Show Draw Weight for Avatars (needs work?)

[Medhue Simoni]

I'm wearing 3 items and my Draw Weight is 9250. This is nuts, if you ask me. More than half of that weight is all because I needed 3 faces on 2 of the mesh to go transparent at different times. I have on a head display thing that stays transparent all the time, that is over 1k. The other 2 items are both mesh, 1 is a holster and the other is a tool that goes in the holster. The 1 in the holster goes transparent when the 1 in your hand shows.

Today, I decided to finally take a look at the Draw Weight for Avatars. WTF!!! You basically can't wear a mesh and stay under 4000. Anything over 4000 puts you in the red. I wore 2 mesh shoes, and that almost pushed me over 4k. A mesh jacket that I bought is over 7k.

Ok, so I think, let's see what a full sculpty avatar cost. OMG, 150k, and take notice that my frame rate dropped a good 4 fps. Then, I try on some hair and that was almost 40k and my frame rate dropped a good 2.5 fps. Granted, these are extremes in the case of rendering, but not extreme for what the normal SL user wears. This shows just how terribly inefficient most non mesh items are, but I still think those numbers are crazy high, especially when you compare this to our old ARC.

Now I'm thinking, what if I load my avatar up with mesh, would my frame rate drop? I procede to pile mesh on until I reach 40k. My fps is showing almost no drop in frame rate. This does not make sense. 40k should equal 2.5 fps drop irreguardless of what type of items they are. Obviously, mesh is further being penalized for some extremely loose rendering weights, which do not reflect what is actually happening. Plus, even tho I was wearing almost 20 mesh items, they were only using about 10 total textures.

As I played around with the mesh items, trying to figure out why some were really high in Draw Weight, I noticed that the majority of all the weight came from transparent parts of the mesh. So making anything transparent will give you a major hit. If you as me, this is probably way over penalized than it should be. This is probably why my 40k of mesh did not result in any frame rate drop, cause the penalty for transparent is way too high.

Overall tho, if Draw Weight is left like this with 4k being considered laggy, then it is almost impossible to wear almost any object and stay under that 4k. I personally think all the penalties are way too high, especially considering what I see.

What do you think about Draw Weights for Avatars?

[Medhue Simoni]

OH, and just a bit more on the items that I'm wearing in the picture. All 3 items total less than 7k triangles, and I'm using 2 texures totalling about 400kbs. That's it! If you compare this to the base model, which LL is counting as 1000, then my total should really be somewhere around 2500, or 1500 for just the worn items. 9250 is just nutty.

[Chosen Few]

---

Medhue Simoni wrote:

I'm wearing 3 items and my Draw Weight is 9250. This is nuts, if you ask me.

---

To me, what's nuts is the poly count on those three items.  If the draw weight is reflecting that, that's a very good thing.

Take a look at the polygon density on that holster strap around the leg.  Now take a look at the density on the rest of the avatar.  Notice a difference?

You've got WAY too many polys in that strap.  It's a cylinder.  There's no reason on earth it needs to be so poly heavy. 

First of all, if nothing else, the strap certainly doesnt need to be double-sided, since the avatar's leg hides the inside at all times. Just removing the inside would cut the poly count in half, without diminishing the appearance in any way.

Second, realize that the avatar's leg is itself is a cylinder, that it has only ten sections around the circumference, and it looks plenty round enough.  That strap could have ten sections, too, and it would look just as round.

The deinsity on that strap in the screenshot is so high, I can't even count how many sections it's got. The wireframe looks almost like a solid object, there are so many.  That's never a good thing.

From the screenshot, I'd say you could remove at ieast 95% of what's there, and that strap would still look great.  In fact, it would likely look far better than it probably does right now, since it would be of consistent styling with the avatar body.  When the denisity is consistent, the attachment can look like a real part of the outfit, as opposed to the way prim attachments always look, which is just look like something foreign that was stuck on after the fact.

