Jump to content
Sign in to follow this  
Vivienne Daguerre

Prim Equivalencies for Mesh

Recommended Posts

I have looked at the complex calculations for prim equivalency, and understand that they are complex, taking into account not only the number of triangles, but size, physics shape and LOD. I understand that mesh should result in less lag on the grid, but also that the data that must be streamed is greater to include UV maps and such.

Given the above, we cannot perhaps expect mesh to be as low prim as we would like it to be. To the little guy, the average Joe Resident who is our customer, all that will matter is prim count, especially if he is sitting on a 512 m piece of land with only 117 prims to furnish it.

I wanted to remake a stew kettle that I have on the main grid that is 3 prims, using one sculpty and two normal prims, using mesh. I wanted to keep the polygon count very low, hoping for a 1 prim or less than 1 prim stew kettle.

Here is the result:

Stew Pots 2.jpg

I uploaded these on Mesh Sandbox 15, and again on 29 to make sure that the server version on 15 was not outdated. From the upload window, I had the following data:

Low Resolution: 370 vertices, 576 triangles, resource cost estimate 2.373 prims

It rezzed at 4 prims.

Higher Resolution: 1347 vertices, 2376 triangles, resource cost estimate 5.878 prims

It rezzed at 10 prims.

The low resolution version is blocky, and I expected that it would have a prim equivalency of perhaps .25 or .5 prim, considering that a sculpty has 1024 vertices and this one had only 370 vertices. I think 4 prims for this is completely unacceptable. If it had come in at 1 prim or less, even blocky it would have been welcomed by those wanting low prim count stuff (almost everyone).

The higher resolution version is nice, and still low resolution at 1347 vertices. 10 prims for this is outrageous. I think it should have come in at 1.5 prims perhaps.

I know I will get lectured that you can't really compare mesh to sculpties in terms of prim count and vertices. The problem is that our customers will be doing that kind of comparison and will have low prim expectations. Given the choice of the three pots, the customer will want the 3 prim sculpty and normal prim original. We content creators will create and deliver what sells, what the customer demands.

I see this as a huge problem. I love mesh, prefer to make mesh, and want to see mesh succeed. This issue could impact greatly on how well mesh is accepted and used on the main grid. I hope prim equivalency can be improved.

Share this post


Link to post
Share on other sites

"I see this as a huge problem. I love mesh, prefer to make mesh, and want to see mesh succeed. "

You do not love mesh, you love the supposed low prim factor. And you do not want mesh to succeed, but you want yourself to succeed.

Btw., do your homework in game design. You actually CAN do this in one prim, but it might take some ...skills. Dolars never were earned easily.

Share this post


Link to post
Share on other sites


...You actually CAN do this in one prim, but it might take some ...skills. Dolars never were earned easily.

 

Prove it.

Share this post


Link to post
Share on other sites

I'd risk to say that it's a result of the genrally curvy form of the pot.  This is a good reason why I tend to tell people that sculpties aren't going away.  Even for the kludge they are, they're quite nice for making reasonably basic organic shapes.  Which, in my point of view, that pot is a perfect example.  And, yes, I can surely make that pot with handle and soup from a single sculpt as well. (^_^)

But, make a "precision" sculpt.  Say... Walls to a house, timbres for a bench, or a bookshelf... And keep it LOD safe.  A lot of faces are lost in clustering LOD culled vertices so that square objects stay square from a distance... Or even to make square edges in the first place.  I personally make sculpts that can't be impacted by LOD to spite the (frankly horrible) sculpt makers who suggest people have to jack up renderVolumeLOFactor for their sculpts to look right..  Ask me about my park bench and my 'One heck of a 1 prim shelf'. =^-^=

I don't think anyone is considering mesh the proverbial 'cure for cancer'.  Mesh, in combination with prims and sculpts will make for the most efficient builds.  To focus on any one tool as "the best" is sheer ignorance.  We don't use any one material to make things in real life, why do we think we need to do so in virtual?  A table can be contructed of wooden slabs and poles, steel screws, and a polymer finish.  While you can make an entire table using each material alone, each will have their shortcomings.  But, the combination of the right parts and tools for the job will genuinely optimise the final product. (^_^)y

 

 

Share this post


Link to post
Share on other sites

The prim count on the edit window have been very variable, at least for me.  A terrain object (a hill) uploaded as 58 prims one time, and 12 prims the next time.  The upload window generally has not matched the edit window.  My conclusion is that part of the software just isn't stable yet.

