The importance of LOD

[ChinRey]

My mesh students and others sometimes ask me why LOD is so important for a mesh maker. I think this overview picture of a full sim, full mesh fairy tale village - taken with default graphics settings, says more than a thousand words...

 

 

[Chic Aeon]

Agreed, BUT it isn't only mesh.

A couple of days ago I had a project making a 1960s Christmas tree for an event. I could have made it in mesh but prims were going to be just fine and LOTS quicker. My tree ended up at 137 prims. Oh my!  Yes, but after a year on grids where you have 45,000 or more prims per sim and even 1000 or 3750 for a rental spot, I finally GET that "prims don't count THERE" LOL.

 

Anyway. Long story but I needed to transfer the prim build over to another grid using the backup function in TPVs. Because of some technical issues I had to use Singularity (xml) rather than Firestorm (oxp). When the tree arrived the spheres (just PRIMS now, not mesh) looked like a polyhedron, NOT ROUND at all. I was not happy until I realized that it was the LOD setting in Singularity (I rarely use that viewer).  At LOD 4 the ornaments looked perfect. I just went back and checked and the spheres are OK (shortish distance) at 1.25, fine at 2. They fall apart even within a few meters at 1.

 

I didn't check what the default in Singularity is. I know that FS is 2 and I test using that often. The Linden viewer is 1.25 unless that has changed.

So while I wholeheartedly agree that LOD is important -- it is also important to realize that even the prim world looks VERY different when settings are turned down or left at default. Many folks know to change things. Some cannot because their computers are not hefty enough. So for those folks many things are going to look less than optimal -- not just mesh.

 

[Drongle McMahon]

Good illustration. This sort of thing happens mostly/worst when people try to cheat the LI system, not caring about those who can't afford or don't know how to set high RenderVolumeLODFactor. The one-triangle medium LOD trick should have been prevented somehow. Too late now. For me, the challenge of making good LODs is part of the enjoyment of making mesh. I guess others don't share that. and it is hard work.

[Pamela Galli]

---

Chic Aeon wrote:

Agreed, BUT it isn't only mesh.

A couple of days ago I had a project making a 1960s Christmas tree for an event. I could have made it in mesh but prims were going to be just fine and LOTS quicker. My tree ended up at 137 prims. Oh my!  Yes, but after a year on grids where you have 45,000 or more prims per sim and even 1000 or 3750 for a rental spot, I finally GET that "prims don't count THERE" LOL.

 

Anyway. Long story but I needed to transfer the prim build over to another grid using the backup function in TPVs. Because of some technical issues I had to use Singularity (xml) rather than Firestorm (oxp). When the tree arrived the spheres (just PRIMS now, not mesh) looked like a polyhedron, NOT ROUND at all. I was not happy until I realized that it was the LOD setting in Singularity (I rarely use that viewer).  At LOD 4 the ornaments looked perfect. I just went back and checked and the spheres are OK (shortish distance) at 1.25, fine at 2. They fall apart even within a few meters at 1.

 

I didn't check what the default in Singularity is. I know that FS is 2 and I test using that often. The Linden viewer is 1.25 unless that has changed.

So while I wholeheartedly agree that LOD is important -- it is also important to realize that even the prim world looks VERY different when settings are turned down or left at default. Many folks know to change things. Some cannot because their computers are not hefty enough. So for those folks many things are going to look less than optimal -- not just mesh.

 

---

I discovered that when I export say, a house, as .dae in Singularity I need to put my avatar in the center of the house, because some meshes will export as lower LOD if they are too far away from the avi.

[ChinRey]

---

Chic Aeon wrote:

Agreed, BUT it isn't only mesh.

---

That's true but with mesh and sculpts builders can do something about it. Expecting the visitors to increase their LOD factor is hardly an ideal solution since it add extra strain on the computer's graphics processing unit and I've yet to hear of a GPU that is too powerful for SL.

 

---

Drongle McMahon wrote:

Good illustration. This sort of thing happens mostly/worst when people try to cheat the LI system, not caring about those who can't afford or don't know how to set high RenderVolumeLODFactor.

---

Or don't have computers powerful enough to increase RenderVolumeLODFactor or would rather use their GPU's limited resources to improve some other aspect of SL graphics or have been tricked into setting it too high. (Raising the LDO factor will increase the switch point delay and if it's set too high for your computer, you effectively end up with worse rather than better LOD)

Maybe I should have mentioned that this jumble of triangle has a land impact of 10 500. That long gray triangle in the bakground is a tower, just a single huge hexagonal mesh yet it has a land impact of 8! I can't for my life understand how it's even possible to make mesh as inefficient as that. Pinching the LOD is not only a very destructive way to reduce LI, most of the time it's also a very inefficient one.