The holster looks very poly heavy as well.  It's hard to tell what the shape is from the picture, but if it's anything like a gun holster, you shouldn't need more than just a small handful of polys.  20 or 30 would be more than adequate.

As for the tool itself, it appears to be essentially cylindrical, which means all it needs to appear round is six or eight sections.  But it looks as at least as dense as strap.  If it's the shape I think it is, then it really doesn't need more than 100-200 polys, tops.

It's hardly fair to skyrocket the poly count like that, and then complain that the rendering cost is too high.  If you want a lower cost, use better built models.

 

 

---

Medhue Simoni wrote:

More than half of that weight is all because I needed 3 faces on 2 of the mesh to go transparent at different times. I have on a head display thing that stays transparent all the time, that is over 1k. The other 2 items are both mesh, 1 is a holster and the other is a tool that goes in the holster. The 1 in the holster goes transparent when the 1 in your hand shows.

---

If I'm reading you right, it's not just three faces that are transparent, but three entire surfaces.  Is that right?

In any case, that's exactly how I'd expect the draw weight to work.  If it didn't go up dramatically when transparency is applied, I'd say the numbers were unrealistic. 

Here's the thing you have to keep in mind.  Every transparent polygon requires an extra render pass.  So, in simplest terms, it's only logical that transparency should at least double the draw weight for any given object.

 

 

---

Medhue Simoni wrote:

You basically can't wear a mesh and stay under 4000. Anything over 4000 puts you in the red. I wore 2 mesh shoes, and that almost pushed me over 4k. A mesh jacket that I bought is over 7k.

---

Have you tried it with a mesh that is actually well constructed for real-time use, and not an super-poly-heavy FPS vampire like the ones in the picture?  Sorry to put it a little harshly like that, but I really want to emphasize this point.  If you want good real-time performance, you absolutely have to build your models with that goal in mind.  It's not like CGI film, or static imagery, where poly count doesn't matter in the slightest, because you've got hours to render every frame.  Everything has to render in a tiny fraction of a second here.  That means everything absolutely has to be as uncomplicated as possible.

If that mesh jacket looks anythng like that leg strap, then 7K is hardly surprising.   If that jacket has any more than a few hundred polys in it, then as much as I hate to have to tell you this, its creator does not (yet) know what it means to be a real-time 3D artist, and you wasted your money.

 

 

---

Medhue Simoni wrote:

Ok, so I think, let's see what a full sculpty avatar cost. OMG, 150k, and take notice that my frame rate dropped a good 4 fps.

---

Again, hardly surprising.  How many sculpties were in the outfit?  Mutiply that number by 2048 (assuming regular sculpties), and the poly count is probably somewhere between 50,000 to 150,000.  That's about 8-25 times the poly count of a typical character model in a game. 

 

 

---

Medhue Simoni wrote:

Then, I try on some hair and that was almost 40k and my frame rate dropped a good 2.5 fps.

---

 

Another no-surprise.  If the hair is full of toruses and cylinders, then it's probably got a poly count well into the five digits.  If it's got transparency, then it gets even harder to render.

 

 

---

Medhue Simoni wrote:

Granted, these are extremes in the case of rendering, but not extreme for what the normal SL user wears. This shows just how terribly inefficient most non mesh items are,

---

Yes, absolutely.  I distinctly remember a couple years ago, a discussion with a key member of the LL graphics team about why there were no shadows in SL.  The response from said Linden was, "Three words: prim... hoochie... hair."  (Of course, we do have shadows now, finally.  But it takes an ultra high end machine to render them without crashing.)

In other words, SL is so disgustingly poly-inefficient, rendering technologies that are par for the course everywhere else simply cannot be employed here.  That's never going to improve until and unless we as content creators start being just as responsible as the professionals who build efficient content for all other platforms.

Now that we have the same standard modeling tools available for SL that we have for every other platform, we've got a chance to do just that.  We now have an opportunity (and a responsibility) to make SL run a thousand times better than it ever has, while looking almost indescribably better at the same time, a massive win-win scenario if we all take advantage of it.  But we've also got a chance to make things far worse than they've ever been. 

