Avatar Complexity too high ?

[Orwar]

4 hours ago, AyelaNewLife said:

The complexity calculations are simply outdated.

   I'm not quite techy enough to be able to get behind how exactly the calculations work or how precisely they preform, but from my experimenting with complexity tolerance I've noted that fully rendering an avatar with 200K complexity does indeed appear to have a larger impact on my FPS than fully rendering someone who clocks in at 80K does (by toggling "render fully" off and on on random avatars with different complexity values whilst in an already taxing setting). I know that it might be a bit of a primitive form of experimenting, but since FPS is usually the go-to measurement for graphical performance, unless my results are just totally random chance - it appears that avatar complexity is at least somewhat on the right track.

   I do prefer to see the people around me, as that sort of feels what clubbing in SL is largely about - but if a person doll up their avies to the point that they alone can cost me an absurd FPS loss, not caring about the performance of people around them, I think that jelly dolling them is only fair. It feels all the more pathetic when someone with well over 200K complexity IM me something like 'nice avi'; I usually tell them that they have 'a nice colour to their jelly doll' - some don't understand what I'm talking about, others get pissy about me not wanting to render their trainwreck of an avatar (especially if you scan them and find that they're running around wearing their high-scripted, highly complex genitalia hidden in their trousers).

   The one thing with the current system that I think might need improvement is that you should get a warning when you attach something with an extremely high amount of complexity. It's easy enough not to notice the complexity prompts when you have your inventory open and focus on picking out an outfit - suddenly you're just way over the top, and don't know why. Often it can be as simple as a single attachment having a few hundred thousand complexity to it, and removing that one thing (usually it's shoes, hair or sculpted jewelry) would put you well within a reasonable complexity level.

[AyelaNewLife]

5 hours ago, Orwar said:

   I'm not quite techy enough to be able to get behind how exactly the calculations work or how precisely they preform, but from my experimenting with complexity tolerance I've noted that fully rendering an avatar with 200K complexity does indeed appear to have a larger impact on my FPS than fully rendering someone who clocks in at 80K does (by toggling "render fully" off and on on random avatars with different complexity values whilst in an already taxing setting). I know that it might be a bit of a primitive form of experimenting, but since FPS is usually the go-to measurement for graphical performance, unless my results are just totally random chance - it appears that avatar complexity is at least somewhat on the right track.

   I do prefer to see the people around me, as that sort of feels what clubbing in SL is largely about - but if a person doll up their avies to the point that they alone can cost me an absurd FPS loss, not caring about the performance of people around them, I think that jelly dolling them is only fair. It feels all the more pathetic when someone with well over 200K complexity IM me something like 'nice avi'; I usually tell them that they have 'a nice colour to their jelly doll' - some don't understand what I'm talking about, others get pissy about me not wanting to render their trainwreck of an avatar (especially if you scan them and find that they're running around wearing their high-scripted, highly complex genitalia hidden in their trousers).

   The one thing with the current system that I think might need improvement is that you should get a warning when you attach something with an extremely high amount of complexity. It's easy enough not to notice the complexity prompts when you have your inventory open and focus on picking out an outfit - suddenly you're just way over the top, and don't know why. Often it can be as simple as a single attachment having a few hundred thousand complexity to it, and removing that one thing (usually it's shoes, hair or sculpted jewelry) would put you well within a reasonable complexity level.

Yeah, this is why I find it awkward to talk about this, because it's not a black-and-white issue at all. Complexity is not useless or totally inaccurate, it's just... calibrated for the status quo of 15 years ago. And could be better; in the way that the calculations used by the Black Dragon viewer are better.

If you take two avatars with the same mesh body (probably Maitreya, on raw numbers alone), and then add 60k complexity of attachments to one and 180k of attachments to the other; chances are the latter is much more resource heavy than the former. Not always though, as there's a chance that the 180k complexity comes from an older mesh object that has a skyhigh complexity, but isn't actually that resource-intensive, while the 60k could be made up of texture-loaded monstrosities that clock in at 1k complexity each.

It's messy. But because it's messy, strict rules regarding complexity limits will throw up a whole bunch of false positives, and really should be avoided by people and especially locations.

As much as I'd love LL to update the complexity calculations, I wouldn't be too surprised if they didn't. There will be a whole lot of anger when anyone that wears a body made by the second most popular creator suddenly finds themselves at a high complexity when completely naked... but that's their fault for wearing a horrifically optimised body.

Oh, and don't get me started on those people that wear a certain brand of scripted mesh lady parts and cry when they're kicked from sims for having a script usage larger than the two dozen other visitors combined. Dagger emoji.

