welcome mesh, goodbye prims?

[Zslash Cyberstar]

lol (no this is not a trollpost) look at the pic :P

http://img830.imageshack.us/img830/5673/goodbyeprims.jpg

 

[Darien Caldwell]

I don't get it. Why should I care about some number printed on a frame? A prim is still a prim, no matter what arbitrary number which may be associated with it.  Try harder.

[Drongle McMahon]

The number on the left of the slash is the actual cost, usually the "streaming cost", that will be charged against the parcel prim allowance. The number on the right is the physics cost. For meshes, if the physics cost were larger than the streaming cost,  it would replace the smaller cost and become the actual cost. For standard prims, the actual cost is always 1 (unless you change the physics shape type to "Convex Hull").

I think the physics cost is left there for standard prims just to let people know how bad some standard prims are for the physics engine. The costs are all due to change before release anyway.

Here is a table/picture of four near-identical toruses (tori?). The ones on the left are standard prims and the ones on the right are meshes that have exactly they same triangles as the standard prims at every LOD level. They also have a physics shape so that the hole is open. At the top, they are set to physics shape type "Convex Hull", and at the bottom to "Prim". In the former, the hole is not accessible.

At this size (not if it is much bigger) using the "Prim" type makes the mesh more expensive because the physics cost [6.8] exceeds the streaming cost [1.8]. It is, however, much less than the "real" physics cost of the standard torus. This is because it is a set of simple convex hulls based on the low-LOD mesh. Unfortunately, nobody would use the mesh one because it is much more expensive although it is much better for the physics engine.

Both standard prim and mesh are the same cost when they use their convex hulls for the physics shape, as one would expect. Alas, even if the physical hole was not needed, people would resist using the convex hull form for the standard torus because it would cost more prims, despite the fact that it is much less strain on the physics engine. Maybe having the unused physics cost there will encourage decisions that are better for the efficient use of the physics engine.

This is rather an extreme example. The cost calculations are necessarily a compromise. In most cases they don't give anomalous results like this.  Fortunately, the physics costs get lower as the triangles in the mesh get bigger. This is good for buildings and landscapes. Anyway, these are transient observations as the costs are likely to change anyway.

[loni Lynwood]

Hmm a very interesting observation. I wonder if we can all do what is right for the rendering cost versus the prim limit?

A fine line to tread, but as most would (& should) agree, lower prim count isn't necessarily always better than render cost.

I for one am willing to give the ass to every megaprim I own if the 64mtr prim comes in, though that is a side issue lol

Thanks for a well thought out post, too many kneejerk reactions at times, so yours is a refreshing change

[Drongle McMahon]

Thanks Ioni, but please note that the numbers in this post are out of date now. The costings have changed, and are pretty certain to change again before release. I think getting the the costs adjusted is the main difficult task left for the developers.

[DanielRavenNest Noe]

Indeed, they need to find the right balance between making the costs low enough that people use the mesh feature, but not so low that mesh objects end up lagging everyone from being overly complex.  The information we have been given on the triangle load per sim (250Ktris/15000prims) at lowest LOD indicates we should try to make the lowest LOD at 16 triangles per prim, which is enough for a simplified "billboard" version.  A "billboard" is one or more flat planes with an image of the object on it.  At very long distances you cannot tell that from a full 3D model, and the technique is used often in other graphics software to simplify rendering.

At the nominal scaling of 4x per LOD level, that results in 16 x 64 = 1024 triangles per prim at the most detailed LOD level, which is quite generous.

[Drongle McMahon]

But look at the table I made concerning the effect of object size. You only get any saving at all from thev lowest LOD if the mesh is smaller than about 5x5x5, and is progressively less from 1 to 5m. So for meshes that size, using your calculation you would have to have the low LOD at 16, and that's only 256 triangles (128 quads) per prim. The next critical size is 10m. Then you only get 64 triangles per prim! At 40m, 16 triangles per prim.

 

[arton Rotaru]

The ignorance of the lowest LOD at >5m, <10m bugs me the most. Now I'm going to split my meshes in smaller pieces, to benefit from the lowest LOD. Results in more assets to load...

[Drongle McMahon]

I did that with the windows for my gallery, as a test case. It was 13 windows and a doorframe in one mesh. I split it into 14 separate meshes and linked them. That came out at half the cost. All the LOD meshes were identical, but I had to have a separate physics shape for each piece, while the original had one for all of the. So more overhead and more physics. (OK, I know; I could make a prim collision mesh, link it and set the pieces to "None")

[arton Rotaru]

Well, that sounds promising. :matte-motes-sunglasses-3:

[DanielRavenNest Noe]

If it was 13 copies of the same window, that actually makes some sense.  The window asset only needs to be fetched once to your viewer cache from the asset database, and then once from there to your graphics card memory ( assuming a sane graphics engine).

[Drongle McMahon]

Yes. In this case it was three different windows that were repeated, so it would save download. However. that wouldn't generally be the case, and It isn't a factor in the prim equivalence calcultions. I just had this handy to do the test.