 

---

Drongle McMahon wrote:

The one-triangle medium LOD trick should have been prevented somehow.

---

I'm not sure if I'm familiar with that trick. Please tell!

 

---

Drongle McMahon wrote:

For me, the challenge of making good LODs is part of the enjoyment of making mesh. I guess others don't share that. and it is hard work.

---

I think it's about understanding how mesh - or rather how SL mesh (which is not really the same as mesh elsewhere) - works. One of the many great things about prims is that they're so easy to learn and to work with. You rez a few of them, twist them a bit and if you have enough artistic skills and pay attention to texturing you can turn them into wonderful 3D models. Mesh requires a bit more understanding how things actually work. A model can look wonderful in Blender or as prims in SL but still be horrible when meshed and uploaded. The most basic skills aren't that hard to learn really but they have to be learned.

 

---

Pamela Galli wrote

I discovered that when I export say, a house, as .dae in Singularity I need to put my avatar in the center of the house, because some meshes will export as lower LOD if they are too far away from the avi.

---

That is interesting. I haven't used the Singularity converter much but I may well in the future.

---

One thing I should add in case somebody somehow manages to identify the build: As far as I know, the person who made this village is a competent mesh maker. But it was made more than a year ago and only intended as a temporary build for a special, short time event.

However, a month or two ago somebody else copied the whole thing (apparently trying to get people to believe they made it themselves) as a permanent replacement for a lovely old prim/sculpt marketplace that used to be there.

She didn't make that impossible tower either - that's one of the few things the new owners have added themselves.

[Medhue Simoni]

---

Drongle McMahon wrote:

This sort of thing happens mostly/worst when people try to cheat the LI system, not caring about those who can't afford or don't know how to set high RenderVolumeLODFactor.

---

Well, that is 1 way to look at it. I would look at it as a result of an extremely bad prim/land impact system, which forces creators to adjust LODs to get land impacts to a reasonable level. Now, personally, on builds that I don't plan on selling, I don't care at all what the land impact is, but that is because I own the sim. Just the fact that size changes the land impact is kind of crazy. It's not more for my computer to render. Of course you can correct me if I'm wrong here, but I suspect that size changes the land impact because of LL's servers, which is kind of ridiculous. The cost of servers, and their capacity goes up every single year, besides the extra efficiency LL got from fixing bugs and changing things to be more efficient. On top of all this, if the mesh is physical, there is more land impact added. If you have a mesh use pathfinding, it automatically is physical, and is a minimum of 15 prims. So, if you want rats running around an area, each will cost you 15 prims. LL handcuffed us.

When I make items for other game engines, I just make the LODs. They come out to be whatever they come out to be. People buy the models and they optimize their games however they want. Yeah, there are some rough guides, but in the end, it's up to the developer to decide how to do things. When I do stuff for a developer, they tell me how they want it. SL is totally different and unique, and LL set the limits. IMHO, that image is completely predictable considering the system we have.

[ChinRey]

---

Medhue Simoni wrote

Well, that is 1 way to look at it. I would look at it as a result of an extremely bad prim/land impact system, which forces creators to adjust LODs to get land impacts to a reasonable level.

---

Yes but again, reducing LOD isn't really a very efficient way of reducing LI. There are so many far more effective and less destructive ways to do that.

 

---

Medhue Simoni wrote:

Of course you can correct me if I'm wrong here, but I suspect that size changes the land impact because of LL's servers, which is kind of ridiculous.

---

Land impact is all about how much load the object puts on LL's servers and network. They couldn't care less about how much it loads down our poor li'l computers.

Size changes the land impact because it changes the switch points. Download weight, which is the one relevant here, is basically an estimate of how much data that needs to be transferred for the object. That depends on how much data it actually has (of course) and also how often it is viewed. A larger object has its switch points further away so the higher resolution models will be used more often, increasing the amount of data required.

It's fairly easy to counteract this by optimizing the mesh for its actual size and switch points.

 

---

Medhue Simoni wrote:

So, if you want rats running around an area, each will cost you 15 prims.

---

You don't really want to use pathfinding anyway. Even beside the LI issue, pathfinding is usually way too laggy to be useful and there are always better options.

[Medhue Simoni]

---

ChinRey wrote:

---

Medhue Simoni wrote

Well, that is 1 way to look at it. I would look at it as a result of an extremely bad prim/land impact system, which forces creators to adjust LODs to get land impacts to a reasonable level.

---

Yes but again, reducing LOD isn't really a very efficient way of reducing LI. There are so many far more effective and less destructive ways to do that.

---

There are? Maybe if you are building a house or a tree. If you are making anything else, then besides making an efficient mesh, there is nothing else you can do.