Share this post


Link to post
Share on other sites

Hi.

I am currently working on a tutorial about "optimizing mesh-objects for SL". I would realy like to take your kettle example as the example object and use it to show some optimization strategies.

If you want so, then i'd appreciate if you send me your blend file of the mesh-kettle so i can make it the "star of the optimization tutorial" ;-)

Cheers,
Gaia

Share this post


Link to post
Share on other sites

Gladly Gaia. What is the best way to send it to you?

I decided to take some pictures of front, top and angled views of the mesh and UV map as they appear in Blender. As you can see, the sides are very chunky, not truly rounded. It looks round because of texture light and shadow baking (done in ZBrush).

Stew Kettle Mesh.jpg

Share this post


Link to post
Share on other sites

"This forum is for serious mesh discussion among mesh creators."

Oh, i thought the forum for serious mesh creators is over at Autodesk, but you certainly prove me wrong.

Share this post


Link to post
Share on other sites

I will be the first to admit that I certainly do not know it all. I would love to learn whatever I can from whoever I can.

I am going to declare this kettle to be public domain. If you want the blender file, Zbrush file, collada file, texture .jpg to play around with, just let me know the best way to send it to you. Once mesh hits the main grid, I will make it a freebie.

It did not take long to make, and if I can learn from this or help someone else learn from it, it would be well worth the time invested.

Share this post


Link to post
Share on other sites

Did you upload with a simple physics mesh? I know the complexity of the physics shape and the LOD levels contribute a lot to the prim count.

Fiddle with the uploader, looking at the physics and LOD parts. I am fairly sure you can lower the prim count a lot.

 

((in my diner in Bay City, there are 2 sculpted one prim objects very similar to your kettle. A bowl of soup(with spoon) and a larger serving bowl on the stove. This may be the type of objects where sculpts still win the prim count battle))

Share this post


Link to post
Share on other sites

I let the uploader make the physics shape. I believe the default is the convex hull, which is the simplest. Please correct me if I am wrong.

You may be right. This may be one of those areas where sculpties win. Perhaps mesh is not for low poly stuff.

I had hoped the mesh would work out for this because it was easier to make it for mesh. I could twist and turn a sculpty shape to wrangle it into the right shape plus handle, but that will be much more time consuming and less elegant.

Share this post


Link to post
Share on other sites

Try making a simple cube that covers your kettle(in blender). Save it as a new.dae and use it as the physi shape when uploading. It also good pratice to make your own LOD files. The uploader gets kinda random creating the lod levels itself. For giggles, use a one or two poly shape for the lowest LOD level, and see how that affects prim counts.

Share this post


Link to post
Share on other sites

There was some major changes in the viewer, how the Convex Hull Physics Shape is create on upload. A couple of weeks ago, they changed it so that the Convex Hull was created always from the Hi LOD Mesh, regardless you provided a selfmade physics mesh or not. This was a problem for small lowpoly meshes, because the Convex Hull Physics mesh could be much more costly than the actual visual mesh. See SH-1537

Now the importer creates the Convex Hull only from the Hi LOD if you don't provide a selfmade physics mesh. Maybe that's why your lowpoly kettle is 4 prims? A simple box for physics should be fine for the kettle, and it probably brings the prim count down.

The big cost savers are Low and Lowest LOD. If you make the lowest very low, you have a good chance to bring the cost significantly down. I achieved best results with all handmade LODs from Hi to Lowest.

