Linden Magic Math!

[Failed Inventor]

So lets talk about the magic of linden render costs with prims, vs. mesh :)

For all you die hard linden fan boys try to keep your rage to a minimal.

 

SL Prim Cube is 120 triangles, and has a cost of 1 prim.

Uploaded cube with 12 triangles ( 1/10th of the linden prim cube ) has a cost of 1.5 prims.

 

Now when you look at a regular render engine, such as ogre, unreal, cryengine, etc. Ignoring all rendering factors with focus selectively on cost of the pure number of triangles.

A uploade cube should be 1/10th the cost of an SL prim. This is both logical, and mathmatically sound. 

 

How does Linden Labs determine that the same shape with less triangles cost more to render? No more textures than sl cube, materials or other attributes to set it apart.

Additionally how does a script, code side function increase rendering graphically? I don't have to rez the text, nor does code in any manner have faces, verts, or edges. 

Additionally, shouldn't cost be stable?  How does a 120 triangle cube cost the same as a 12 triangle cube? mathmatics to me would state that an amount 10x should have a render cost 10x more.

to me the solution is pretty simple, 1 + 1 = 2 this is simple math we have used for years. 

@ Linden Labs 1  = 10 and 120 = 12!  Not to sound punny but Linden Math just doesn't Add Up!

 

 

 

 

 

 

 

 

 

 

 

 

 

[Failed Inventor]

Also did you know that a model which has 137 + 137 + 137 + 137 for all LODS which totals = 548 triangles all together has a 2x cost verses a 4000 + 2000 + 1200 + 500 for all LODS which total = 7700 triangles. .... 

since when did... a 1/14th of the total triangles ever... cost 2x the render cost?

[Gearsawe Stonecutter]

No one ever said they got good grades in math. Also as you link togetter the same low cost prim the cost goes down for each prim. Take 256 1.5 cost prims link them together you get about 256 prim cost. So the more you link the better off you are. Weird? It like you get a bonus for "instancing" the same prim

 

[Poenald Palen]

There software has to do stuff before it turns it over to OpenGL.  I assume they are using this and also taking into account the "average" or mean range of PC hardware.

To be honest, you bus speed and GPU all work at different rates AND we also have different thresholds. So this number is known, from the start, as a sort of generel guide. It can help a creator who wishes to weigh in how much impact he might have on some of the customers. It is a quick way to sort of do it, obviously user testing and getting your users to use ctrl+alt+1 and ask them all the info might help a bit more. Screen shots and checking out there settings help, or simply adding in a quick script to let them change some things. Like shiney. It is different under the newer lighting system AND that lighting system might change. But, videos, movies, and pics are all over the net with that system! So, all I can do is add a menu item to my models and send out updates. Now I need to put in a update script to handle any more future updates easier, hand sending is hindered by the new system that allows profiles to be locked from public view! When I started out I don't remember this being an option. So, things change, I have to change my product. If people use ARC, then I would have to be a bit concerned or offer an alternative of the attachment, right? Well, maybe not BUT if enough people have issues it helps.

That is why it is good to have freinds. They can test stuff for you...and you all figured you where just loved freinds of the 1000+ they had on thier facebook? They could always pay a pro, so you might still be simply loved...heck, maybe they have a great deal of love for humans in general...or furries or whatever!? I am not trying to put anyone down, but if you know people you can know the market a bit better, and ARC's math is irrelevent. It has something as a base, but no one cares about it...so whenever anyone mentions it they see it as propaganda to put them down as if they are not a "green" or community concious person. The funny thing is, SL was supposed to take up as much electricity as it took to feed a brazillian....so, I mean...you could feed a person of the joules of energy you use to power you PC, the network hardware and LL's servers etc. So, really and truly you could argue they are all evil, no ARC needed to prove anything! It is all perspective, there are so many things to consider and the total economics of energy use, what entertainment costs in joules that is NOT as simple as SL and so on....well, you can get into some long equations that have these odd bits that is where you almost make stuff up. Then you remember the phrases and qoutes about statistics and figures being lies or as good as them.

So, it is all much ado about nothing...which is why epic threads can be made out of it! If it where simple people would accept it and carry on. It if funny to see what is NOT on forums here! It is funny to try and think about what is NOT asked!

[Poenald Palen]

http://wiki.secondlife.com/wiki/Arc

I didn't think I would have to post this. But since I looked it up out of interest myself, here it is!

 

I mean, there is this to: http://wiki.secondlife.com/wiki/Culling

and maybe a few other things that are easy to find and understand about the rendering pipeline. As well as the open source code.