---

ChinRey wrote:

---

Medhue Simoni wrote:

So, if you want rats running around an area, each will cost you 15 prims.

---

You don't really want to use pathfinding anyway. Even beside the LI issue, pathfinding is nearly always way too laggy to be useful and there are always other better options.

---

My experience is that pathfinding works much better than anything any coder could ever do, hence why the system was implemented. I actually helped develop an npc movement system before we had pathfinding. They work good, but they also produce some lag. To be able to make them do what can be done with pathfinding, now you are talking alot of code, and waypoints to place around the sim, hundreds of them. Plus, even if you didn't use pathfinding, the npc would still be physical, and still subject to the extra land impact costs. IMHO, pathfinding is great, except for the land impact costs, and the fact that we don't have custom skeletons.

 

 

[Medhue Simoni]

I really hope they do not use this same system, or anything similar, in the new world LL is making.

[ChinRey]

---

Medhue Simoni wrote:

There are?

 

---

 

Oh yes. Here are some:

1. Optimized LOD models. The uploader generates the LOD models simply by removing triangles from the original. Unfortunately it's horribly bad at it and starts removing important ones long before it's done getting rid of the ones not really needed for that particular model. By making those models yourself, deciding which tris to remove and which to keep for each one, you can save huge loads of LI with only minor LOD reduction.
2. Balancing the weights. Generally a linkset made from few, relatively large, meshes will have a lower server weight but a higher download weight than one made from more and smaller ones. Since only the highest of those two weights can count as LI, it's a good idea to find the sweet spot where the two are just about the same. Advanced weight balancing is a lot more than that of course - we have to decide which parts to include in which mesh, we have to take physics weight into account and so on. But even the basic, crude, equalize-download-and-server-weight aproach will usually save loads of LI.
3. Smooth normals. Can greatly reduce the number of tris needed for smoothly curved surfaces. (Smooth normals can also make a complete mess of your build so better be a bit careful and test on beta before uploading to main when you're in doubt.) Smooth normals will affect LI in other ways too but only to a very limited degree and the effect can go both ways, sometimes you gain a tiny bit of LI with them, sometime you loose a tiny bit.
4. Equal tris. I mentioned that download weight depends on the object's size and complexity. If you combine small details and large low poly surfaces in a single mesh, you get the worst of both worlds. The details add complexity and the large tris add size. Unless you really, really know what you're doing, it's probably better to mesh them separately.
5. Efficient building. Be strict with yourself: which details does your build really need and which can it do without? It's always tempting to try to emulate Real Life but that's simply not possilbe. Real Life builders seem to have unlimited LI so they can spend lots of it on even the most minute details (anybody figured out how they do that btw?) In SL we are always on limited budgets and have to find ways to use our resources as effectively as possible. There's an old C&W song, called "The Builder". Second verse starts with "Every builder knows that the secret to survival / is knowing what to throw away and knowing what to keep." (Actually the commercially released version was renamed "The Gambler" but those lines make no sense in that context and we all know the true story, don't we?)
6. Textures. Well chosen, well applied textures can add a lot of quality to a build, reducing the need for overcomplicated mesh. Textures can be quite laggy though so you better be a bit careful. To quote texture master Sea Warcliffe: "Use your texturing powers for good, not for evil!"
7. Effect maps. The old bumpiness and shininess maps and the more recent normal and specular ones are all seriously buggy and should be aproached with extreme caution. But if you're fast and if you're lucky (errr... I mean if you're good and if you're lucky) you can save quite a bit of LI by using these for some of the finer details. We have quite a few other simple and cool surface effects to improve our builds too but you don't expect me to give away all my secrets here, do you? :matte-motes-wink: )
8. The WOW factor. A while ago Charlotte Holmeforth reminded me of one of the essential principles for decorating on a budget and it has helped me building ever since: Use the WOW factor for all it's worth! Make it all cheap and simple and then add one or two really outstanding features! Actually, you don't really want too many stars in your show even if you can afford them - they just start fighting each other for attention.
9. Work with you materials and surroundings, not against them. One of the most fundamental principles of all good art and craftmanship.
10. Use all available means. I was lucky enough to run into Aley just when I was starting as a builder and she taught me this important lesson. Mesh is not always the best solution. There are jobs that prims can do better, that sculpts can do better and that particles can do better. Only by selecting the most suitable material for each specific job, can you build truly efficiently. Oh btw, if somebody says you can't link sculpts with mesh, take a look at #2 on this list and tell them you know better.
11. Compressibility. The server doesn't transfer raw data to the client, it's compressed. You can shave off a tiny bit of download weight making the mesh as easy as possible for the compressing algorithm to deal with. Not much and usually not worth the bother but sometimes a few quick and simple tweaks are just enough to flip the LI down by a point.
12. Umm... Seems I've forgotten what #12 was supposed to be right now. I'm sure it'll come back to me though and besides, I have to stop at some point anyway. This list could go on forever...