The amount of polys in that leg strap of yours would be an example of the latter, no offense.  I strongly recommend you remake it, with an eye toward the former.

 

 

---

Medhue Simoni wrote:

but I still think those numbers are crazy high, especially when you compare this to our old ARC.

---

ARC was never representative of true rendering costs.  Five points for every texture, no matter what the size;  ten points for every prim, no matter how many polygons are in it; etc.  None of it was in any way accurate.  At best, it was a good idea, very poorly executed.  At worst, it only fostered the same giant problem SL has always suffered from, its users thinking about all the wrong things.

If the new, more accurate, display costs shock you, GOOD!  They should.

 

---

Medhue Simoni wrote:

Now I'm thinking, what if I load my avatar up with mesh, would my frame rate drop? I procede to pile mesh on until I reach 40k. My fps is showing almost no drop in frame rate. This does not make sense. 40k should equal 2.5 fps drop irreguardless of what type of items they are. Obviously, mesh is further being penalized for some extremely loose rendering weights, which do not reflect what is actually happening. Plus, even tho I was wearing almost 20 mesh items, they were only using about 10 total textures.

---

 

You're trying to apply anecdotal evidence in a way that can't really mean anything.  Just because under one particular condition, a display cost of 40K cost you a couple of FPS in no way means that that's now a fixed ratio that must always apply.  That's just not how it works.  The display cost of any one thing, or group of things, is only one factor among many that affects your real FPS.  There are a million other chaotic things going on all the time.

If it helps, think of it like this.  If you haven't eaten in six hours, your blood sugar is going to be low.  If you double that time to 12 hours, will it be twice as low as it was after the first six?  Maybe, maybe not.  There are a million other things we'd need to know, in order to make a reliable prediciton, many of which are inherently unknowable.  What exactly did you eat?  What activities have you performed during the 12 hours in question?  How accustomed is your body to fasting?  How's your health?  Etc., etc., etc.  All we can ever know for certain is that the longer you go without eating, the hungrier you're gonna be.  We can never just issue a blanket statement like, "You haven't eaten in X hours, so you must have a blood sugar level of Y."  It just doesn't work that way.

By the same token, just because we know there's an inverse relationship between display cost and FPS in no way means we can ever say X display cost must yield Y FPS.  All that can ever be said for certain is the higher the display cost, the worse the performance.

 

 

---

Medhue Simoni wrote:

As I played around with the mesh items, trying to figure out why some were really high in Draw Weight, I noticed that the majority of all the weight came from transparent parts of the mesh. So making anything transparent will give you a major hit. If you as me, this is probably way over penalized than it should be. This is probably why my 40k of mesh did not result in any frame rate drop, cause the penalty for transparent is way too high.

---

Again, in simplest terms, it's technically twice as hard to render a transparent surface than an opaque one, so it correlates perfectly that transparent objects should carry a much higher display weight than opaque objects.  But how exactly that one factor will combine with all the other factors that affect FPS at any given time, no one can reliably predict.  Once again, all we can say for certain is that the harder it is to render a scene, the longer it's going to take.

 

 

---

Medhue Simoni wrote:

Overall tho, if Draw Weight is left like this with 4k being considered laggy, then it is almost impossible to wear almost any object and stay under that 4k. I personally think all the penalties are way too high, especially considering what I see.

---

A statement like "4K is considered laggy" really can't be applied.  At any given time, 1K might noticeably slow things down one computer, while it might take 10K to noticeably slow down another.  At another time, the reverse might be true.  Just like with that blood sugar example, there are a million other things we need to know, before we can say whether any particular number might be acceptable or not.  What else is in the scene?  How many other avatars are around?  What are the avatars doing?  What are the other objects in the scene doing?  How's your computer feeling today?  What else is it running?  Etc., etc., etc.