[Nalates Urriah]

3 hours ago, AyelaNewLife said:

...calibrated for the status quo of 15 years ago. And could be better; in the way that the calculations used by the Black Dragon viewer are better.

...

As much as I'd love LL to update the complexity calculations, I wouldn't be too surprised if they didn't. .

About 3 or 4 years ago ACI was made from a revision of ARC which is older but came in after I joined. So, what we are using isn't 15 years old.

I am not sure Black Dragon's values can be called "better". They are different and a good try. But, they are causing problems. People will use whatever is available to hopefully solve a problem they have. Pushing people to lower their ACI is being done by using scripting to check ACI values and kick those they arbitrary think are too high. BD users are getting some surprise kicks. Going the other way if BD users are early to an event it throughs off the detectors and they let in higher ACI avatars.

A project named ARCTan is in the works. It is somewhat delayed as other projects have hit bumps. But, several Lindens tell us it is coming and will reset ARC/ACI. Because of the high impact on the community they are working out how to bring it in without causing massive problems. I assume if its values are higher than the current ACI all ARC/ACI values will be pushed down so the new ARCTan numbers will be in the range of the Current ACI. But, that is just my guess.

 

[AyelaNewLife]

8 hours ago, Nalates Urriah said:

About 3 or 4 years ago ACI was made from a revision of ARC which is older but came in after I joined. So, what we are using isn't 15 years old.

I am not sure Black Dragon's values can be called "better". They are different and a good try. But, they are causing problems. People will use whatever is available to hopefully solve a problem they have. Pushing people to lower their ACI is being done by using scripting to check ACI values and kick those they arbitrary think are too high. BD users are getting some surprise kicks. Going the other way if BD users are early to an event it throughs off the detectors and they let in higher ACI avatars.

A project named ARCTan is in the works. It is somewhat delayed as other projects have hit bumps. But, several Lindens tell us it is coming and will reset ARC/ACI. Because of the high impact on the community they are working out how to bring it in without causing massive problems. I assume if its values are higher than the current ACI all ARC/ACI values will be pushed down so the new ARCTan numbers will be in the range of the Current ACI. But, that is just my guess.

This is honestly the first time I've heard of the difference between ARC and ACI; but as I wasn't around 3 or 4 years ago, I'll take your word for it. That makes it even worse for LL, as they can't even use the "it used to work!" excuse.

ARCTan was shaping up to be far more similar to BD's calculations than ACI, last I heard; although with all values scaled down to the ranges we see using ACI, as that is what the userbase is more familiar with. This was almost a year ago, with ARCTan supposedly in a release-ready state... and I've been out of the loop since then. This wouldn't be the first project that has gotten to a functional state and then left on someone's desk for a year or so; or it could have been overhauled entirely, I have no idea.

[Lawrence Celestalis]

On 7/9/2019 at 4:37 PM, Eva Knoller said:

Hello! You many cause rendering issues to others, but you may also not be rendered at all depending on other user’s settings. I have mine set to ‘jellydoll’ anyone with a complexity over 200K. I have never seen sim rules listing complexity limits, but that doesn’t mean they don’t exist. Usually script usage is the issue.

I also always try to keep mine under 200 as well.

[Nalates Urriah]

9 hours ago, AyelaNewLife said:

This is honestly the first time I've heard of the difference between ARC and ACI; but as I wasn't around 3 or 4 years ago, I'll take your word for it. That makes it even worse for LL, as they can't even use the "it used to work!" excuse.

ARCTan was shaping up to be far more similar to BD's calculations than ACI, last I heard; although with all values scaled down to the ranges we see using ACI, as that is what the userbase is more familiar with. This was almost a year ago, with ARCTan supposedly in a release-ready state... and I've been out of the loop since then. This wouldn't be the first project that has gotten to a functional state and then left on someone's desk for a year or so; or it could have been overhauled entirely, I have no idea.

The changes in ARC and ACI are from hardware changes. As video card tech changes how the render tasks are weighted needs to change. They didn't get ARC or ACI wrong. No excuse is needed. The world has changed and it is time to revise what was once a decent rating system to a better one with more appropriate meaning for today's users.

ARCTan has never been in a 'release ready' state. I have no idea where you came up with that. 

Also, the numeric value chosen isn't about user familiarity. The consideration is in how the numbers will impact the over all SL system and its various components. ARCTan will also change Land Impact. So, the numbers, if poorly chosen, could force a massive return of prims on land. Part of the delay is in choosing numbers that work based on objective data observed in SL.