 

(Edits: just correcting tyops really)

[Jake Koronikov]

I am not answering here to any specific question, but just some considerations:

- In games generally, average triangle count in a scene is anything between 50 000 - 1 000 000, depending on FPS requirements. In an acton game the scene is never as high as 1 million. In a slow motions strategic game it may get as high as 1 million

- Avatars are usually quite dense. It is not rare that an avatar has even 20K tris, or more. But: usually in games, there will not be dozens of avatars in a scene at once. In SL there is...

- In games, there are possibiliies to handle geometry more efficiently. In SL all the objects are loaded via network at sim setup. In games, the scenes can be opitmized so that only camera direction is loaded. Behind your back there can be used low-poly substitutes for physics and so on

- In general. One big mesh object is less expensive than several smaller objects linked together - in the GPU point of view

- In game engines, LOD handling can be different for separate object sets. There can be layers for small detailed objects that will disappear faster. For large buildings LOD handling can be such that they are visible from a distance. In SL we do not have this kind of functionality...

- According to my knowledge, SL uses a method called Frustum Culling. Even those objects behind large objects consume GPU. In games, there is possibility to use Occlution Culling whitch is less GPU expensive. Stuffs behind large objects are not drawn at all.

I believe LL has faced a lot of problems during the development of SL. How to combine reasonable user experience and still keep things high FPS. SL is not an action game, it is very slow game where people mostly just stand still, zoom and admire amazingly detailed builds.

[Drongle McMahon]

There can be layers for small detailed objects that will disappear faster. For large buildings LOD handling can be such that they are visible from a distance. In SL we do not have this kind of functionality...

I'm not sure I understand what you mean here. If you design your LOD meshes appropriately, you have quite a lot of control over when details, or whole models, disappear. You can also add another level of control with invisible geometry and/or joining/splitting objects. Of course, if you stick to the automatically generated LODs, that is pretty hopeless. I did once do a jira asking for a object-by-object settable LOD distance modifier factor - so I must have agreed with you a bit, but the existing syatem is pretty flexible.

[Drongle McMahon]

I would look at it as a result of an extremely bad prim/land impact system, which forces creators to adjust LODs to get land impacts to a reasonable level.

It doesn't force anybody to do anything. What it does is penalise those who don't pay attention to making things efficient for real-time use. The intended penalty was incresed LI. Unfortunately, it also allowed the alternative of terrible LOD behaviour, and many took that way out as it was easier and/or more lucrative. Allowing that was the mistake.

The original intent, reflected in the size-related calculations, was that the LI cost would reflect the burden of data downloading, which uses both server and clients bandwidth. Later, the developers shifted emphasis toward the rendering strain on the gpu, but retained the calculations as the two factors were (claimed to be) highly corellated.

If I remember correctly, the LI calculation was based on a scene budget (excluding avatars) of triangles for a viewer with vertain low graphics settings (effectively RenderVolumeLODFactor=1 and FarClip=188). I can't remember the figure, but it was well below a million and over 100,000; maybe 150,000 or 250,000*. In fact, the calculation was rather generous. It effectively assumed that content was uniformly distributed in two domensions. In fact, cameras and objects are spatially correlated, and distributed in three dimensions (albeit less spread verically). Accounting for either of these effects more accurately would have increased LI and made it even more drastically dependent on size.

I thought originally that the LI calculation it was overly drastic. However, having seen the extent to which every opportunity to avoid the effort of making efficient content gets exploited by at least a few, I can appreciate that some for of control was absolutely necessary. Without it the world would be unuseable for anyone without the fastest broadband and the latest gpu (even with it, for some - almost there now for my 1.75Mb/s connection). The system adopted was at least based on reasonably rational quantitative criteria. Of course it's not ideal, but I don't think it's worthy of "extremely bad".

*the figure is hidden in the old content meeting minutes somewhere.

[Drongle McMahon]

The one-triangle medium LOD trick should have been prevented somehow.

I'm not sure if I'm familiar with that trick. Please tell!

Setting the max triangle count for medium/low/lowest LODs to zero, so that you get one triangle per material at all but the highest LOD. Often recommended here for LI reduction for stuff always inside buildings, on the basis that you never see them further away than the first LOD switch. But if you have RenderVolumeLODFactor set lower than the designer, this can lead to horrible collapses to random triangles while stuff is within view. Tha earlier collapse of smaller objects is also often forgotten, and woe betide those who like to look through windows.

