LL Blog: Avatar Rendering Cost

I doubt it's going to get that far but it would be interesting if all of us who have lower number avs (I'm about 770 which is on the high end of the green zone) could sell their excess to people with higher scores for a profit.

Okay, when I finally get around to doing that pulsating, spinning, thumping neon pixel-supernova city called Respublica Blingtardia, ONLY those with 'red' scores will be allowed in the region!!!

I'll only relent if you are a well known scammer, griefer or adfarmer notorious for using a noob avatar...

It never ceases to amaze how people will complain about absolutely anything. Here we are, presented with this highly useful new tool to help people learn some of the most basic fundamental principles of how 3D graphics work (which hopefully includes some of the essential do's and dont's), and all some people can see fit to do is complain about the potential for yet further complaints about what the numbers might reveal. Wow.

Has it occurred to anyone who's complaining right now that your avatar causes just as much lag whether there's a number over its head or not? The only thing the number does is quantify the amount for you, so you can see for yourself what effects your decisions have. Given the choice, can you honestly say you'd prefer to remain ignorant, and just pretend that what you do doesn't make any difference? Come on. The system doesn't work by magic, guys. It works according to a very real set of computational laws, the same set which governs every 3D application in existence.

No one is saying that the system itself shouldn't be made to run faster, least of all the Lindens. It absolutely should, and we all know that. But the fact is even the best written, fastest, most stable, most high-performance 3D engine in the world would slow to a crawl in the face of what some people routinely do in SL.

The workings of the system are obviously beyond any of our control as users. But the things we create in it are absolutely under our control. This tool simply gives you feedback you need in order to help find the optimal balance between visual quality and performance cost. And that's ALWAYS what digital art is all about.

The point is there are lots of ways to achieve any one look. Some of those will always be more computationally expensive than others. We now have a tool in place to allow us to see first hand which approaches to any given art problem are most efficient. There's a tremendous opportunity here for people to learn to lower lag, while still having all their stuff look just as good. So rather than complain, which accomplishes nothing, how about getting excited about the fact that content creators will now inevitably compete to see who can make the best looking stuff with the lowest rendering cost. The end result is your avatar can look just as good as it did before (or better), while your frame rate goes up and up and up. It's a win/win all around. Anyone who can't see that is sadly missing out.

Don't forget to invite (saying company names instead of av names due to naming names policy) World Stock Exchange, Ginko Financial, Allenvest Financial, and about half a dozen other ponzi and scammer runners... oh and all the bot runners as weoll and of course Prok.

Does anyone know if the generated number is based on an average of the typical type of GPU running SL or is it dependent on the individual's system? Does a low end computer return the same number as a high end one?

I agree, Chosen. This is prime time for content creators to take their creations into the beta grid and test them out for efficiency. To really get a handle on video memory and make the best quality textures at the lowest resolution possible. The rendering lag is the worst thing for me, and we have FiOs at home.

It's not dependent on your system in any way. All computers will display the same numbers. The formula is explained in the blog post.

A bunch of people have asked that question and we haven't gotten a response yet which is a shame because until we know these numbers are only good for comparison between two different avatars without a baseline.

I am unimpressed by this "feature"..

LL, how about adding a feature to the client that lets me track how many sales I'm losing due to lost consumer confidence? how about spending some time, Oh.. I dunno.. fixing the grid.

I had three customers needing refunds today due to failed transactions this morning.

Winter, you have been here long enough to know that the work being done is compartmentalized so complaining about that work being done every time another section posts about it doesn't do any good since the people posting about client tweaks and stuff are entirely separate from the team working on the server issues. I understand the frustration that many people (myself included) feel but the I'm going to shout out about the server problems everywhere I can because the lindens have ignored me everywhere so far and I have no where else to shout out" mentality is useless and does no good not to mention that it makes people frustrated at you for making them read my rants about your rants.

Yes I understood how they ranked things. I am curious though how they decide that a number is in the red zone when certain machines can certainly handle that with ease while others can't. What level of computer system are they basing whether a number is in the green, yellow, or red zones as far as rendering cost is concerned? Sorry. I'm trying to understand this.

I also have been here long enough to know that work on projects is "voluntary" at the Lab, and some programmer worked on this, rather than on the actuually USEFUL things. They invented a whole new way to discriminate against people based on their avatars.. rather than actually making SL a better place.

My frustration, at what I feel is a "stupid" and "useless" addition to SL, (another tool enabling discrimination and exclusion) when there are more pressing matters that I would rather "the company" were paying it's people to address... is exceptionally on topic for this thread.

We didn't need this feature.

Will the hair makers and clothing designers start putting out more info about the Rendering Cost of their items? If I actually decide to try and keep my number down, it would piss me off to buy something that looks simple, only to find out that it will boost my number like crazy again.

I already figure that I would probably have to throw out the majority of my wardrobe if I want to keep my number really low -- and I never really thought of my outfits as that complex. I already toss or donate outfits after I wear them once - or most outfits anyway. So the thousands of items in my inventory still need to be worn at least one time - regardless of their rendering cost - in order to get some value for my spent lindens.

As much as I love some of the things that have come out of voluntary work that probably never would have come about otherwise I seriously hope that will change with the coming in of a new ceo so as you said some real work can be done before new features are added... I hope though that the general assumption that Lindens work on these things without regard to core projects is wrong and if it is wrong.

Winter, I feel your pain. Really, I do. But you're barking up the wrong tree. The guys that added this feature (Pastrami and his team) only do graphics work. They are not responsible for "fixing the grid".

