Distorted Sculpt

Thanks for the explanation. Sounds like a good looking build.




Whether or not it's a good idea depends on your goal. If you're thinking purely in terms of best possible performance, then yes, it's a bad idea. If you're thinking only about aesthetics, though, then it's great idea. If you're thinking about both, then the answer is maybe.

My only point in all this is that people should weigh all the factors when making decisions. Educated, deliberate choices are what it's all about.

Chosen, firstly why do you have me quoted for things that weren't said by me? And for the record, I don't like to argue for no reason, I just don't like misinformation spread by someone with a hamfisted grasp of a subject I know well.

As for spoon feeding you a how-to, sorry, you'll have to teach yourself, I'm given more than enough info for people with more than a rudamentary grasp of how sculpties work the basis to go about doing it themselves and if this post sounds arrogant, it's in direct responce to your know-it-all attitude towards a subject you don't fully grasp. I have a great deal of respect for your work, but when it comes to optimisation, or the simplest oblong techniques, you haven't the faintest idea of what you're talking about.

@Chosen and @Bonk: Can it be, that you are missinterpreting each other ?
I have the strong feeling, that both of you have good and correct opinions from what i understand. Maybe the truth is somehow depending on the viewpoint ?

What i understood so far is, that Chosen points to the fact that sculpties allways have 2048 tris/triangles, while standard prims typically have much less. He says: That is evil when you think in terms of many hundreds of sculpties due to summing up of triangles in the graphic card. Now here is my question: How can you measure the effects ?

I further understood from Bonk that as long as sculpties are not visible (hidden from other prims) their meshes do not add to graphic rendering (due to culling). So is it so, that as long as 50 sculptie doors are hidden inside a star ship, their existance can be neglected ? But as soon as all sculpties come into view, the graphic card starts boiling up (like Chosen has pointed out with this giant sculptie tree...)? Is that correct too ?

Chosen was so kind and posted a table of triangle counts for each standard prim, see

/8/81/315464/1.html

I have taken a few minutes and found a way how to make these triangles visible in the SL viewer (just use Advanced->Rendering->Wireframe and you can see them) So now i/we can immediately see the differene in terms of triangles. But what does that mean to us ?
I still have no idea how much of degradation i get when i use sculpties instead of prims.
I think, Chosen has given a theoretical answer but what can we learn here ? Can this be elaborated to more detail ? :



Finally, without wanting to get deeper into your argumentation, maybe Chosen has made a simple cut/paste mistake when quoting Luke's questions ? That was clear to me at the first glance ... So maybe things can boil down again, if Chosen just edits his contrib ?

Just my 2 cents here hoping it helps a bit ?

Gaia

No, I'm saying you don't have to use every last vert on a mesh to make an object.

Whoops, mistake with the old cut & paste there. I've corrected it.



You still have yet to dispute any facts I've posted. What you've done so far is four things. First, you mistakenly assumed I was talking about something other than what I was actually talking about (NURBS vs. polygons). Then you complained about the length of my posts, pretended I was somehow trying to keep people out of SL, and then made an absurdly outlandish claim about what you could make from a single prim. If your goal was to correct "misinformation", where's the correction? Tell me, what specific "misinformation" did I spread?




I never asked you for a how-to. Where are you getting this stuff?



Once again, if you feel something I said is incorrect, explain exactly how. Otherwise all you're doing is hurling insults around, pointlessly. If you've got facts to contribute to the discussion, let's hear them. With the one exception of the alternate construction technique you offered to the OP, you have yet to contribute anything on topic to to this thread at all. Sorry but just saying "you don't know what you're talking about" doesn't mean anything. Back it up with facts, lest your comments remain irrelevant.

As for my alleged "know-it-all attitude", I've invited you at least three times now to explain what you think it is I don't actually know. Clearly if I felt I "knew it all" I wouldn't do that, would I? Again, if you've got something factual to contribute, go ahead.

I suspect part of what's got you riled is the fact that I didn't respond to your comment about how collapsing vertices can possibly reduce the poly count of a sculpty. If that's what it is, you're right, I should have addressed that. As far as I know, there's never been any hard evidence made available one way or another on that. Some people say "I looked at the code, and I know those triangles don't get drawn." Others say "I looked at it too, and I know they DO get drawn." Since I'm incapable of understanding the code myself, and since I've never gotten a good answer from a Linden on the subject either, I've got no way of knowing what the truth is. And since I can't make recommendations based on a maybe, I don't go around suggesting to people that sculpties can have variable poly counts (or that they can't). I only deal in what I actually do know to be true.