If done really well, with an explicitly one-triangle LOD models where the triangle is always invisible (e.g. facing into the ground), this can work well, as long a you can accept things disappearing altogether, but the zero auto-LOD method will never work like that.

As long as there are examples inworld, you can inspect the LOD behaviour of items you consider purchasing by dialling in lower RenderVolumeLODFactor while looking at them. You can't do that for things only shown on the Marketplace. That is a serious limitation of the latter that facilitates bad practices.

[ChinRey]

---

Drongle McMahon wrote:

I'm not sure I understand what you mean here. If you design your LOD meshes appropriately, you have quite a lot of control over when details, or whole models, disappear.

---

You're right of course. You can't determine LOD on an object-to-object basis in SL but there are lots of ways to manipulate things to get the result you want.

But - and this is a big but - how are builders going to find out about those tricks? It's not as if LL provides much info useful and understandable to the average builder. They were never big on documentation in the first place and the practical application of mesh is just as big - or bigger - a mystery to them as it is to the builders. (There's one exception but Runtai Linden doesn't write tutorials and besides, although he shows meshing skills well beyond what you'd expect from a Linden, I don't think he's quite at that level yet.) People like me and Drongle and Ele and Aquila and a few others have spent countless hours studying the topic, trying to dechipher the documentation that actually exists, experimenting, thinking, testing... and I don't think any of us can claim to have really got to the bottom of it yet. You just can't expect the regualr builder to spend that much time and effort figuring out how to mesh effectively.

I wanted this to be a thread about the builders' responsibility and hoped not to bring LL into it, partly because it would shift focus away from the main topic, partly because for the most part hey've done a good job this year, slowly but surely improving SL in all ways.

I guess I have to come clean though but first:

---

Medhue Simoni wrote:

I really hope they do not use this same system, or anything similar, in the new world LL is making.

---

I'm sure they won't. That would make the whole SL2 project pointless.

However, the village in my picture is currently known as Meauxle Bureaux, one of the latest great LL builds and "Home of the Moles". What it really is, is the backdrop from the 2013 Fantasy Faire exhibition. They've added the inventory for the houses of course and a few other details, mostly made by good mesh makers but the moles have also made some pitiful attempts at meshing themselves, including that tower I already mentioned. But all those houses, streets, landscaping etc. are simply the Fantasy Faire backdrop copied down to every detail. There's no mention of the creators anywhere, unless you actually check each and every object, so the casual visitor is led to believe that this is a genuine Mole build.

There's a pub in that village of course. The bar counter there is the generic rounded corner design everybody-and-his-uncle used to put together from a dozen or so prims. Clumsily texturized and meshed up to 24 LI. No attempt to clean up the mesh either, it looks exactly like it did when it was made from prims, only with twice the land impact.

This, my friends, is what LL considers to be top quality mesh building today and that doesn't bode well for SL2. :matte-motes-frown:

(In case somebody don't know who the Moles are, they're the people employed by LL to do public building works. They used to do great work but their laurels are quite worn down from all the resting they've done on them by now.)

 

---

Drongle McMahon wrote:

Setting the max triangle count for medium/low/lowest LODs to zero, so that you get one triangle per material at all but the highest LOD.

---

Oh, that. I didn't know the trick was known by that name, I just call it "zeroing out" a model. It doesn't usually reduce the model down to a single triangle though, there'll be one for each mesh face and sometimes one or two extra you can't get rid of too.

Also, if you've already uplaoded the mesh with a custom made model before, you'll have to delete the slm file or the triangle count will bnever go below that of the custom model.

[Pamela Galli]

---

Chic Aeon wrote:

 

I didn't check what the default in Singularity is. I know that FS is 2 and I test using that often. The Linden viewer is 1.25 unless that has changed.

So while I wholeheartedly agree that LOD is important -- it is also important to realize that even the prim world looks VERY different when settings are turned down or left at default. Many folks know to change things. Some cannot because their computers are not hefty enough. So for those folks many things are going to look less than optimal -- not just mesh.

 

---

Everyone's computer capabililty and graphics settings are so different, I have found I must build for a certain target capability, when it comes to a number of things, including LOD.  I dont, for example, build for those with the default LOD setting of 1.25, and with draw distance set to 256, as seems to be the case with the village picture. But I have found that most ppl whose computers can't take a higher than default LOD setting probably also dont go around with a DD of 256.  If they kept it at 96 or so as I do, then at a distance things simply disappear. 

My houses and most furniture are rezzed inworld, anyone can check LOD and know what they are getting before buying -- and no one should every buy anything rezzable without seeing it inworld first. 