Of course, the point of the post is to find the stuff that is maybe NOT seen in simple wiki entries. But, since I was looking this up and it points to a few examples (someone jsut mentioned copying a prim...um, it depends what is in view maybe? I mean, if some verts are not visable because they are behind another prim....does this get culled? These are some questions I am asking..so, here are the links above might help people seeing what I as saying about "other stuff" happening with ALL of the components and also teh file...I mean, how to all those prims get stuck into a link? It doesn't happen by magic, there is a file to make this happen. And if you copy one....well, my thought is "Compression!" but everyone elses is "Stupid LL math" so....I can't really say, because maybe the linked objects are NOT compressed and I don't know this....so, I hit the wiki! This is part of my job with cars, so I am reading it over again lol.

[Feynt Mistral]

ARC should technically go down by a factor of 10 for a low poly mesh avatar (constructed of 12 tri cubes) versus a typical prim (cubes in this case), but the ARC is modified by the streaming cost as well if I recall.  It should still be a significantly lower cost.

Culling on the other hand would function almost the same if not faster with the lower poly cube.  Less triangles means less back face culling to test, but that's almost automatic now a days because the check is so easy to make in OpenGL.  The octree culling should technically remain the same because the corners are still the same sizes, so visibility of the object doesn't change.

 

With regards to the render cost, yes a prim has more polys than the one you made, but the one you made also has a fixed form.  Prims are generated programatically.  That means the cost of storing them in memory is far cheaper.  Now, whether it's 10 times cheaper or not I don't know, I haven't gone hard core graphics programmer to find out.

[Failed Inventor]

@ Poenald: you have walls of text and I can barely understand what you have typed after rereading a couple of times. 

@ Feynt: I understand and have already considered that prims are program generated, and have less storage cost.

But if you think about it, just cause an object has higher server cost to store doesn't mean its rendering cost is higher. rendering is handled client side, the server simply tells the client what to render. 

Additionally when we think of the inefficient wasted faces to build in both prim and sculpted, mesh should be hands down superior by leaps and bounds as objects can now be made by professionals without wasting a single face.

Of course if you build inefficient you will have atronomical cost, say a 2 million tri dolphin. But that is why there is polygon optimization, and those who use it shouldn't be seeing their models unfairly penalized compared to sl's inefficient and outdated prims system or the fps killing sculpt.

Sculpts were the worst thing invented, a **bleep** child, grew monster. I say this and I am the guy selling them cause there was really no alternative. We have mesh now, and SL can operate smoother, cleaner, rez faster and be close to up to date with the rest of the industry if only we don't penalize it in order to balance out for those unwilling to take the time to learn mesh and that continue to cling to sculpt and prim.

I know there are people who don't wish to learn mesh, love to create in SL, that is why there are people like myself which will make mesh kits, which will allow them to do just that. But we shouldn't penalize progress with fictional numbers about rendering cost just to balance for politics and keep prims from 2003 viable.

 

 

 

[Gaia Clary]

Why are you so 100% sure that what we currently see will become the final result ? Could it not be that it is a glitch in the software ?

I mean what is the reason to add meshes at the end ? So this is what i belive are some good reasons for mesh:

 

• optimize objects (for less lag)
• allow more acurate content (better precision)
• open SL for main stream development of 3D objects (closer to industry standard)
• ...

If the costs that we see now would be the final costs which we will get at the end, wouldn't that simply mean: Making objects with meshes will get very unattractive ?

But as i understood LL in the past, they want to fight lag and they want to increase the platform's look and feel. So i am still confident that we will see lower prim costs for meshes shortly.

Well, i might be completely wrong and the truth might be completely different. So i'll lean back now and wait and see what happens.. maybe we get something to know on monday's mesh office hours.

[Drongle McMahon]

"That means the cost of storing them in memory is far cheaper. "

Less asset server space and less download bandwith, but surely they still need the same gpu memory per triangle, and the triagles have to by put in cpu memory first (well, maybe not have to be, but I think they are ... all the triangles are made in one pass as far as I can see, both for prims and sculpties).

[Drongle McMahon]

"there are people like myself which will make mesh kits"

As long as you keep the components small, I think mesh can still work for this, as long as you accept really horrible low-LOD meshes (not as bad as sculpties though). It is the size effect on cost that is going to kill large individual meshes. People making large builds will have to make them of many small meshes if costing stays the way it is going, which will annoy them. I think this is going to shift the balance from finished product mesh builders to component users. There are a lot of people who would prefer that anyway.

[Medhue Simoni]

Looks like the same math they use to calculate rankings in the search engine.

I agree, sculpts are a nighmare compare to even the least efficient mesh.

Considering that prim cost is probably the most important aspect of mesh, I don't understand why it was not an ongoing adjustment while it was in beta. Instead we were told the whole time that these numbers would change, so no1 complained about them much. Now, we are coming up to crunch time and I believe these algorythms are bunk. Sure, I'll be fine with most of what I make, but unless the costs change, there is little reason to make a big deal out of mesh. Plus, IMHO, LL never addressed what has got to be the biggest lag producer, which is how many prims you can wear.