You might as well go into an auto body paint shop and ask them to fix your transmission. After they're done laughing at you, they might take a minute to explain to you why it's not their job to do as you asked. Their job is to paint the outside of your car, not to mess with anything under the hood. If you want your transmission fixed, you go to a mechanic, not a painter.

By the same token, if you want the grid improved, you talk to the network people and the server code people, not the graphics people.

I think that's a great idea.


Absolutely.

It's 10 base points per attachment, not per prim.

The blog also doesn't explicitly state it, but each unique sculptmap counts as a texture (5 points) as well.

It still has the problems from RC0 as well:
* texture size isn't considered: an attachment with a 32x32 and a 128x128 texture (random small size examples) will show as more "costly" than an attachment with a 1024x1024 slapped on it
* prim count/type isn't considered: I doubt a tortured twisted torus renders as efficiently as a plain box and it's plain silly that a 255 prim attachments has the same base "cost" as a 1 prim attachment

Avatar "shame"/"cost" doesn't really reward building efficiently, an efficient creation can even end up being more costly. If people build with ARC in mind, things won't necessarilly get better.

As far as I can tell while the point totals are based on a formula (albeit one that's apparently flawed as pointed out above) the green/yellow/red designation is entirely arbitrary probably based on what the lindens each scored on a game of darts at the bar the night they were brainstorming this.

That's a darned shame. Gonna be tough enough to get people to understand how/why to use this thing without it being markedly inaccurate (even misleading) in some critical areas.

I hope there will an opportunity for refinement down the line, and I hope the voices of those who would recommend useful changes aren't drowned out by the clamoring hordes outraged by the whole idea.

Nope, it is PER PRIM.

Quote: " All attachments are then looked at on a per-prim level. The prims are weighted as follows:"

I tried on several of my furry avatars while watching those stats. The prim heavy ones that had the same number of attachments and SIMPLER texture schemes were ranked way worse.

My avatars ranged from nearly 4000 for an incredibly detailed wolf with lots of flexi-prim fur, to just one point for my "Low-lag Human form, that has non-prim hair and no attachments or scripts. (She still has a very nice skin, and has a high-quality texture for her non-prim hair...).

Ny normal avatar was 1122 points, for a three-tailed Kitsune with flexi tails and a talking muzzle. Some of my other furry avatars were as low as 300.

I actually appreciate the tool. But I am concerned that, since you also see everyone else's rating, it will be used as a reason to discriminate against non-Human avatars. "Take off that prim heavy furry form if you want to be here...".

I've been playing with this since I downloaded the latest RC a week or so ago, and it's kind of fun to find out how different outfits and accessories affect my rendering cost.

I also like that it gives me some level of control over how much lag I'm generating for myself when I hostess or go places that might be crowded. I've been surprised to find that what I would have thought were costly attachments or avatars were sometimes much lower than I thought, and less primmy attachments were sometimes quite high.

I was *dismayed* to discover that my walk replacer, which I always thought was a single-prim attachment, is something like 20 alpha-textured prims and inflates my cost dramatically. Sigh. But without the rendering cost display, I'd never have known.

I hate to be the bearer of bad news but there already are places that discriminate against furry avatars, not just prim heavy ones but furry avatars in general, although they are few and far between because most places realize that it doesn't make good business sense to turn away customers for stupid reasons like whether they're in furry or human form.

Some observations:

A three position tail, made of 28 prims and two textures, added 92 points to my score.

A 4-position tail, made of 185 prims and using 3 textures, added 1553 points to my score!

The HUD for my avatar did NOT add to my score.

My favorite hair costs me 400 points, and it is NOT a terribly complex style.

The tool on avatar rendering cost is a great debugging tool.

The problem isn't with the tool, it's with that strange blog post announcing it.

The tool is good for content designers who want to make resource-efficient content. It is a useful tool for Linden Labs programmers trying to figure out how to stabilize the grid.

Then there is this blog post. The blog post says little about the useful features of the tool. The blog post says "It's not our (Linden Labs's) fault that Second Life runs so poorly, it's your (the residens') fault, this tool proves it, so stop complaining to us and start complaining to each other."

This kind of tool could help Linden Labs understand Second Life's resource problems better so they can find a solution. It looks like they aren't going forward with the find-a-solution part of the deal.

Linden Labs, as a company, just makes no sense.

It's 10 per attachment and the amount of prims don't count, I looked at the code for RC5 to make sure it was still the same I posted when RC0 came out (/327/4f/252877/1.html#post1948756).

(Prims *do* count if the size along any axis is >1m, but that's rather rare for an attachment)

You can easily check for yourself:

* A) rez a prim, take it into inventory and wear
* B) rez 2 prims, link and take into inventory and wear instead of A

Your ARC for A and for B will be the exact same. You can make a 200 prim linkset consisting of plywood boxes, cylinders, torus, sphere, etc it won't change your ARC as long as every prim is smaller than 1m in any dimension.

Some things do count per prim (flexi, particle source, etc), but the amount of prims itself (or their type) does not count.

---

One thing I always noticed is that a lot of attachments (particularly prim skirts) alpha faces that aren't directly visible (the inside of the skirt) which makes up the big difference in ARC spike two seemingly identical items.

Invisible attachments are some of the most costly as well since every face will have an alpha texture/transparency on it (if you have a - complex - attachment you can hide/show on demand, look at the ARC before and after hiding).