 

 

[Jake Koronikov]

---

Drongle McMahon wrote:

There can be layers for small detailed objects that will disappear faster. For large buildings LOD handling can be such that they are visible from a distance. In SL we do not have this kind of functionality...

I'm not sure I understand what you mean here. If you design your LOD meshes appropriately, you have quite a lot of control over when details, or whole models, disappear. You can also add another level of control with invisible geometry and/or joining/splitting objects. Of course, if you stick to the automatically generated LODs, that is pretty hopeless. I did once do a jira asking for a object-by-object settable LOD distance modifier factor - so I must have agreed with you a bit, but the existing syatem is pretty flexible.

---

My apologies, I didnt express myself very clearly above. What I mean is about this:

http://docs.unity3d.com/Manual/class-LODGroup.html

The properties section in above link explains the idea what I was looking for.

I was not giving critisism for SL implementation, I was more like considering the possibilities in other platforms. And I agree with you that object-by-object setable LOD would be a great idea.

[Drongle McMahon]

Ah. I think I see what that's about - the ability to coordinate the LODs of several objects together. That would be a very nice addition. The uncoordinated switching of objects that are part of the same thing, because of either size or distance differences, can be very disconcerting and difficult to mitigate. It might be less easy to implement LOD groups in a system like SL, where, unlike a Unity game, stuff from different makers gets mixed up and the contents of a scene are not all controlled by one author. If we are right in assuming that the SL successor will be using something like Unity, or a comparable engine, it may well inherit much more flexible LOD control.

[Drongle McMahon]

an extremely bad prim/land impact system

I should have qualified my defence of the LI system by restricting it to the dowload weight component. The triangle-based physics weight system, on the other hand, is entirely deserving of your assessment!

[Theresa Tennyson]

---

ChinRey wrote:

---

However, the village in my picture is currently known as Meauxle Bureaux, one of the latest great LL builds and "Home of the Moles". What it really is, is the backdrop from the 2013 Fantasy Faire exhibition. They've added the inventory for the houses of course and a few other details, mostly made by good mesh makers but the moles have also made some pitiful attempts at meshing themselves, including that tower I already mentioned. But all those houses, streets, landscaping etc. are simply the Fantasy Faire backdrop copied down to every detail. There's no mention of the creators anywhere, unless you actually check each and every object, so the casual visitor is led to believe that this is a genuine Mole build.

---

It was profiled in a blog post by Inara Pey before it opened as Meauxle Bureaux and she made it clear exactly what it was - it's no secret. The LL blog post describes it as being by "resident artists" as well. And they didn't really "copy it to the last detail" - basically, they moved it. All objects have the original creator's name. You make it sound as if it was copybotted.

[Medhue Simoni]

---

ChinRey wrote:

---

Medhue Simoni wrote:

There are?

 

---

 

Oh yes. Here are some:

1. Optimized LOD models.
   The uploader generates the LOD models simply by removing triangles from the original. Unfortunately it's horribly bad at it and starts removing important ones long before it's done getting rid of the ones not really needed for that particular model. By making those models yourself, deciding which tris to remove and which to keep for each one, you can save huge loads of LI with only minor LOD reduction.
2. Balancing the weights.
   Generally a linkset made from few, relatively large, meshes will have a lower server weight but a higher download weight than one made from more and smaller ones. Since only the highest of those two weights can count as LI, it's a good idea to find the sweet spot where the two are just about the same. Advanced weight balancing is a lot more than that of course - we have to decide
   which
   parts to include in which mesh, we have to take physics weight into account and so on. But even the basic, crude, equalize-download-and-server-weight aproach will usually save loads of LI.
3. Smooth normals.
   Can greatly reduce the number of tris needed for smoothly curved surfaces. (Smooth normals can also make a complete mess of your build so better be a bit careful and test on beta before uploading to main when you're in doubt.) Smooth normals will affect LI in other ways too but only to a very limited degree and the effect can go both ways, sometimes you gain a tiny bit of LI with them, sometime you loose a tiny bit.
4. Equal tris.
   I mentioned that download weight depends on the object's size and complexity. If you combine small details and large low poly surfaces in a single mesh, you get the worst of both worlds. The details add complexity and the large tris add size. Unless you really, really know what you're doing, it's probably better to mesh them separately.
5. Efficient building.
   Be strict with yourself: which details does your build really need and which can it do without? It's always tempting to try to emulate Real Life but that's simply not possilbe. Real Life builders seem to have unlimited LI so they can spend lots of it on even the most minute details (anybody figured out how they do that btw?) In SL we are always on limited budgets and have to find ways to use our resources as effectively as possible. There's an old C&W song, called "The Builder". Second verse starts with "Every builder knows that the secret to survival / is knowing what to throw away and knowing what to keep." (Actually the commercially released version was renamed "The Gambler" but those lines make no sense in that context and we all know the true story, don't we?)
6. Textures.
   Well chosen, well applied textures can add a lot of quality to a build, reducing the need for overcomplicated mesh. Textures can be quite laggy though so you better be a bit careful. To quote texture master Sea Warcliffe: "Use your texturing powers for good, not for evil!"
7. Effect maps.
   The old bumpiness and shininess maps and the more recent normal and specular ones are all seriously buggy and should be aproached with extreme caution. But if you're fast and if you're lucky (errr... I mean if you're
   good
   and if you're lucky) you can save quite a bit of LI by using these for some of the finer details. We have quite a few other simple and cool surface effects to improve our builds too but you don't expect me to give away
   all
   my secrets here, do you? :matte-motes-wink: )
8. The WOW factor.
   A while ago Charlotte Holmeforth reminded me of one of the essential principles for decorating on a budget and it has helped me building ever since: Use the WOW factor for all it's worth! Make it all cheap and simple and then add one or two really outstanding features! Actually, you don't really want too many stars in your show even if you can afford them - they just start fighting each other for attention.
9. Work with you materials and surroundings
   , not against them. One of the most fundamental principles of all good art and craftmanship.
10. Use all available means.
    I was lucky enough to run into Aley just when I was starting as a builder and she taught me this important lesson. Mesh is not always the best solution. There are jobs that prims can do better, that sculpts can do better and that particles can do better. Only by selecting the most suitable material for each specific job, can you build truly efficiently. Oh btw, if somebody says you can't link sculpts with mesh, take a look at #2 on this list and tell them you know better.
11. Compressibility.
    The server doesn't transfer raw data to the client, it's compressed. You can shave off a tiny bit of download weight making the mesh as easy as possible for the compressing algorithm to deal with. Not much and usually not worth the bother but sometimes a few quick and simple tweaks are just enough to flip the LI down by a point.
12. Umm...
    Seems I've forgotten what #12 was supposed to be right now. I'm sure it'll come back to me though and besides, I have to stop at some point anyway. This list could go on forever...

 

(Edits: just correcting tyops really)

---

These are all nice examples, but they are the basics that everyone will or should eventually learn. I assumed that all of this has already been done, and you are just uploading that model. At this point, if your model is outside the bounds of what would be exceptable on the market for such an item, then you have no choice but to change you LODs to fit the land impact you are going for. I think the marketplace is the big difference here. The market creates an environment of competition, which generally leads to better products. The question is, what is better? Is less prims better than models with LODs you can see from a 500 meters away? For many, they don't care if the LOD is perfect at that level. Yeah, we can disagree on the importance, but the customer is the final judge, and the sales you get from them.

[Medhue Simoni]

---

Drongle McMahon wrote:

I would look at it as a result of an extremely bad prim/land impact system, which forces creators to adjust LODs to get land impacts to a reasonable level.

It doesn't force anybody to do anything. What it does is penalise those who don't pay attention to making things efficient for real-time use. The intended penalty was incresed LI. Unfortunately, it also allowed the alternative of terrible LOD behaviour, and many took that way out as it was easier and/or more lucrative. Allowing that was the mistake.

The original intent, reflected in the size-related calculations, was that the LI cost would reflect the burden of data downloading, which uses both server and clients bandwidth. Later, the developers shifted emphasis toward the rendering strain on the gpu, but retained the calculations as the two factors were (claimed to be) highly corellated.

If I remember correctly, the LI calculation was based on a scene budget (excluding avatars) of triangles for a viewer with vertain low graphics settings (effectively RenderVolumeLODFactor=1 and FarClip=188). I can't remember the figure, but it was well below a million and over 100,000; maybe 150,000 or 250,000*. In fact, the calculation was rather generous. It effectively assumed that content was uniformly distributed in two domensions. In fact, cameras and objects are spatially correlated, and distributed in three dimensions (albeit less spread verically). Accounting for either of these effects more accurately would have increased LI and made it even more drastically dependent on size.

I thought originally that the LI calculation it was overly drastic. However, having seen the extent to which every opportunity to avoid the effort of making efficient content gets exploited by at least a few, I can appreciate that some for of control was absolutely necessary. Without it the world would be unuseable for anyone without the fastest broadband and the latest gpu (even with it, for some - almost there now for my 1.75Mb/s connection). The system adopted was at least based on reasonably rational quantitative criteria. Of course it's not ideal, but I don't think it's worthy of "extremely bad".

*the figure is hidden in the old content meeting minutes somewhere.

