Calculating "Prim Equivalency"

[DanielRavenNest Noe]

Instanced geometry and textures save in a couple of places.  The data only has to be sent from the server once per UUID, and it can be stored in the graphics card memory in less space than unique objects.  When rendering a frame from the geometry and textures, it makes no difference if it's instanced or unique.  For people with small graphics card memory, having it all fit as instances rather than some stored in main computer memory for unique items will save fetching the data into the graphics card, so they will see faster rendering.

In the rendering process, "triangle strips" are more efficient.  When two triangles share an edge and two vertices, the only new information that needs to be loaded is one vertex position.  If you can organize the model into a series of triangles that share edges (a triangle strip), that loads more quickly into the shader pipeline.  Good graphics engines and importers will do that.  Changing textures requires fetching the new one to use in the shader pipeline, so many texture changes slow things down.  But instancing does not affect things at this level.  The shader pipeline is always coloring one triangle with one UV mapped texture at a time.  What makes graphics cards fast is they have many shader piplelines (up to 1000) all working at once.

[Eata Kitty]

Selecting a cost from the greater of the three main costs and having a minimum cost sounds like a realistic approach to object costing. The problem is bolting this on top with legacy costing kicking around, sculpts are obviously underpriced for their streaming cost and have no server cost.

The size costing also appears to be quite punishing on mesh at the moment which is worrying.

[Psistorm Voxel]

imho, mesh in general is being butchered, cost wise, when you look right at it. mesh is the only object type that gets more expensive by size, and the only one that gets a cost slapped ontop - doubled even! - of it for containing even a single script.

If a new cost system is needed, then please do balance it better, so that mesh is actually turning out to be viable, without having to completely bend over backward to get close to what a - less efficient - sculpt or prim build would cost. Mesh should be sold as something good, not something hideously expensive, when already, a lot of fears are attached to it (content theft, copyright infringement, horrible performance from people importing 5 million poly poser meshes.. and the like).

If this cost system HAS to be done, then weigh it properly, and dont try and put mesh on the new system whilst prims and sculpts remain on the old. This will destroy mesh retention rates for sure, it certainly will keep me from using it for anything but attachments.

 

Edit:

If I may think up a general prim weighting system, i have some suggestions here:

- Allow decimal weights. A torus could sit at 1.5 perhaps, maybe even 2, a sculpt could be the same. Decimal prim weights are NOT rounded, but instead are added up to the total sim prim usage. That way, you dont have a prim or mesh rated 1.3 costing 2 on the sim. so a mesh rated 0.7 would still fit.

- A bit more finely scaled weighting. A box costs 1.0, sphere 1.25, torus 1.75 maybe. Sculpt 1.5 or something.

- Fair mesh weighting. Prim weight is based off of physics shape complexity, LOD0 triangle count, and a weighted average of the other LOD`s triangle counts (this should be balanced so that an ultra simple object with only one lower LOD level - due to it otherwise being TOO simple to ffeasibly render - isnt punished massively.)

- Seperate script count limiting system. Script limits do NOT affect build cost, up until a certain treshold. Maybe 16 to 32 scripts, from then on up, a gradual cost factor is introduced into the prims containing those scripts, perhaps 1.0 prim weight for every 8 scripts ontop of it or so. This will allow low scripted builds to remain cheap, and introduces a slow cost-increasing factor for heavily scripted builds. Additional script limitation work can be done via other means, as LL did have plans before the big layoffs, such as variable script memory and such (memory could also be a limiting factor, btw).

[Drongle McMahon]

Now I am confused. I mjust made a three-prim ordinary object, whith a tortured tube, a tortured torus ans a plain cube. Linkset had a streaming cost, from Show Render Info, of 11.9. The pieces on their own were 0.4 for the cube, 2.9 for the torus and 8.7 for the tube. Then I took a mesh that had a streaming cost of 2.9, and thus a PEwt of 3 on it's own, and linked that to the three standard prims. Now those prims satisfy Nyx's condition (3) ... they are linked to a mesh. So they cannot be costed at less than theire streaming costs. But the linkset hade a PEWt of 6.0. That's just the sum of the stadalone costs of the three prims and the mesh. Surely it should have beed 11.9 + 3, which would mround to 12.

What am I missing, or is this a bug?

And incidentally, assuming it is a bug and the cost should be 12, then we are going to have fun when people link a nice mesh chair into their expensive house with nice sculpty windowframes and all that tortured prim furniture.....

[Eata Kitty]

A 1400 vert mesh with a prim equivalency of 5 at 3M in scale becomes gets a PE of 56 at max size, a ten times increase. I don't know if tweaks to LOD or physics (I think both were on automatic defaults) might have brought that down but large meshes appear to be very heavily penalized.

[Ashasekayi Ra]

I'd also have to add my objection to raising the prim cost when a script is added to a mesh object. Mesh already has too many handicaps that do not exist for prims and sculpts. 

[Drongle McMahon]

"...large meshes appear to be very heavily penalized."

Yes they are. The  streaming weight (presently) becomes maximum at an object radius of 34.66m. The reason is that the LOD switches depend on the ratio of camera distance to object radius. At 35m, the highest detail mesh is all you will see with default settings. So there is no resource savings from the lower detail meshes.  If the mesh is big but with small details, it will still try to display those details way beyond where they are less than one pixel wide. To overcome this, we need to hve a variable LOD factor.