Gaia, thanks for the attempt at playing peacemaker. I hope you can straighten this out, because honestly, I'm not sure what's got Ponk so upset as he seems to be over this.

I'm not annoyed in the least, I just find it a bit tedious that you're attempting to field all questions with no means to answer them fully. As I've mentioned before I've got no intention of explaining, there are some things you'll just have to learn for yourself. Starting with oblongs, which you've stated before you have no interest in for your own perfectly acceptable reasons, the rest of what I said will fall into place once you play around with them.

As for 1 dimensional faces/verts, I doubt they get drawn in the typical sense because they don't have any visual parameters. I might be wrong in this, but I can make a single prim go a lot further than most, so a ship, a house a car an entire sim build is all the same to me and it'll have a lot less performance load than with regular prims.

The thread about trees will cover all aspects of tree creation, of which i will get knowledge about, so as the thread name allready says

"About the principles of making trees"

will also cover making trees out of standard prims. As far as sculpties are involved, of course i will again go with blender. And as far as this thread here is involved, i eagerly want to learn from here, what relavant information i can use for making better trees regarding to resource usage

So lets get to practice and think a bit about resources:

1.) Assume you make a tree with 3 crossing prims (flat cubes, no trunk for simplicity), then you will end up with 6*18 triangles = 108 triangles.
2.) Assume you make the same tree out of one single sculptie, then you end up with 2048 triangles.

conclusions (first part):
- From counting the triangles, sculpties feel much less optimal.
- From counting the prims, i can make 3 times as many sculptie-trees on my land, which sounds not too bad.

Now we shift from knowledge to religion :

- If we believe that collapsed mesh points (aka collapsed faces) are not drawn, we can get away with only 24 triangles when using sculpties compared to 108 using prims so sculpties would be 5 times better.
- If we believe, that collapsed mesh points contribute to the face count, 3 prims are 20 times better then one sculptie.

Conclusion, secnd part: If i calculate correct, we do have a factor of 100 here depending on what we believe. So for me it is absolutely evident to find out the truth here, don't you think so too ?


ok, a little further thinking: What happens when i add textures...

for both construction types i can create one single texture of size 512*512 and partition it into 4 parts (of which i would use only 3) and place it on the prims (ok for sculpties i might need one extra baking loop to place the textures at the correct position, apologize here, this is not well thought, maybe it is easier but anyways...). So from the textures viewpoint both technologies seem to be equal weighted.

Now what ? For me 2 questions are evident:

1.) how are the resources distributed among mesh and texture ? If texture outweighs mesh by far, we might simply forget polygon counting... So what is true here ?

2.) Which religion is the "true" religion ? "collapsed faces are drawn" /vs. "collapsed faces are skipped". A factor of 100 sounds like we should investigate to find it out.

I hope, what i write makes a bit of sense. Otherwise please feel free to correct me and guide me/us to the true knowledge (believal?)

If you find it tedious, don't read what I write. It's as simple as that.

As for whether I've answered "fully", OF COURSE I can't do that. No one can. There's never a point at which there's nothing more to say on a topic. That's called life. Once again I invite you to add whatever you feel is missing.



OK, so it's "you're wrong because I said so", and that's that. Got it. Who was it again you said wasn't answering fully?



Just because I recommend against using them, and I avoid using them in my builds doesn't mean I've never used them or that I don't know how they work. You'd do well not to be so presumptuous. I know you don't want accept this, but the fact is I don't comment on topics I don't know about, unless it's to ask questions of my own. When I'm answering other people's questions, it means I've done my homework.

Look, there's nothing inherently different about oblong sculpts than square ones. It's just a matter of proportion.

In any case, I'm not sure how the topic of oblongs even came up. If you're suggesting that the OP's door should be made from an oblong, how about saying so? Otherwise, it's not even relevant to the discussion.