There's no way to just draw a line in the sand, to separate the laggy from the unlaggy.  Again, that's not how it works.  It's never about any particular fixed point.  It's about relativity, about balance, about doing as much as possible with as little as possible.  If 4K is what it absolutely takes to make your avatar look the way you want it to look, then 4K is the right number for that avatar.  However, if the same look could be achieved for 2K, by doing a better job building the models, then 4K is way too high for that avatar.

 

In any case, my avatar has all of 16 small prims attached to it, and its display weight is 4096.  Attachment-wise, it's practically naked.  So, yeah, it's proably impossible to stay under 4000, with ANYTHING attached.  I'm not concerned.

 

 

---

Medhue Simoni wrote:

 

What do you think about Draw Weights for Avatars?

---

I think it's about time LL has finally started encouraging users to think about how these things actually work, rather than the utter works of fiction they've always promoted in the past.  The fact is SL users are very, very spoiled by the false impressions LL has allowed them to cary for so long. 

Now that there are tools in place to more accurately present the facts, a lot of people are upset.  They just found out there's no Santa Clause, and they're understandably unsure what to make it of all.  At times like this, some children (rightly) get angry at their parents for having misled them, while others get mad at them for having told them the truth at all, because the lie was more fun to believe.  Still others just get mad at Santa himself for having the gall not to exist.

I'm firmly in the first category.  I think it's more than silly that LL willfully allowed people to think so incorrectly about resource usage all this time.  While I applaud them for finally starting to get it right, I can't help but be bitter at the lack of faith they showed all along in the intelligence level of their user base.

Had they made it clear from the beginning that prims are not equal in terms of real resource costs, and that there are plenty of other things people should be thinking about instead of just prim count, we wouldn't be having this discussion right now.  We wouldn't need to.

[Medhue Simoni]

First, let me say that I generally value your opinion. Here tho, I don't think you are analyzing this correctly. The leg strap, yes, is a bit too heavy, and I added that quite hastily. The other models are basically at their bare minimum. The tool in the hand can't really be any less polygons. The holster, without the strap, might have a few extra polys, but not many. The total prim counts on those parts are 8 for the tool, and 4 for the holster.

To say that the polygon density should match the SL avatar is not right at all. Smaller objects require a higher density. This is just a fact. Look at the density of the mouth. My models are actually less dense than the default avatar in most areas, excluding the strap. How can LL say that their base model only is 1000 and then say these attached items are over 8k. That doesn't make any sense at all. The avatar is actually using 2 more textures than my meshes. Plus, my meshes are using less polygons than the avatar. The only major thing that could add complexity is the transparent parts, which I pointed out consumes more than half the total weight.

While I agree that the old ARC did not come anywhere near realistic rendering costs, it still was a guage to evaluate the avatar. You can't say that ARC is your standard and then create something else that has vastly different results. People created their items based on that old ARC. I'm not saying that we should not show the "real" cost. What I'm saying is that SL survived with every1 wearing these lag monster items, and now LL has gone completely to the other side of the spectrum. Plus, I would say that these costs have not been tested fully. To do these costs right, you have to do thousands of comparisons. Thousands and thousands. This can only be done by a large group of people and alot of trial and error. How many meshes did LL test their equations on?

4000 is not a realistic number under the current penalty system. Being in the red should mean that your avatar could cause some1 lag or have some impact on my machine. As I pointed out, even at 40k of worn mesh items, I still saw no relative impact. Maybe, if you base this number on there always being a half dozen other avatars around, but that still doesn't make a whole lot of sense. I am using a decent machine right now, but I do have lower end machines to test this on also.

The bottom line is, that LL has basically made their new Draw Weight irrelevant. No1 is going to use it as a guage for anything. People are not going to just stop wearing anything, and that is basically what you have to do to stay under their limit. If anything, people will ignore the LL 4000 and just make up their own that is more reasonable. Personally, I want people to have an accurate guage for their avatar, and I fear what LL has released goes WAY, WAY overboard, and does not reflect anything close to reality.