[DanielRavenNest Noe]

EDIT: Scratch the below, there seems to be no upper bound on PE weight any more.  Large complex models can have truly horrendous values, over 1000, when I tested a "forest" made of multiple copies of a tree but uploaded as a single object.

Since the formula appears to limit at 82 prims, this will encourage people to make maximum detail objects (64K triangles) on the theory that if you are paying for it in upload and prim cost, you may as well use it.  So instead of getting efficiently modeled houses, you will get highly detailed houses with some landscaping thrown in to use up the polygons.  Paradoxically this will make the load on servers and viewer software worse.

On a 4096m parcel, which allows 937 prims total, it would not be absurd to have a highly detailed landscape (terrain, rocks, and trees) as a single 3d model costing 82 prims, and a highly detailed house costing another 82 prims.  That leaves 773 prims for furniture and customizing.

 

[Jack Abraham]

Nyx;

This is brain damaged.  If implemented, Linden Lab will once again snatch defeat from the jaws of victory.  You've managed to make mesh objects completely unattractive to me both as a creator and consumer, by making them cost more in prims than equivalent non-mesh objects, despite the non-mesh objects being more resource intensive.

Brilliant.

[Vivienne Schell]

"On a 4096m parcel, which allows 937 prims total, it would not be absurd to have a highly detailed landscape (terrain, rocks, and trees) as a single 3d model costing 82 prims, and a highly detailed house costing another 82 prims.  That leaves 773 prims for furniture and customizing"

 

COOOOOOL!

Why do you complain? You´ll finally have your beloved Blue Mars revisited in SL . Crash, crash, crash....boom. Avatar Fashion Show! Yay!

:matte-motes-nerdy:

[Madeliefste Oh]

cYo is a succesful supplying business for builders and creators, we make sculpties with editable textures. With the promise of mesh coming to SL we made plans for the future of our business. The idea was to start offering meshes with editable textures to our customers when mesh would be finaly available on the main grid. In first instance next to the existing sculpties, and maybe later change all our initial sculpties into mesh versions. Depending on how mesh is adepted by the user base of SL.

We did some work on the beta grid, but did not invest a great deal of time yet in making meshes. It is still satisfying and profitable to make sculpties. And now after we have read about this prim count for meshes, I'm glad we didn't. We still can turn back on our plans for the future, we still can be a sculpty supplyer.

For the market we work for are two aspects very important: high quality and low prim count, at least when it comes to rezzable objects. For attachments it's a different story, attachments don't cost prims on land, there is high quality the main argument for buying. But for rezzable objects the combination high quality and low prim count is what makes an object economical valuable, all experienced merchants of rezzables know this.

So the conclusion for cYo is simple. When mesh will have a higher prim count then sculpts, we will stick to sculpts. We don't see that there will be a large market for the more expensieve mesh versions. We might do a mesh line with avatar attachments in the future, but when it comes to rezzable objects you won't find any mesh at cYo.


[Ann Otoole]

If people step away from the organic shapes and build a house out of mesh parts that look like it is made of prims then the prim cost can be far less than the equivalent building made from prims and a few sculpts. I rezzed an example house in mesh beta that demonstrated this. I doubt it is still there as last time I looked that area of mesh beta was missing regions and in general the entire mesh beta area was in disarray with simulators running different versions causing sim crossings with mesh vehicles to result in being ejected into the void forcing a manual viewer crash. Maybe when LL makes all the mesh regions all run the same version then I will try again.

Everything just depends. Don't write off mesh yet. However there will still be a use for sculpts for a long time. I expect mesh will not be "mature" in SL for at least a year or two on release. The initial release will be the basics to meet the stated goal of mesh in 2011.

That said people that build full sims for specific full sim environments are going to be delivering some awesome builds.

Just a matter of perspective and use case.

[Gaia Clary]

This matches with my recent experiences. As long as you only look at 1 Prim mesh Objects and compare these with one Prim Sculpties or regular prims, mesh looses. But as soon as you compare a mesh with a multipart build, mesh gets much better. And as far as i can see, it seems possible to make cost effective low prim meshes which can replace high cost regular builds.

We only have to say good bye to 1 prim meshes as replacement for one prim Sculpties.

What still makes me real trouble are all issues with meshes linked to regular prims. When they continue to end up in unexpected high costs for the link sets, then this may(will) become a big source of frustration.

[Drongle McMahon]

Yes. Looks like Mash Sandbox 27 suffered some sort of mass return. Probably someone linked a few ordinary prims to a mesh! I'm going to put my walls back. I think they show you are right for mesh vs prim, but not for mesh vs sculpt. However, the greater difficulty in texturing the sculpt version may still tip the balance in favour of mesh. That ahving been said, however, I expect the sculpty version is less goiung to be about 30 times the number of trriagles for 2/3 the prime equivalents. That makes the mesh 45 times the cost per triangle. Ridiculous.

 

[Robert Galland]

Well, the lesson from this is maybe, maybe, (simply because of Ann's post) I can offer great housing options using mesh that will advance the look of architecture.  In the end, I need to know if I can still make something that my customers can find reasonable for their prim budgets as well as satisfy my creative juices before knowing if mesh is marketable for housing.  I have to admit, I am worried about the time I have invested and planned to in this endeavor now.