Yeah, I know LL said that the algorythms will change, but that does not really make me feel any better about it. Change could mean that it gets worse, which is probably more likely.

[Drongle McMahon]

The cost calculations have been changed repeatedly during beta development. This has included complete overhauls of the underlying rational. So it has been a continuous subject of investigation. The current streaming cost algorithm is quite sensible as long as streaming resource was ther overriding factor. It will be more or less similar if it represents the triangle burden. The ver real problems we have are because standard prims an sculpties are not costed according to resource use, and that results in a huge practical bias against using mesh, even where its use would benefit performance.( I have no understanding of the workings of physics engines and therefor cannot comment on the physics costs.)

[Fenix Eldritch]

Isn't mesh still in beta? Aren't the resource costs still being worked on? I've been reading the user group meeting notices each week and last I saw, nothing was nailed down yet - and probably wont be right up until the very last minute.

No need to get too wirried about costs just yet.

[hibit Spad]

If you go through the group meeting notes you will see where the Lindens said they want mesh cost to be somewhere between a sculpt and a prim such that houses were not going to be made with mesh but detailing in the house would be. As soon as I saw that I shelved several projects that I had been working on. Basically LL does not want to upset their billing scheme by introducing something that lets people cram more stuff onto a lot. So don't expect any kind of major change in the prim costs from what you are currently seeing.

[Medhue Simoni]

Sorry, I only saw a couple of prim count changes during the beta. I mostly played with mesh for the first 3 months of open beta, and then I had to get back to making products to sell. I only came back to the beta grid to check on changes that I read about in the Office hours.

As I was saying on another thread, on current prim objects, some1 can use dozens of textures to create a build. Although some1 can use as many textures as they want with a mesh build, in practice, mesh creators tend not to overload mesh with many textures. It is just not the nature of the process to make more textures than you need. The only time this factor displays itself to the public, is in the ARC. The only way a consumer is going to know that a prim object uses 20 textures is by detaching and attaching things while their ARC is displayed. With buildings and things you don't wear, there is no real way to know how many textures make up a build. If I were LL, I would add a texture amount section to the specs of objects on the marketplace.

[Drongle McMahon]

Ah yes. The texture thing is a whole extra story. I think the biggest difference, which applies equally to non-mesh prims, is the cjoice between using small simple tiled textures, which can be re-used on different shapes, and custom (often baked) textures. The latter generally have to be large to give sufficient detail on large objects. More detail can be gained by using all (up to 8) available materials. Tiling re-usable flat textures requires special ways of making the UV map, but then they can be small and used on multiple different shapes.

Thus we might find on the one hand a castle, for example, using ten different mesh components, each with eight different baked 1024x1024 textures, and on the other hand the same build with all the components using one 256x256 tiled stone texture. The former, 800,000,000 pixels, has much better overall appearance, while the latter, only 64,000 pixels, may have more detailed surface in close-up view!

Sculpties are somewhat different because they can only have one texture,but lack of UV map control gives a bigger incentive for custom textures, but otherwise this applies equally to ordinary prims and meshes. With ordinary prims it may be worse if custom textures are used, because more are used for the same complete build. However, the preparation of custom textures may be much more likely to be used for mesh builds because it is part of the 3D software. With normal prims, it requires special effort outside of the building tools.

It will be very interesting if/when we can see the texture resource information for objects and compare it with the other resource loads. I hope that's what the resource cost tool will do. Meanwhile, I compared my gallery (Mesh City) with my underground pool (Mesh Sandbox 15). I'll use a texture size unit of a 128x128 = 1 unit. The gallery is mostly textured with tiled repeats. These come to 61 units. There are two uses of custom textures. The windows and skylights have alpha textures for low LOD. These come to 37 units. The doors (for no good reason) use baked textures and come to 85 units! Now compare that with the much smaller underground pool (excluding the grass mound). This is one mesh with six faces which all have baked textures. The UV maps use stacking where possible to get better close-up resolution, but you can still see it's fuzzy close to, compaed to the bricks on the gallery. These baked maps come to 320 Units !!

Here is a simple example. The same mesh steps with a baked texture on the left, and with a general-pupose tiled texture on the right. That's 1024x1024 (64 Units) vs 256x256 (4 units). I can't resist using the baked texture. Is it worth the difference?

[Medhue Simoni]

Personally, i think baking it is worth it, but of course there is some room to haggle here.

First, let me say that I'm learning just like many people here. Please, do not think I'm trying to put any builder's work down, but I do think every1 would benefit from a Best Practices section for all creating sections in the wiki.

A resource cost tool that takes all resources into account would be very much appreciated. If this works well enough, I'd really like to see the results in the build window so that every1 can see it.

[Ashasekayi Ra]

Worth the difference to me. Thanks for all the examples.