I would encourage you to keep testing this system. Although, I pretty much agree with what you are saying, but you are making assumptions. Test this Draw Weight system on items that you or an everyday SL user are actually going to use. Things like realistic looking weapons and combat systems, interactive items that make SL fun. I looked at the jackets mesh when I bought it. The polygon density is quite reasonable for what it is. There is no reason it should cost 7k when LL gives their own base model a 1k score.

[Kwakkelde Kwak]

I'm not going to give a full reply on this, it will be more fun to let Chosen do that...

One thing I have to say though. The fact LL only counts 1000 for the avatar is probably because (unless you want to kill your entire 5 year wardrobe) you have no choice. I would even say they could set it to zero. This doesn't reflect the costs, I know, it's a formula and you can't state 2000 is twice as good as 4000, all you can say is the lower the better.

 

BTW don't get me wrong, I fully agree a 10 faced cylinder around the leg wouldn't look good just because it matches the SL avatar... Objects in SL, well most of them anyway, can be seen as stand alone items, not as avatar extensions, even when worn. Your objects do look awfully dense though, could you post close ups maybe? both in mesh view and rendered view...

[Medhue Simoni]

I will happily post whatever people what me to post, but this is not the point. The point is not what my meshes are getting, it is what all items are scoring. I have dozens and dozens of meshes that I've made. All of them have different properties. Currently, I have mesh that are high poly and low poly. I'm looking at it from the overall perspective, not why these items are high. High poly and low poly are relative tho. A high poly gun in the 3d circles could be 30k polygons, but in SL, 5k polys is probably considered high for a gun. At the same time tho, you need at least 1k in polys to get a decent looking gun with any detail.

The tool in the hand needs detail, but I did not waste too many vertices. The tool is round, and I reduced this as far as I can go without it looking too block. It is at around 3250 triangles, and has 4 buttons, a gem, and clamps that open and close. Oh and the texture work is not really done for these yet.

[Kwakkelde Kwak]

I really replied for the 1000 points for the standard avatar.

 

---

Medhue Simoni wrote:

A high poly gun in the 3d circles could be 30k polygons, but in SL, 5k polys is probably considered high for a gun. At the same time tho, you need at least 1k in polys to get a decent looking gun with any detail.

 

---

I'm not the polypolice, nor do I want to be:) But 30k polygons can not easily be justified I think. It depends on a great number of factors though. Some you have control over as a builder, some not. You can set the lower LODs to keep the pressure on the servers and graphic cards within normal proportions (whoever decides what those are). I have never built for games, but I really doubt the guns in shooters are 30k polys, that sounds like art, movie or stills numbers to me, as Chosen mentioned, you can let your computer do the rendering in an hour, a day, a week, not in a fraction of a second. As I told him, I don't think you can compare SL with a game one on one, for various reasons, but SL is certainly closer to games than to high poly modelling. There has to be a balance between how detailed the model looks and how much resources it takes, I'm not the one to tell anyone what that balance should be. LL is, to a certain degree, other than that every builder or user should decide for themselves what is acceptable or appropriate, taking in mind that what they wear DOES affect everyone around them. Do you want your model to look good when playing wargames or do you want it to look good on display? The two would result in different models I think. Furthermore if the object gets too detailed, it won't match anything around it, so all one accomplishes is making everything look bad, by making something nice... how odd is that?

 

EDIT I see you edited in the picture, thanks:)

I'm not suggesting you should rebuild the entire thing, but I see a couple of things that can easily lower the polycount.

The strap for the holster you already agreed on, 1024 triangles is a bit much. The holster itself looks like it has got the entire inside modelled. I don't think that's the way to build it.

Purple arrow..same as the holster strap, I think you won't disagree plus it looks like a bunch of faces are buried in the model, invisible unless you rotate your camera inside it.

Green arrow..looks like you have one straight edge divided into several pieces, that's not nececary, joining the vertices won't affect the shape at all.

Orange arrows..it's hard to see, but on the right arrow you have the flat circular piece with one vertice in the center, this uses half the faces of the similair plane by the left arrow.

