This week has seen discussion on the ARC/ACI values and how broken they are. See the SL Forum thread: Rigged mesh LoD bug. I knew ARC/ACI weren’t all that accurate. But, I thought they were reasonable comparisons. I still believe once up on a time they were. But, in light of this new information I can’t prove that.
ARC = Avatar Render Cost
ACI = Avatar Complexity Index
The purpose of ACI is to influence users and designers to buy and create more efficient mesh content. Something along the lines of Consumer Reports, but free and in your face.
A couple of things have happened that has not only changed ACI’s usefulness but distorted it and pushed us away from its goal. Possibly even making it counterproductive
The SL system has been changing. Things it did NOT do well it now does well. Things video cards had to work hard to do are now much easier for them… or looking at that another way, the cards have so much more power they are no longer problems, and a bug (feature?) in the viewers. And the fact designers have found ways to game the ratings, while in their interest aren’t in the best interest of Second Life™.
So… what is going on with ACI and how do we know? Those questions are answered in the SL Forum thread linked to above, Rigged mesh LoD bug. The thread was started by OptimoMaximo. He and Beq Janus discussed/debated the problems with ACI. ChinRey has embellished the information.
The thread is long-ish and definitely technical, but worth the read. I’ve had to change my thinking.
If you are non-techy and wanting to avoid glazed eyes, the short version is simply ACI reports values that are way too low for mesh and items and those that use materials while reporting values too high for prim attachments.
A prim hair from years ago may report as 170k and a new mesh hair report as 2k. So, which is better and less laggy for the system and viewer? Well, it isn’t that easy to know. ACI is telling us the mesh hair is better all around.
In the thread we find the thinking is ACI is off by a factor of 100 to 300. However, there are a number of variables that affect video cards differently and affect the load on the SL servers and network differently. Combining all those to give a generally comparative number is pretty trick. So, forget precision.
Black Dragon Viewer is made by a contrarian and uses a different ACI algorithm. My previous version was showing ACI and polygon count. After an update I can’t find the freaking setting to turn that back on. I think the poly-count was probably the most useful for comparing items.
Whatever, in BD my FS-ACI 23k mesh hair has an BD-ACI of 196,614. Putting on a 2013 hair with a FS-ACI 90k the BD-ACI is 196,819. Digging I find a 2010 hair with a FS-ACI 159k and BD-ACI 193,146.
I don’t know how NiranV does his calc for ACI. If I remember correctly he did explained it some time ago. Whatever, we see difference factor of 8+ times.
The size of the numbers is not important. It is the difference between them. The LL and FS viewer give a difference between the hairs of 159k – 23k = 136.
While BD difference is 196,614 – 193,146 = -3,468… which says the old high ACI hair-do is better than the new mesh hair-do.
So, the foundation on which we base many of our purchasing and design systems is broken.
The Lindens are aware of the problem. They are collecting data on what things are worn and how the different viewers rate the items. The plan is to use this data to change the ACI algorithm to better reflect render cost and eliminate some of the loop holes that allow the system to be gamed.
We are a ways off from seeing a revised tool in our hands. For now, there just isn’t much we can do. Plus, those boot-you-out scripts use the numbers we have today.
I recommend going back to the old pre-ACI trick of trying on the demo and pressing Ctrl-Shift-R to switch into Wireframe mode. The smaller the polygons, the harder it is to render the item… and that is a VERY generalized and not that accurate test. But, until the coming ARCTan (name for new ACI and LI algorithm) is out…