---

Maybe I am a bit harsh toward LL's system, as we know there must be some kind of system. That said tho, I doubt any of us would totally agree on any system. I created lots of different kind of things. Some of them, were highly detailed and very realistic. Other items, I've used just the bare minimum. I've made just about every kind of item possible, and I have yet to see the genius of LL's system, where the benefits of efficiency seriously outway the benefits of detail. You add in a marketplace, where the consumer is comparing detail to prims, and you get what we have here. People are going to ignore good LODs for lower land impact and higher detail in the original models. For the most part, the consumer doesn't care if the item looks good from that far away, or they don't understand enough to know it can be different. Chosen and I had this same discussion. He is a great modeler, but like I told him, if he was an SL merchant, he would not sell much because he'd be competing with items that are highly detailed. That is what people want. If something is supposed to be round, they don't want to see the hard edges, and you will sell less because of that.

Plus, LL completely ignored the real culprits. It's not really the items on the sims that are going to be problematic, but the avatars that play there. None of LL's system does anything at all to address this.

[Drongle McMahon]

I don't disagree. Poor choices by consumers are as much to blame as the producers who choose to exploit those choices. No difference between RL and SL there, except maybe that RL does have a few laws to eliminate the worst abuses. Still the cynical exploiters can still make the most money (ever heard of Gerald Ratner?). Fortunately for me, I'm not trying to sell anything. So I can ignore the destructive realities of commerce.

[Medhue Simoni]

---

Drongle McMahon wrote:

I don't disagree. Poor choices by consumers are as much to blame as the producers who choose to exploit those choices. No difference between RL and SL there, except maybe that RL does have a few laws to eliminate the worst abuses. Still the cynical exploiters can still make the most money (ever heard of

Gerald Ratner

?). Fortunately for me, I'm not trying to sell anything. So I can ignore the destructive realities of commerce.

---

Are they poor choices? Some, most likely are. Others, are making a conscious choice for what they think is better, for them. Just like in real life. Most of the time, even with our marketplace of millions of items, the consumer doesn't really have a choice. Yes, it seems crazy, being that there are so many items, but that is ultimately the truth. This is also why I like SL's system of allowing anyone to create things. Goto any other 3D marketplace, and try to find something specific. It's impossible to find even a few items, with the specs you need.

Recently, I was thinking about setting up a small theatre, as a place for people to watch my videos in SL on my land. When I went looking for chairs, I found a few, but they were all much more detailed that I needed them, and many more prims than I thought was needed. So, of course, I just made my own. My main requirement was that I didn't want them to cost more than 1 prim each. To make that happen, the lowest LOD is 2 triangles, 1 for the back, and 1 for the seat. These are for theatres, so no1 will ever see the last LOD. In a stadium tho, they might not look as good zoomed out.

Really, it's a shame you and Chosen Few don't sell on the marketplace, as I'm sure some of those items would be huge sellers for those that need it.

[ChinRey]

---

Theresa Tennyson wrote:

It was profiled in a blog post by Inara Pey before it opened as Meauxle Bureaux and she made it clear exactly what it was - it's no secret. The LL blog post describes it as being by "resident artists" as well. And they didn't really "copy it to the last detail" - basically, they

moved

it. All objects have the original creator's name. You make it sound as if it was copybotted.

---

Oh, I hope I didn't give that impression. It wouldn't have made sense anyway since LL has the right to use anything we upload to SL more or less any way they like.

Of course I could argue that a post in a blog hardly counts as "widely publicized" - a simple plaque at the entry point would have been far more effective - but that's beside the point really.

The point is that LL doesn't build anymore, they just buy the stuff they need and the building skills their staff used to have is rusty and way outdated. That would have been OK in itself - more than enough of independent builders in SL to fill the palce up with content - but it also means they don't understand building. We were talking about all those tricks necessary to make good mesh for SL. Who's gonna teach the builders those tricks? LL can't because they have no idea how to do it themselves.

Nor is there much contact worth speaking of between LL and the content creators who fill this virtual world. The places where dedicated volunteers spend their time teaching newcomers how to build (and everything else they need to know about SL), Builder's Brewery, Happy Hippsters, NCI, Helping Haven, Oxbridge just to name a few; those are the places where you can meet and talk to the actual content creators. You won't find any Lindens or Moles anywhere near those communities though, they are all but ignored by LL - as long as they pay their tier in time of course.

The result is that the people who build and maintain the software, the core of SL, are working out of sync with the people who put that core into practical use. It's not that LL is unwilling to assist when a builder related issue comes up. Quite the contrary, they're always eager to help. But they're just not able to see SL from a content creator's point of view so often the solutions they come up with are simply irrelevant, sometimes they do more harm than good.