[Poenald Palen]

Ah, my post is to much befudelling. A clearer post will follow...with the power of titles!

Um, correct me if I am wrong, but 'ARC is not a GPU benchmark.'

Info about it is on their wiki, but it also says stuff that is like...well, it is magic and they can't tell you lol. I

But, you can tear apart the software and use bits of it turned off and make a code for testing it to benchmark different assets loading and further evaluate this. Sort of reverse engineer thier ARC value...if you want lol.

 

Look to the software for anwswers? ARC is NOT a GPU benchmark, so software is all that is left to turn to?! Is overall rendering experience of the viewer is taken into account adn NOT only post loaded content? Different software libraries/components have to handle mesh than handle sculpties and images, they have different CPU time and different effect the overall quing of stuff flowing back and forth? These ques are software based, not GPU an dnetwork adapter...so, sofware again as you end up with a less effiecient quing with some items? Lots of other stuff going on BUT the next bit is a key to it all, maybe?

 

The users experience is not with a "loading" screen that is blank and then it is all loaded and ready, BUT live and streaming...how can software not be taken into account?

 

Heck, maybe you could get the spies to nab all of thier code!

I mean, maybe the CIA already did and they loved the whole secret of it all and possibly it has a sort of number in there that is purely to mess with peoples heads as LL does socialogy experiments to further it's understandings of who a social system can be built to rely on SL, so they can seperate from the US government?

 

In snowcrash the characters ended up with thier own country.

You never know how some of the folks in cybertopia mindsets might sound when they get excited about a future...i mean, the EFF will tell you of the harsh and odd treatment given to free citizens who where enjoying playing a game! So, you never know...possibly a social experiment and you are here to sort of work for the CIA to stick it to a few with dreams of sticking it to the man and using thier knowledge to become kings of thier own cybertopian kingdom...or anarchdom or whatever.

I don't know, nor care. I ramble here to some, while others read this with amusment...so I work is done at least to some...they are amused? Good!

 

But, maybe LL is working FOR the CIA and the US government to find cyber anarchists or maybe even lag crazed sabatours who wish to kill lag at all costs to advance SL!?

The perfect weapons where to lure them into attacking high ARC avatars in a bid to make SL work faster? Maybe they are cybercommies who wish to make the world less laggy, and even have been sabotaging the PC's or makers of laggy objects!!

 

Some people are serious about cyber worlds...so I will go off on a tangent and speak of thier yearnings and struggles...doesn't this hint at me being a sympathizer?! Should I expect some odd gittering on my screen as recording spyware is dropped from it's injector and I get to chat with new friends of a friends friend while enduring odd looks from a few of my old time ones?!! Eek! Ok, read on if you are bored with ARC value BS and wish to read of the strife of living in meatspace that I (horribly) authored/rambled here for lurkers who are laughing and bored to enjoy. Mr. Gibson I am not!

I am sure some are serious though, they wish to break away. Even if SL doesn't wish to. Do these people care enough to make this all work by bashing or even resorting to underhanded techniques to stop SL from becoming to slow? Was this meant as a joint effort to find these guys!?? They wish to NOT be part of a world in which they have so many restrictions and so on. They wish to be freed from thier body as it the flesh is a horrid prison...some of them read books all day between takinig medicine to numb the pain thier nervous system hurls at them in a automaton fashion to "help" them survive. Others just want to be furry animal, feel real love and be safe at home at the same time, some with to cheat death only in a virutal way and still feel the jaws of a dragon chew them up after roasting them! They are trapped, unable to really feel like they can live! Who knows what might be wrapped in a joke or hidden in smile in the valleys and allays, the silicone places that are NOT! They are bricks, glass and metal. They are rusting, decaying and sometimes even under threat while people sit in them and look out wondering "why" and simply only are able to point to the government. So...why not learn as much as you can to make a better virtual world? Why not slip in a joke or a fun sociology experiment that hurts no one? It is not terribly unethical, they are not using to cost anyone any money, hurt anyone (they are saying it is NOT so big of a deal, to keep people from pointing fingers to much?) and basically....well, they can do it and we would not know. But the CIA might. So....so, what? In the end, they all kiss and make up, so no biggy!!?

 

[Failed Inventor]

@ Poenald, your post started out... ok.. and then you fell into a retarded rambling in which I once again couldn't read your WALL text.

[Haye Abbott]

It would be nice if you could overlay a lightmap/occlusion texture over a tiled texture. That way you could get away with a much less detailed baked texture since the tiled texture would handle the details.

[Drongle McMahon]

Yes. That is the best resource-saving texturing development I can think of. After all, it's essentially what you do when you bake a huge custom texture. I think this should come with the materials project that we all hope will follow mesh. As far as I can see, all the basic requirements are already there in the image interpolation and av texture baking functions.