliders are heavy on the CPU while model complexity is heavy on GPU. I tend to join objects with the same material where I can to reduce to amount of draw-calls of my models.

[liquidhype]

2 minutes ago, Enceos said:

@liquidhype Hmm, I have a model with 9 colliders >_<, 2 mesh colliders and 7 Unity primitives, is that bad?

I know that colliders are heavy on the CPU while model complexity is heavy on GPU. I tend to join objects with the same material where I can to reduce to amount of draw-calls of my models.

Yes the wise thing to do is always to join as many meshes together as possible. The only time you should ever have multiple meshes in one part is when you are making an engine, solar array or parts which needs to move independent of each other. 

I never, ever use the part itself as a mesh collider. I always use the most simple low-poly shapes possble if a part needs more than 1 collider. Even if you are making a high-fidelity engine, you can still stick with a simple 12-sided cylinder as your collider. There is no need for separate colliders for each engine nozzle (assuming it has more than one nozzle), and a collider for the engine base, turbo-pump, etc. Keep the colliders as simple and as few as possible. And try to have your parts consists of as few independent meshes as possible. Doing that for all parts will help boost performance, especially if you have a lower-end PC.