I think you're misunderstanding what I meant by the word "drawn". I'm not talking about simply occupying space on screen or not. I'm talking about how various pieces of data are processed. You make it sound as if you're attempting to apply real-world human logic to the internal workings of computer functions. That's a mistake. Just because something might be invisible doesn't mean it doesn't get drawn. There are lots of items that have no visual parameters that still get processed in any rendering pipeline. It's a question of mathematics, not of vision.

Oh well then, my mistake.

Yes. That's one I have wanted to know for a long time. I did look at the code and found something that took them out of the triangle list before it was sent to opengl, but I couldn't find anywhere that code was called! I asked Qarl Linden and he didn't seem to know either. Anyway, opengl has another chance to prune them too before it does its work, but someone else said they actually slowed opengl down! For now, let me advertise the forthcoming small-map-sculpties which will certainly have less triangles.

Gaia, what you said makes perfect sense. Yes, the answer to the drawing question is certainly worth finding out. But after having gotten so many conflicting answers to the question myself (even from Lindens), I have no idea how to find it.

Drongle, thanks much for the input. I've heard that same thing before about the part of the code that snips out the zero-dimension faces. Some people have said it's there. Others have said it's not. If it's indeed there, but it's not being called, that would probably go a long way toward explaining the back-and-forth on the subject.


One graphics expert I know told me that those collapsed faces actually increase, rather than decrease, the rendering overhead. I didn't quite understand his full explanation, since I'm not the mathematician he is, but he was quite certain. One part that I did understand was this. When you have a face with zero dimension, where's the normal? A lot of 3D engines will crash instantaneously when confronted with that question. He said the fact that SL doesn't is actually quite impressive. The math to try to answer the question is still being done, though, and the way SL interpolates normals for sculpties makes it even harder. He said the processing overhead increases by as much as a factor of 10 for each zero-dimension face. That's significant.

But again, that's just one explanation from one person, and even though he's a bona-fide expert on graphics, whose expertise I trust and often depend on, I can't operate solely on the basis of one person's commentary without any confirmation. I'm more inclined to believe my friend than total strangers on forums, of course, since I know his credentials. But still, there's been so much "yes/no/maybe" back-and-forth on this subject, it's really hard to know what to believe.

I've heard the opposite from people who work in the games industry coding graphics engines, so apparently it's horses for courses.

It's a huge non-issue anyway, the performance hit mentioned is negligible at best.

It's only negligible until you've got a ton of them in view, which brings us right back to where this discussion started.

Yes, which was nothing to do with the question posed.

I would like to say that for what it's worth, just because you can make something with a sculpty, it doesn't mean you should. Just stick with a regular prim wherever possible. It winds me up no end when people whittle away at a sculpty for something that can be done in 5 seconds with a regular prim.

Yes. No doubt it varies from engine to engine and from version to version. When all else fails ... I might have to have a go at doing some experimental investigations, if I can think up how to .... results only valid for one viewer version on one platform, of course, so a waste of time but ....

It didn't directly have to do with the question as worded, no. That's why it was presented simply as a forward to the answer, not as the answer itself. I thought I had made that perfectly clear at the time.

It's not uncomon or unreasonable to say "Sure, I can tell you how to do that, but before I do, let's make sure that's actually what you want to do, because there are some drawbacks you should be aware of." It should and does happen all the time.



We're in agreement on that.

I've learnt something here: the question of whether or not collapsed vertices get drawn needs a firm answer.

I would sculpt the living daylights out of my place if I knew for certain that collapsed vertices do not get drawn. For instance, only somewhere between a quarter and a third of the available mesh is visible in the windows I mentioned earlier in this thread. A back-of-the-envelope calculation suggests the meshes I created would have roughly the same polygon count if I were to construct them from cubes.

Looking at both versions in wireframe, the sculpted version is still a little richer in polygons than its object-based fellow but not markedly so - and it looks a hell of a lot better both in terms of aspect and proportion.

However, I am inclined to err on the opinion that collapsed vertices do get 'drawn'.

Chosen's point about mathematics as opposed to the simple issue of visibility is very relevant here since the collapsed vertices in my windows have a specific purpose in maintaining LOD. The Lindens no doubt employ a little gnome specifically to work on the sums for that every time someone cameras in on my house.

