On Lod issues li count

[Naiman Broome]

I am trying to reduce the the most the li count of my object , but when I use precise and well arranged lods the prim count is 3 , when I use the messy automatic lod creator of SL uploader the prim count is 1 , yet , the lod levels have the same exact triangle faces... same size etc .. so why that?

 

Linden uploader generated Lods :

 

My custom created lods :

Edited July 9, 2021 by Naiman Broome

[arton Rotaru]

What about the physics shape? Clicking on the "More Info" link in the build floater will reveal all the different weights.

[Naiman Broome]

I did add indeed a physic prism shape that helped reduce further of one li bringing it to 2 but no less ... still dont get how to reach 1 , also cose I will need do split the mesh in 2 and have it be 2 li prims but right now would be 3 or 4 .

Edited July 10, 2021 by Naiman Broome

[Naiman Broome]

13 hours ago, arton Rotaru said:

What about the physics shape? Clicking on the "More Info" link in the build floater will reveal all the different weights.

This is with the regular uploaded item with generated lods

 

 

and this is mine with precise made lods :

 

 

I dont understand why mine, that has even better collision prim and physics it counts as double???

[Naiman Broome]

So basically if you Autogenerate the lods you save Li and primcounts , while if you manually do your lods you get penalized ???

[arton Rotaru]

1 hour ago, Naiman Broome said:

So basically if you Autogenerate the lods you save Li and primcounts , while if you manually do your lods you get penalized ???

No!

First of all, this time your triangle numbers aren't identical between the two.
Even identical triangle numbers doesn't necessarily mean identical vertex counts. The in-game vertex count is driven not only by the geometry, but also by UV splits, hard edges and material borders. So to compare models you will have to look at the actual in-game vertex counts. Which will be shown in the importer, and can be inspected in-world by enabling the Develop > Show Info > Show Render Info display.  By selecting an object, and zooming in and out to switch through the LODs. (There is also a free script available for 3ds Max, called Gamefomizer. Which will show in-game vert counts inside 3ds Max.)

So the difference between the 2 models here is, that the auto LODs look like crap, and yours probably not. That at the cost of just one LI. Most often pretty much worth it IMO.

On a personal note, a Land Impact of just 2, for an object with a complexity of almost 13k triangles is super low actually. With models of this rather small size, I usually aim for a 50 to 60 % reduction for the medium LOD. While being more radical on the Low and, especially on the Lowest. On rather small objects, most often, this doesn't even increase the land impact at all. 
With a default LOD factor setting, which is something between 1.0 and 2.0 (not 4.0), the swtich to the medium LOD happens rather early, and I would imagine the transition is rather harsh, going from 12600 to just 100 tris in one step. And if it's not harsh, I would question why the High LOD is 12600 tris to begin with.

Of course, without seeing the model, or knowing what it is, it's just stabbing in the dark indeed.

 

 

Edited July 10, 2021 by arton Rotaru

[ChinRey]

7 hours ago, arton Rotaru said:

And if it's not harsh, I would question why the High LOD is 12600 tris to begin with.

I haven't seen the item so I don't know for sure but it may make perfect sense.

For an object as small as this, the LoD system works pretty much as Ye Olde Gods of Second Life intended it to: the high LoD model is the "special occasion" one only used when somebody cams in on the object to study it in detail while the mid LoD is the regular one. That means you can afford to fill up the high LoD model to your heart's content without affecting the LI or performance very much. There's an important lesson here: Optimisation is not only about saving LI and reducing lag, it's also about making room for all those lovely details that really count.

10 hours ago, Naiman Broome said:

I dont understand why mine, that has even better collision prim and physics it counts as double???

Your physics model has a lot lower weight than the auto one so you defiitely got that one right.

As for download weight, I do agree with Arton that one extra LI is well worth it if it means much better LoD. However, 2.1 is only 0.6 above the threshold between 1 and 2 LI and with such small difference we're close to the realms of micro optimisation with all those little factors that are hard to predict. Tweaking those factors may well be enough to get it down:

• Arton mentioned the vertice count and that may be the most significant factor. Most of the mesh data is stored in the vertices, not the triangles so there can be a lot to gain if you can get rid of only a few of them, especially in the low and lowest models.
• See if there are any tris you can get rid of. In this case it is probably the low LoD model that is the most critical and every single tri counts there. But any tri you don't really need in any of the models will help a little bit too.
• Is the model triangulated in the dae file? That can add quite a bit of download weight. Merge as may triangle pairs as possible into quads and deselect the "triangulate" button when you export. (I'm assuming you're buiding in Blender btw, the method for other programs may be slightly different)
• Try to align the vertices as well as possible both in the 3D model and on the UV map. A quick and dirty trick that can often shave of a bit of download weight is to select all, go to the UV map, check the "Snap to pixels" option, make sure the other snap function (the one with a magnet icon) is unchecked, type S to scale but don't actually resize. That way all vertices will move to the nearest whole pixel on the UV map. It will simplify the file a bit and it shouldn't make any noticeable difference, at least not if you only do it on the lower LoD models.
• The sorting order of the vertices makes a difference too. It's hard to predict what is the best option but try different ones, and see how they work. Of ocurse, ake sure you keep a backup of the original in case that turns out to be the best one after all.

 

9 hours ago, Naiman Broome said:

So basically if you Autogenerate the lods you save Li and primcounts , while if you manually do your lods you get penalized ???

When the difference is as small as this, it can both ways.

[arton Rotaru]

21 minutes ago, ChinRey said:

When the difference is as small as this, it can both ways.

When people get the impression that they get penalized by using custom LODs, they usually compare apples to oranges. In fact, it's the other way around, you can save a lot of LI with custom LODs. Because, the reason you make custom LODs in the first place, is because you do care about the aesthetics of your model throughout the LODs. Of course, when using auto LODs, you will still have to apply the same aesthetic expectations to your model. With auto LODs you will need to dial in pretty high triangle numbers to meet similar aesthetics as to what can be achieved by making custom LODs. Resulting in fairly high land impacts.
Even inexperienced modelers are able to outperform the generator quite easily. Maintaining a better, cleaner shape with fewer triangles.