Second Life Still Struggling With a Crisis of Poorly-Optimized User-Made Mesh Items -- But a Pro Game Artist Has Some Solutions

Above: A single Second Life fashion item with nearly 50,000 triangles (Courtesy Penny Patton)

Longtime SLer and professional game artist Penny Patton recently bought a Second Life shirt and then saw something that shocked her. That one mesh item (see above) had nearly 50,000 triangles.

As a point of comparison, characters in AAA console games with high end graphics rarely have this many triangle. The ultrarealistic lead character in God of War, for instance, has on the PlayStation 4 version only has 32,000 total triangles. For the entire character and their clothing items, while looking even more detailed than that one Second Life shirt. (Click Kratos at right to see.) 

But when Penny contacted the creator of this shirt in Second Life, she got an interesting response: 

"They informed me that they could not reduce this because it results in a 'decrease in quality'.

"The [shirt has] 11 faces. Each face has three maps. Diffuse, Specular, and Normal. It's even worse because the specular actually only uses one map, a black 1024x1024 texture that was uploaded 11 times instead of reusing the one.

"It's absurd. It's also, sadly, common in SL these days.

"This is why SL needs hard caps on avatar resource use."

It's not only common, it's an ongoing problem, and hasn't really been sufficiently addressed. As a former Linden Lab staffer told me about poorly optimized mesh, "These items are effectively DDOS attacks on everyone around them." 

Five years ago, we pointed out that most of the top-selling mesh bodies in SL are poorly optimized, and have even more triangles -- 200,000 or more. (Penny has also been writing about the issue on her own blog for years,)

Fortunately there are ways Linden Lab can address the issue head-on. 

"I really want to stress there is not one thing Linden Lab should do to fix a problem like this," Penny tells me. "There's no 'silver bullet' solution that will fix the problem and make everyone happy."

That said, here's the solutions she recommends:

There's really no reason for a plain T-shirt in #SecondLife to use more than 3 textures. The diffuse, the normal, and the specular maps. The same goes for a pair of jeans or pair of simple shoes. What's more, you can typically make the specular map half the size of the diffuse.

— Penny Patton (@PennyPatton) March 23, 2023

Content Creation Blog Which Also Corrects Misinformation

First is education, which a lot of people dismiss because education by itself won't do much.

But you need to put the information out there so people can understand the other steps you take. I've always felt Linden Lab should create an official blog of sorts that focused on content creation. Showing good examples, creating bad examples and explaining why not to to do that, holding contests to encourage the community to get involved, that sort of thing.

Importantly this would serve as an official source those who understand the importance of optimization could point people towards when we see misinformation being spread.

Plenty of content creators in SL, either out of ignorance or laziness, will try to tell you their high-poly, VRAM bloated content doesn't impact performance and right now there is no official Linden source to contradict that. It's also important that having LL coming forward about content and performance would help spark a broader community discussion about the issue.

Better Tools and Optimization Warnings in the UI

Residents need better tools, both to manage their own resource use, but also tools to better understand where the lag, FPS drops, and other performance issues they experience are coming from.

For example, if I attach an item with an excessive amount of textures or polygons, maybe there should be a warning about the performance impact? Perhaps if my avatar in total exceeds a certain amount of triangles/VRAM I could get a warning about that?

"Your avatar is using over 200,000 kb textures, this can result in lower FPS and longer rez times for yourself and those around you." and this warning could escalate the larger your performance impact is. If avatars around you are the problem, perhaps the viewer could recognize that and suggest you enable the "jelly dolls" feature, or lower the caps on it.

If the environment is the problem, perhaps the viewer could suggest capping texture resolution at 512x512. Give people the information they need to understand the problem, and the tools they need to deal with the problem.

System Limits on Poorly Optimized Content

Finally, and this is the one everyone hates to hear, but we need hard caps on resource use.

There is absolutely no reason any avatar should be walking around using 500,000 kb of VRAM, let alone 1,000,000. Ideally avatars wouldn't use more than 100,000 kb but I don't think Linden Lab should be that strict. I'd give avatars 200,000 or even 250,000 kb.

But what happens when you reach that cap? I don't necessarily believe you should be prevented from wearing any more beyond that. Instead SL could put a maximum texture cap on just your avatar. Reducing the maximum texture size of your avatar's textures until you fall below the cap.

For most people, this would be a 512x512 cap and they'd probably never notice. For more excessive avatars it would become more obvious as they hit 256 or 128 caps. For triangles we'd probably need a system more like Land Impact.

It's important to note that Linden Lab is currently working on rendering pipeline improvements that will give us some substantial framerate improvements. We're talking double the framerates you currently see.