Nevertheless it would be good if we could have a JIRA that gives a firm answer to this issue (I'm not clear if that's within the remit of JIRA). If no such JIRA exists, could someone with sufficient understanding of rendering engines set one up? I could only find the following when I combed through searches for 'render vertex' and 'collapse vertex draw' and 'collapse vertex' and 'collapse sculpt vertex':

https://jira.secondlife.com/browse/VWR-2248

The triangle rendering count in the Statistics Bar that Feynt Mistral referred to by its shortcut (Ctrl+Shift+1) is interesting but I don't see that in the Statistics Bar of the current viewer. The thread is from September 07 but I was a newbie then - can any of the more experienced long term residents recall whether or not this was a feature that is no longer available or if I am missing something?

I'd be fairly interested to find out as well, since we're on the subject we might as well get a definitive answer.

I must say though, that there are quite a few examples of full sims that are completely sculpty centric and appear to perform a great deal better than similarly created prim sims. Aminom's Sculptomancy sim and Black Swan are two which spring to mind, but there are a fair few RP sims whose names escape me with masses of sculpties which don't even cast a thought to optimisation and they're no worse than any mall or large store in terms of lag.

Well, I did my first experiment. Made a sculptmap mostly grey but with pixels to make a LOD1 shape with 16 visible triangles (all grey is recognised and rendered as a sphere). All the other triangles are zero in all dimensions. Went to beta grid, up to 200m on a platform, turned off all except simple and volume under Advanced|Rendering|Types and rezzed 20 sculpties with this map. With only these visible the triangle count was 51.3K, just about 2048 per sculpty. Zooming away to LOD2, only one triangle visible per sculpty, count fell to 12.6K, one quarter. That means the count is the number after generating the sculpty triangle list (but could still be culled before passing to opengl?). So, at least at the point the triangle count is done, the zero dimension triangles are still there. My guess is that this means they ARE passed to opengl.

It's still there in viewer 1.22 release. View the statistics bar, then click "Advanced". This should reveal the details including the KTriangles per frame and the KTriangles per second.

Thanks for the advice, Drongle. I can't get to the Beta Grid tonight but I'll try something similar tomorrow as soon as I have the opportunity.

However, I flew up 1500 metres above my place tonight where I had previously rezzed a prim and a sculpted version of the window. My draw distance was down to 64 metres since I logged in but I didn't note much difference as I deleted each version in turn. I turned away after I drew level with the examples for comparison with an otherwise empty view of the stars and again as I deleted each example in turn. I also allowed a couple of minutes at each stage and each change of view to let the readings 'settle down'.

I didn't notice much difference at any point: KTris drawn varied between 69 and 67 per frame and 4550-4800 per second all the time.

Mind you, I am thick as two short planks when it comes to the nuts and bolts of technology: bytes are things I take out of sandwiches and frames are the things I deleted in my half-arsed experiment earlier tonight.

I have always relied on the kindness of technically gifted strangers for advice in such matters.

Thanks, Drongle. It's nice to know this.

Of course, the total number of triangles is only one of many factors that influence performance. For example, one common technique for increasing OpenGL performance in some applications is to purposely add degenerate triangles to link different objects together and allow them to be passed in a single OpenGL call.

I would be interested in experiments like yours that compared two alternatives as measured by the average frame rate. For example, a comparison between a sculpty plane with no degenerate triangles versus a sculpty plane with only one or a few non-degenerate trianges would give us an indication of the overall cost of processing degenerate triangles. Some other interesting comparisons might be

* Sculpty cubes with just 12 degenerate tris versus conventional cubes.

* A more complex geometric sculpty versus one with the same shape made from traditional prims.

Also, since the results could well be differernt for different hardware, it would be nice if any tests that seemed to be revealing were packaged up so that others could repeat them on their own hardware.

Yes Omei. I guess that could be approached by recording the Ktris/second in similar circumstances, or the frame rate with a huge number of objects to make it slow enough.

That's a lot, as if it's rendering some other stuff. You need to turn everything off in the Advanced|Rendering|Types menu. Especially Character, which includes your own avatar. Then you need to make sure you are only looking at the object under study....take it away and the KTris should fall to zrero.

Thanks for reporting your findings, Drongle. It seems we're finally making some headway on this. Of course, there are still a lot more related questions to answer, and more experimentation needed, but this is a really great start.

I actually never even noticed that that triangle counter was there in the stats bar. If I had, I'd like to think it might have occurred to me to do the same experiment myself. Thanks again.