Blue arrow..Not sure, but it looks like you have hidden faces there.

Circle...well I think even within a high poly model this is a high poly part, especially since most of it is partially hidden.

 

I'm sure there's more, but they are probably similair to the ones I described.

Anyway, i don't mean to put a "bad model" label on your object at all, just trying to show where things could be improved. I'm sure if I made an intricate object like that I'd have a bunch of parts that could be improved, so don't take it the wrong way.

[Medhue Simoni]

Transparent items in a mesh take a very serious hit. Without anything being transparent on the tool, it's weight is 1100. With 1 SL face being tranparent, the weight jumps to 4500. That is illogical.

[Medhue Simoni]

This is hilarious that people are trying to explain to me about low poly. If no1 has noticed, I've been the sole person trying to explain why sculpties are so bad. I've been the 1 in the merchant forum getting blasted because they all think high poly statues with crazy detail is what people want. Please, just spent a minute trying to convince some newbie merchant why their 500 prims sword is not a good use of resources. I've done it over and over.

[Kwakkelde Kwak]

---

Medhue Simoni wrote:

Transparent items in a mesh take a very serious hit. Without anything being transparent on the tool, it's weight is 1100. With 1 SL face being tranparent, the weight jumps to 4500. That is illogical.

---

That depends on the actual number of  faces affected, to me it does sound rather high though indeed.

[Medhue Simoni]

It's about 400 actual faces that go transparent, but 1 SL face.

[Kwakkelde Kwak]

I think the 1 SL face makes no difference whatsoever, it shouldn't... the 1100-4500 for 400 faces sounds more like a bug to me than just being illogical. JIRA?

[Medhue Simoni]

[Medhue Simoni]

Of course, any1 can look at a model and say they could reduce this or that. The pink 1, the loop, yeah, I do wonder why i went so overboard there. Other than probably the holster strap, I'm not likely changing it tho. I think it is a reasonable model for the detail that I want. Thanks for the input tho. I do tend to so overboard on the things i want really smooth. It is likely, that if these number stay the same, I will make different choices when modeling something similar.

[Kwakkelde Kwak]

The thing is not "yes I can reduce this or that" but "hey I can do without this data without changing a single thing about the looks. You say you have to make about 400 faces invisible.

As far as I can see it's only the four flapping pieces, I made a duplicate and it costs me 4x64 faces, which is only 64%. Looks, exactly the same. This is what I ment. I can't argue on the balance of detail vs resources, but it makes no sense to me to waste so many faces on...well nothing but building convenience really. But that's just me, if you can build items that people like and make twice as many because you cut some corners....maybe you are better to the community than I am:)

[Medhue Simoni]

Well, maybe if you don't plan on reducing the model further. I kind of thought that was what LODs are for. Not every1 is standing 2 meters from the object. I'm not the kind of creator that works on a model for a week, especially when you are selling it for 50 cents. I have thousands of things I want to make, and hundreds of things lined up. Plus, I'm really an animator, not a modeler. Animation pays the rent and always my main focus. And believe me, it takes alot longer to make animation than a model. The models, I made because I need them, and I need them to do specific things, in a specific way. No1 else can make it for me, cause even I don't always know how the final product will work. The majority of times, I have an idea and a few hours to bust something out to do some testing with. If I waste a hundred verts, I'm not that worried about it concidering what is currently considered efficient in SL. Overall tho, I doubt any1 can say my mesh are inefficient.

But.... any modeler worth a lick should and can say that they can get the same thing with less verts. I didn't post my concerns to see who could be the best at reducing models. I asked if the Draw weights seemed reasonable and I would appreciate it if we could focus on that.

[Kwakkelde Kwak]

---

Medhue Simoni wrote:

[...] And believe me, it takes alot longer to make animation than a model. ...]

---

I do some animating and it takes lot less time than my modelling... that doesn't mean I am a faster animator or a better one or a slower builder or a worse or better one. All it means is I put more time into my modelling, you put more time into your animating...I'm sure your animations are a lot longer and more complicated than mine.

