Forum Discussion
LOD vs Tessellation?
I work with 2 artists on a small project for the DK2 and the assets we create should be future proof and will be only used for VR. I did some research but couldnt find proper answers to what is better in terms of performance.
Should we either go for the typical LOD system with 3 different levels of detail for the models or should we go with lower poly models and then add tessellation once the player gets closer?
We are also thinking about a hybrid approach as having one very low poly model, one higher poly model and then the tessellation once people come really close. Any thoughts on this?
Should we either go for the typical LOD system with 3 different levels of detail for the models or should we go with lower poly models and then add tessellation once the player gets closer?
We are also thinking about a hybrid approach as having one very low poly model, one higher poly model and then the tessellation once people come really close. Any thoughts on this?
5 Replies
- There is no such thing like future proof assets.
If you want to use tessellation you need to create model with that in mind as it will be created from bezier surfaces and not from simple polls/quads. That means different CP's positions than for classic vertices. With tessellation you still need traditional LOD system and switch back to triangle mesh at some distance (as patches will be still to dense to use them intesselated as low poly mesh). This also requires blend transition between the two to prevent popping of geometry.
Question is do you really need tessellation as this is not trivial stuff to do properly.
On the other hand, parallax mapping and normal mapping is just not working in VR, so you need real geometry if you need more details -> but using tessellation for that means supporting displacement maps and maybe vector displacement maps which is even more complicated. - yeah i agree with above, just go with lod. I am stuck on a gtx650 but tessellation or displacement tessellation isn't as performance savvy at standard draw 60fps let alone taking vr into consideration. Maybe on the later cards it might be better but I haven't been able to test. While cool and if all future cards could do it at a performance level equal to just hiding and showing a different mesh I'd say go tessellation but for assured performance, and simpler modelling pipeline just go traditional LOD.
- It depends on the model. Go with LOD's with a "complex" model, such as a chair or person. Use tessellation for non-complex models, such as a brick wall, leaves, or grass.
"saviornt" wrote:
It depends on the model. Go with LOD's with a "complex" model, such as a chair or person. Use tessellation for non-complex models, such as a brick wall, leaves, or grass.
To save time modeling it or why do you recommend this?
Thanks for all the input!- Saves time on modelling, and, this is a sneaking suspicion that is not tested, but I feel that tessellation works the GPU a bit harder than using an LOD system, that's why you don't want to use it for complex models.
Don't use one or the other, use both techniques, as there will be some models / assets that will be better (in all aspects, time modelling and GPU usage) using one versus the other (both in terms of time modelling and gpu usage).