ARCTan is coming because of how users are using SL and improvements in networking, video tech, and improvements to the SL system. Content Creators need to focus on more appropriate design considerations. Some of the delay is from other projects that are running over time-wise and the need to extinguish conflagrations. Plus, the challenge of getting a better system to fit within the ACI rating scale we have now.

The current deficiency we see in ACI isn't from the algorithm used. It is from the "I didn't see that coming" side of anticipating what users would do. Content Creators have found ways to drop ACI using this algorithm that don't improve the render performance. Essentially circumventing the purpose of ACI and producing a result opposite the intended one. This coming change will improve things. Once it is out users will figure ways to get the result they want and only a few will consider design in regard to render efficiency. So, at some point there will be a new ARCTan replacement once the Lindens learn all those new twists users come up with.

[Penny Patton]

On 7/10/2019 at 2:01 PM, Nalates Urriah said:

On those video cards... My GTX 1060 is seldom pushed to 100% use. I mostly run in the 15 to 50% range. I think anyone checking the utilization on there machine (CPUID's HWMonitor - free- is a good tool) can quickly see if they have a hardware bottleneck.

No argument here, there's definitely a hardware bottleneck, but even if LL managed to solve that tomorrow content would still be a major performance issue. And in any case I'm not holding my breath for LL to make such an improvement any time soon. Content, on the other hand, can be fixed right now if you have the know-how.

On 7/10/2019 at 2:01 PM, Nalates Urriah said:

Yeah, he said those words. But, as written it is implying an out of context meaning. The context (paraphrasing) is the problem is hoped to be solved as a byproduct of other improvements and features being made to the system. If anyone doubts that, visit a Tuesday UG meeting and talk with Oz Linden. Or Corner Ebbe on Twitter and ask about it.

Until I see these other improvements and features making any sort of impact I'm going to remain skeptical. I'm just not convinced LL understands the problem and that comes from years of dealing with them directly. I certainly hope I'm wrong, I'd be extremely happy to see LL make some real improvements when it comes to performance and I will give them all the credit they deserve if I see that happen.

On 7/10/2019 at 2:01 PM, Nalates Urriah said:

The equation was well considered. That doesn't mean it is great. But it WAS the most representative their combined minds could come up with. To say there is a better way... well the open source creed of 'You can do better? Then provide the code or shut up...' might apply. 

One of the biggest problems with how ARC is calculated is that LL has always failed to grasp how textures impact performance and their ARC calculations hardly take textures into consideration at all. This is something they deserve a lot of criticism over until they fix it. That said, as long as LL shoves managing ARC onto individual users setting their own ARC caps, it will never be a useful means of reigning in badly made content. Most people turn off the cap and just ignore it because they don't understand the performance impact, but they do understand that setting the cap means they see ugly pixelated blobs instead of their friends' avatars.

I don't need to submit any viewer code for my stance to have any merit, I'm not a programmer but I have a professional CG artist's understanding of how 3D rendering works which is far more important when it comes to understanding how content affects performance.

On 7/10/2019 at 2:01 PM, Nalates Urriah said:

I think the reason we do not see the un-optimized-content and high ACI avatars clamped down on is the Lindens' support of creative freedom for all. When we object to that it is almost always for personal reasons, a bit over on the narcissistic tendency side. I see few proposing things that add to the creative freedom. Most are much more in the fascist-like controls area. 

This is an extremely bad take. Here are two versions of one of my avatars, with their respective ARC.

Is the avatar on the right "less creative" than the avatar on the left?

Does making a mesh model with a dozen textures that look like this

make it "less creative" than a mesh model with one texture that looks like this?

Of course not, and that's because reigning in unoptimized content in no way impedes creative freedom. It just means content creators would have to be more considerate with how they use resources. Saying people are "fascists" for suggesting LL work in reasonable restrictions on resource use does nothing but derail the conversation into childish name calling.

Edited July 24, 2019 by Penny Patton

[Damian Mills]

You wouldn't cause render issues for me.  You just wouldn't be rendered at all.  I typically set the cutoff around 150.  Even on my serious desktop computer, I really don't want to deal with the 450k avatars every time I go to some club somewhere.

So if it looks good to you, keep it.  Just be aware that for a fair number of users you're going to be rendered in cardboard cutout mode.  You won't look good to them.

[Drake1 Nightfire]

On 7/9/2019 at 11:29 AM, TDD123 said:

So what is it based on ? Vertices ? Textures ?

I don't notice FPS drops ( in my own home alone it's over 100 FPS easilly ) nor high usage of either CPU or GPU.

FLexi prims will crank ARC through the roof. I saw a minotaur with an arc of over a million.. Every bit of fur was flexi.. 

My human av 30523 ARC..