 

---

Medhue Simoni wrote:

[...] If I waste a hundred verts, I'm not that worried about it concidering what is currently considered efficient in SL. Overall tho, I doubt any1 can say my mesh are inefficient. [...]

---

I think it's pretty inefficient:)

But all kidding aside, I think now that we have a tool that lets us build efficiently, the current rediculous standards, where torusses and sculpts are the norm, should be canned and we should take advantage of the new possibilities. That doesn't mean everyone should dig and dig for weeks on end to find that last wasted vertice or face ofcourse, I agree. I rather see a lot of beautiful new meshes than a couple of beatiful ones which are slightly or even quite more efficient.

 

---

Medhue Simoni wrote:

 

[...] I didn't post my concerns to see who could be the best at reducing modelsI asked if the Draw weights seemed reasonable and I would appreciate it if we could focus on that.

---

I know and I already said the 1100-4400 increase or whatwasit sounds high for making 400 faces invisible, test it on some other items and post a jira I'd say, this doesn't sound right to me at all, no matter how your model looks or is constructed.

 

EDIT sorry for hijacking your thread, Chosen and I already stole another one..I'll try and stay there with the nitpicking about verts and tris and efficiency and tricks and all...

[Drongle McMahon]

The wiki says transparency multiplies the "per prim" weight four times. That explains the increase you describe. What happens if you make the pieces that have tp receoive transparent textures a separate mesh (prim) and link it? Then, if I understand it (the wiki is a bit ambiguous)  the multiplier shoud apply only to that tiny prim and that could save you a lot.  If that works, the general principle would be to always make transparency-animted parts sepatate objects, not just separate materials. Also, the LOD distance for that part would be smaller, which reduce the per-triangle cost (probably negligible).

Of course it would be nicer if the multiplier were applied at the submesh (=material/SLface) level, so that such machination was not necessary. I suppose it doesn't because they are essentially using the download cost function here and that doesn't distinguish submeshes. Changing that would not require a huge amount of work, I think. You could make a feature request jira for it.

Having said that, another possible anomaly occurs to me. The download weight algorithm uses the object size (radius) to determine the LOD distances (that affect the triangle count estimate). Attached meshes switch LODs at the same distances as when they are rezzed, but rigged meshes switch at distances synchronised for the whole avatar. I wonder if that is taken account of when the render weight is calculated for rigged attachments?

[Medhue Simoni]

Thanks Drongle! That seems to explain much, as I have other weapons that use similar techiniques, but are separate meshes, rather than just another SL face. It is easy to change on the model, I'd have to create a simple physics box for that too. I will probably take both the open and closed clamps and put them on 1 mesh. It's in the work on the coding that I dread, that makes the action happen, lol. It seemed so much simpler when it is all 1 mesh and you are just changing the face. For the gain in Draw Weight, I gotta do it tho.

Man, If things weren't complicated before, they seriously got alot more complicated.:smileytongue:

[Medhue Simoni]

---

Kwakkelde Kwak wrote:

---

Medhue Simoni wrote:

[...] And believe me, it takes alot longer to make animation than a model. ...]

---

I do some animating and it takes lot less time than my modelling... that doesn't mean I am a faster animator or a better one or a slower builder or a worse or better one. All it means is I put more time into my modelling, you put more time into your animating...I'm sure your animations are a lot longer and more complicated than mine.

 

---

i

I can do any pose in a few seconds, but doing motion capture can take an hour to a whole day or days for 1 motion capture. Setting up, practicing the movements or hiring models, getting into character, recording the movements, setting up the scene in the program, processing the movement, processing more detail, fixing and smoothing, and then I can bring it into the actual bvh editor to loop things, clean more shakyness, and get it ready for SL. Then, I have to adjust that animation on the SL character, which there are no 2 the same size.

[Kwakkelde Kwak]

Exactly what I said....

 

Professional modelling (something which isn't my job btw) works the same, except it can involve airports and hotels..multiple times....

