LOD Ratios

[Rusalka Writer]

Do we have a final required ratio between LOD levels? I'm fiddling with uploading an old structure, and I'd like to know what to shoot for before I make lower LOD models. Right now the automated system looks like

High : Medium — 4 : 1

Medium : Low — 3 : 1

Low : Lowest — 2 : 1

That first step is a doozy and the model looks like hell from a hundred meters. The temptation is to throw caution and cash to the winds and pay for an favorable error limit, which will certainly violate the goal of making mesh the more efficient option.

Maybe I'm just looking at the real cost of trying to make a large structure with even detailed pieces in mesh. I hope I don't turn into a jewelry maker out of frustration.

[Medhue Simoni]

Good luck! Sorry, can't help, as I've not even tried to make a house or structure. I think it would be very tough under the current PE, but I've not checked the latest changes.

[Drongle McMahon]

Unless something change very recently, there is no required LOD ratio. You can load the same mesh in all four slots, but that will make it very high PE. The best strategy is very dependent on size. For something with a cube bounding box, the lowest LOD has by far the greatest effect for small meshes, but its effect decreases as the size approaches 6.25m on a side. Above that, it has no effect. For the low LOD, the effective limit is 12.5m, and the medium LOD matters up to 50m. 

So for smaller things, you can afford to be less stringent with the first ratio, and use a very simplified mesh for the lowest LOD.  When you consider that many people are using the defailt renderVolumeLODFactor, and see how close the first LOD switch is with that setting, you realise that not losing too much in the medium LOD is very important for smaller objects. However, you don't gain anything from increasing the detail on the high LOD so that you can have the same medium LOD with a higher ratio. In fact that will always make the weight higher, though not by as much as you might expect. As Runitai said, lowest LOD triangles are expensive and high LOD triangle are cheap. but this is less and less true as your mesh gets bigger. If it's bigger than a 50m cube, then only the high LOD counts at all.

That all assumes that the download weight (I think that's what we are now supposed to call what used to be the streaming cost) is dominant. If physics weight is higher , the considerations are very different.

[Ashasekayi Ra]

As Drongle said, the importer doesn't force you to decrease the triangle count for each LOD step. However, during one of the meetings, a Linden developer stated that decreasing by a factor of 4 for each step down will give you decent streaming costs. And really, I wouldn't bother with the automated LOD. You can get much better PE and looks doing the LODs (especially the two lowest) by hand.

[Rusalka Writer]

Groovy. I remember a meeting long ago when it was 3:1 and it was a bit of a palaver. I'm happy to hear it's in my hands and on my account. That I can wrangle.

Mahalo!

[Rusalka Writer]

Now the problem I'm having is the realization I can just make 90% of the *$&@(#*$ thing out of regular prims that will look fine at a distance with less work and the same prim count. Hmmm...

[DanielRavenNest Noe]

Despite the fact that a wall can be made from 4 triangles in mesh and needs 12-108 for a box prim, mesh has a higher prim cost when it's large.  For typical buildings, the best use for mesh will be for complex objects like windows and fireplaces.  Windows can reduce to a flat plane at the lowest LOD, and indoor items like a fireplace will never be seen at the lowest LOD, so you can drastically reduce the triangle count and get a reasonable cost.  And items like that which are only a couple of meters in size will not get penalized too much for size.

 

[Rusalka Writer]

Yep, the prim count for my poor ol' lighthouse means it makes no sense to build it in mesh. Getting the count down would require LODs that look dreadful at even the slightest distance. Other than a couple of unusually-shaped parts, there's really no point to building something this size in mesh.

[Gaia Clary]

Although maybe it is (currently?) not a hyper cool thing to do, you still can cut your lighthouse into reasonable pieces, and then use these smaller pieces as "special mesh prims" for "building" your lighthouse. So you can easily create a combination of regular prims (where sufficient)  and mesh prims (where needed for details or nicer look).

I actually plan to do exactly that with the lighthouse example which we introduced about 9 months ago in another video. thinking of "creating building blocks" made out of meshes may in fact be a good way to get parts of your building activities back into SL ;-)

I could imagine to have building blocks for the windows, for the walls, for the light room itself, for the roof... And i can imagine that getting along this route may in fact be a good compromise...

However i must admit that i am talking in theory here, as i have not touched big meshes so far.

[Marcthur Goosson]

Isn't it so that all these small pieces go earlier to next LOD level then a whole structure would do when zooming out?

 

[Gaia Clary]

exactly this is the reason why small pieces are cheaper than big ones ;-) But as long as you know how to make a good LOD decomposition this is acceptable (and even wanted). Drongle McMahon has proposed an idea for customizable LOD transition settings which would allow big pieces to transit to lower LODS earlier and thus get cheaper...

but this will not be available in the first release... maybe later...

[Nix Manx]

Ive been having a good old fiddle with LODs today and there are some interesting results, specifically, the object detail slider has a huge impact on when LODs change over. The most interesting part of this is if your slider is all the way over to the left (lowest object detail) you will never see the 3 highest LODs unless its a small item.

However if its 1 notch up from the lowest it will go, you will see the the second lowest LOD at about 60m away, as many are using billboards for the lowest LOD, anyone on the lowest possible object detail setting will be out of luck for meshes.

Its going to be tough to find that acceptable medium on LODs i think.

[Drongle McMahon]

There is a debug setting called renderVolumeLODFacttor that is one of the setting changed by the object detail slider. It is a simple multiplier of the LOD switch distances. /you can access it directly through Advanced->Show Debug Settings. Type in the name and when it autocompletes you can enter the setting. Lowest object vdetail sets it to 0. Raising it will greatly increase the strain on your graphics card, so experiment to see what you can tolerate.

The switch distance for a particular prim (mesh or otherwise) is determined by it's size. The distances are about 2, 8 and 16 times the diagonal of the bounding bax of the prim, multiplied by renderVolumeLODFactor. It can be difficult to measure exacly because the LODs are sort of lazily updated. You have to do a lot of zooming in and out to trigger the changes.