Will using HR2 in non-dots project help with rendering performance?

Hybrid Renderer is the rendering system for DOTS. If you want to make use of it, you have to use DOTS for the parts that you want to render with it.

If your game is otherwise non-DOTS, this would probably be the easiest to do for things like static background objects and such, which you could place into subscenes so they get converted into DOTS entities.

 

Rendering performance will strongly depend on the type of content, and how well it batches.

HRV2 is the fastest in the case where there are many objects which have the same Mesh and Material, which can be rendered with large instanced draw calls. In this situation HRV2 should be a lot faster on the CPU than using GameObject based rendering.

In the worst case, if everything has a unique Mesh or Material and there are no batching opportunities, we have usually seen that HRV2 rendering performance is roughly equal to GameObject based rendering. In this situation it is likely to consume more memory, as the DOTS chunk fragmentation is likely to be high and a lot of memory is wasted.

On the GPU, HRV2 rendering should usually be approximately the same speed as GameObject rendering, but that is likely to be both hardware and content dependent.

 

In HRV2, a different submesh will behave the same as a different mesh, so it will break the batch.

 

We are working on improvements, but unfortunately I can't share details about them right now.

 

Did I understand correctly, that this cost is only observed when the objects initially enter or leave the frustum, but is not seen while they stay inside and keep rendering each frame? Such a cost is definitely unexpected. Would you be able to share some Unity Profiler captures, where this unexpected spike is seen, and a comparison frame which is otherwise similar but is missing the spike?

Hybrid Renderer does use a different rendering code path compared to DrawMesh, so it's possible it is not affected by this issue, but I'm unable to confirm this without understanding what causes the original performance cost.

 

Hello!

We are working on some features which are intended to help with speeding up parts of GameObject based projects using the same technology as the Hybrid Renderer is using, including a much improved version of the BatchRendererGroup API. When that tech is ready, we will have samples and documentation for it.

If you are looking for something you can use immediately, then using Hybrid Renderer with the ECS should be the easiest way to go.

The simplest approach would be to place some of your static GameObjects into subscenes, which will then be converted into DOTS Entities and rendered using Hybrid Renderer. This process removes the original GameObjects, so there could be issues with things like game logic scripts or collisions. However, if the objects are purely visual, then this is likely fine. Alternatively, you could manually create rendering Entities using the RenderMeshUtility.AddComponents API, while leaving the GameObjects in place with disabled Renderers.