Though, a discussion on prim equivalencies should be started when we have workable numbers, some the Lindens say, 'that's what you finally get'.



 

  • Like 1

Share this post


Link to post
Share on other sites

" If you make the lowest very low, you have a good chance to bring the cost significantly down."

I'd add to that thought, another - Considering that the lowest LOD isn't likely to ever be seen, as it's called primarily when the object is at such extreme range as to be little more than a pixel dot, it's safe to make your lowest LOD out of practically nothing.It doesn't have to even be the same shape as the object in question.

The only limit is that you need at least one face for every texture area you want to use. If you have five textures, you have to have at least five faces.

  • Like 1

Share this post


Link to post
Share on other sites

Well, with the default rendervolumeLODFactor of 1.125, I see lowest LODs all the time I zoom out, espacially on smaller models. But maybe I'm to picky on this. :matte-motes-nerdy:

Share this post


Link to post
Share on other sites

"The only limit is that you need at least one face for every texture area you want to use. If you have five textures, you have to have at least five faces."

 

Oo! I learned someting! Thanks! This must stop the textures jumping around as the LOD changes.

Share this post


Link to post
Share on other sites

I appreciate all the helpful suggestions!

Ok, I made a simple cube to use as my physics shape and saved it as a new .dae. On the physics tab, I selected this new .dae. It still came in as 4 prims :matte-motes-confused: Here are some screen shots to give you more details.

LOD.jpg

This is pretty low at all LODs already.

Physics.jpg

A Physics shape cannot get any simpler than this, and the Physics Cost is showing as 0.

Share this post


Link to post
Share on other sites

Some possibly relevant points...

1) The only number reported in the upload dialog (till now) is the streaming cost. I think Vivienne D is reporting that number. In that case, the physics cost is not yet taken into account in these costs. However, this, rounded in two dimensions, is typical of the type of model that gives a convex hull with very many surface points. A physics custom shape is pretty much essential to avoid high physics cost, especially for a small mesh like this.

2) The cost reported in the upload dialog is still completely wrong** (CTS-536). The costs from the edit floater are more relevant, but are still likely to be changed before release.

3) It is true that the mesh version is always going to be far fewer triangles for the same thing than the sculpty version, and that that means less work for the gpu. However, the streaming (bandwidth) cost is usually the overriding thing for meshes. Sculpties can be efficiently compressed (but...) and that is why they win out here, coming in at less than one byte per vertex, while (IIRC) meshes are >10 bytes per (hi-LOD) vertex even after compression. Unfortunately the net effect is to give sculpties the edge in many cases and thus burden the gpu at the cost of sparing the network.

4) Actually, sculpties are usually not optimized for lossless compressibility. A 64x64 map has 3007 (of 4096) unused pixels which are usually (often?) interpolated. If they are instead made equal to adjacent used pixels*, the maps certainly compress much better with png, and I guess they do with jpeg2000.  Also there are still sculpt maps 128x128 etc, which are completely wasteful since we got lossless compression.

5) This model is in the size range where the streaming cost is dominated by the lowest LOD. This is not the case for larger meshes. There are fixed values of the radius (half-diagonal of bounding box) where successively higher LODs disappear from the calculation. These are currently 4.33, 8.67 and 34.67. So once you reach a 40x40x40 cube, the cost depends only on the high LOD.

*For the pedantic: OK, there are a minority of cases where the sampled pixels at lower LOD are not a subset of those sampled at high LOD, and these could be more accurate with interpolation, but the differences are small.

**at least in the last viewer I checked.

  • Like 1

Share this post


Link to post
Share on other sites

A  Resource Cost of 2.3 in the upload floater should be way less inworld in the edit floater.

Which viewer version do you use?

 

Share this post


Link to post
Share on other sites

That looks much like the convex hull was created from the HI LOD. I can't remember on which viewer version it was changed again, but pls try with the latest viewer if you can. And switch the Physics Shape Type to PRIM in the edit floater under the Features tab.

  • Like 1

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Sign in to follow this  

×
×
  • Create New...