[Medhue Simoni]

After looking at the wiki, I see that it is all tied to triangles:

http://wiki.secondlife.com/wiki/Mesh/Rendering_weight

I was hoping that triangle counts would play more into mesh, but I did not really expect it this way. I still think that some of the settings are wacky. So does the display number have anything to do with prim counts, or land impact? Cause it really seems to me that you can get away with alot of extra triangle as long as you get your Lowest LOD low enough. Or, I guess this affects the Display also. My mind is melting.:smileytongue:

 

[Kwakkelde Kwak]

No you are 100% right with the LOD, as you mentioned earlier already if I recall correctly. It all comes down to data processing cost. When you have your lowest LOD set to next to nothing, there's next to nothing to send to avatars and next to nothing to render for those same avatars, I bet you already grasped that.

It's not a complicated, but a very big formula.

The number of triangles make up the base cost of an object, but only at the LOD they represent, so the distance multiplier for LOD high is nowhere near as big as that of LOD lowest. So indeed you can "get away" with a huge amount of triangles for the highest LOD. The thing is: What do you want your model to behave like? Do you want very smooth transitions between the LODs? That means the detail will have to gradually improve going into higher LODs. Or do you want to make an object as detailed as possible when up close? Then you scrap the lower two LODs and can use enormous amounts of faces per prim equivalent. One way isn't neccecarily better than the other, they each serve a different purpose.

One thing remains though, the less you use, the less it costs.

I think you've struck something quite interesting, I don't understand why LL decided to make the multipliers multipliers by prim, maybe easier coding and faster processing, but it doesn't represent the cost at all, if I read the wiki correctly not only 400 of your faces, but a single one would quadruple the draw weight. Now that is just stupid:)

Oh forgot to answer your question..duh

---

Medhue Simoni wrote:

So does the display number have anything to do with prim counts, or land impact?

 

---

Yes. It does. Everything I wrote also applies for Prim weight. As they say in the wiki, the number you are working with is based on the triangles, the Prim equivalent has one more multiplier (a very small one ofcourse) and is directly linked to the triangles.

 

This is what I make of it:

From http://wiki.secondlife.com/wiki/Mesh/Mesh_Streaming_Cost

 

• Dlowest = R / 0.03
• Dlow = R / 0.06
• Dmid = R / 0.24
• Dhigh = 0.0

These represent the LOD transitions (in one direction) The triangle count used for the draw weight uses the area though, so rather than in one direction it's measured in the area of a circle. ( π r²) So this is π * D[LOD] * D[LOD] 

removing a single triangle at LOD low allows you to add 64 in the highest then.

I'm missing something though, since the maximum radius of LOD lowest doesn't exist.. does the viewer use 256 meters as a maximum then?

[Drongle McMahon]

"...does the display number have anything to do with prim counts, or land impact?"

Well, that's the way I read it, for that component of the cost. It says the same calculation is used as for the download weight, except the final conversion that introduces the region triangle budget. At that point, it is supposed to be an estimate of the triangle count, corrected for what proportion of random observers will see at different LODs.

In fact, last time I looked, for mesh the triangle counts are just estimated from the download data size of each LOD mesh, assuming a constant number of bytes per triangle. As we know, the bytes per triangle can vary a lot with differences in sharp edges, uv seams etc., as well as different compression ratios. So the relationship with actual triangle count is not very constant.

However, it was said** that this was probably a better representation of render load because the sharp edges etc. did increase the rendering engine load as well as increasing the download size. In other words it may be regarded as a sort of effective triangle count, adjusted for those factors.

It will never be possible to have a perfect measure*, not least because of differences between gpus and opengl implementations. So I think this may be as near as they can reasonably get.

*without a real-time profiling, which would kill everything.

**I think this is in an old thread somewhere, but I can't find it. So I leave it anonymous.

[Medhue Simoni]

Well, this does shed some light on things. I bet I could lower the display cost more by simply lowering my medium and low LODs more. Definitely gotta play around some more and look at it from the display cost perspective.