But if you're typically seeing 10-15 FPS that still only means 20-30FPS with the new pipeline. And that FPS can still drop when you're moving around as opposed to standing still.

So, at this point, I still believe encouraging optimization is something Linden Lab needs to take seriously.

Hard agree. Linden Lab's own efforts to optimize Second Life with cloud rendering and so on are regularly undermined by poorly optimized user content.  

Read more on Penny's blog.

Update from Penny on the Kratos comparison (3:40PM PT): "The shirt comes in two pieces because it has a 'regular' and 'pulled up' version. So the entire attachment together is 120,770 triangles. So that one shirt is about the equivalent to four Kratos."

Comments

[Kyz]

They could educate about texel density and perform checks based on specific texel density unit sizes as well. This keeps any small and medium sized props realistic while allowing for large item exceptions. You should be able to break the rules and use higher texel density on objects that are intended to be viewed close up as needed.

This is how it is commonly done in professional studios. The bonus is that all items appear to be at the same visual resolution with triangle and texture size, which makes the entire surroundings look more believable and not consisting of mismatched resolution items all over the place.

[Judas]

With building everones an expert. Most of them only got to be experts from the opputunity of learning to build in sl over a long period of time.
Everone can and should be able to develop their skills in sl, not just the building gods, its why its fun.

[seph]

Customers buy products that look great, regardless of what it does to their performance. If they preferred performance, I guarantee creators would be selling performance.

Linden Lab rolling out PBR and some better lighting features will help creators rely on more modern material maps and rendering instead of mesh density. More modernizing of Second Life's graphics engine and content creation pipeline will help more.

In the meantime, no creator is doing this to sell less but make performance floater watchers happy. Penny can keep calling them ignorant and lazy, but making customers happy is anything but.

[Penny]

@seph Most basic optimization can be done without detracting from how good your content looks. Let's take the t-shirt from the article. Eleven of its textures are a blank, black texture used as a spec map. Are you really trying to argue that somehow looks better than using one 64x64 black texture?

To put it in the simplest terms, optimized content would look just as good as anything you find in SL does now, but you'd have much higher framerates and much less lag to deal with. Sims would be able to handle many more avatars without crashing your viewer. You'd be able to turn on more graphics features, and dial that draw distance up further. There is literally no downside for the customers. And honestly, it wouldn't put much more work into the content creator's lap, either.

[V]

We have newbie meshers uploading straight from programs like marvelous designer into SL without any retopolization or optimizing in between.... And the UV maps are layed out so every face takes nearly an entire map, thus 1 image per face instead of 1 image to use for multiple faces. Hell they could use a 2048 image for all faces and it would still use less memory. I am a clothing creator that cares about optimization, so we do exist. But I also work in the tech industry in RL so maybe that's why...

[seph]

@Penny

Most of SL isn't running around in 50k black t-shirts. Most people are wearing beautiful well made things that are very intricate but yes, very dense in mesh because that's the only way to achieve competitive detail right now without proper modern materials and lighting.

So it's not about how to make a really good black t-shirt. It's about how to be competitive with other designers and always one-upping yourself for your customers. For you its a no brainer to put shirt buttons on a normal map and hope customers have ALM on and happen to be in good directional lighting where maybe that detail will pop. For a working designer in SL though its a no brainer to mesh that button and ensure the detail pops because all their competition does that and a flat unshaded textured button would awful.

If one designer won't use increased mesh density and more material maps to make something better looking, another will, and the former will be at a disadvantage that costs money and livelihood.

The fix? Linden Lab keeps making strides like the upcoming PBR materials. Making better performance and better looking one in the same rather than things at odds is the only way this will ever improve. Accusing ignorance or lack of education is short-sighted. The most popular and most worn designers are incredibly talented and overqualified for Second Life's level of content creation and know exactly what they're doing and why.

[Penny]

@seph Most of the resource bloat plaguing SL content isn't contributing towards better looking content, so the entire argument that it has anything at all to do with "being competitive" goes right out the window. You're creating a straw man.

[0xc0ffea]

There is no world in which "better content" equates to "more draw distance" on the SL platform.

On shaves a few thousands of a second from GPU draw time. The other is an logarithmic progression in the volume of raw content to be fetched, processed and downloaded.

Render time is NOT the bottle neck limiting your FPS.

VRAM usage is NOT the bottle neck limiting your FPS.

That't not what the SL viewer spends most of it's time doing.

[Lara]

@0xc0ffea
Okay, so how come it makes a gigantic difference whether you're surrounded by a group of average avatars with 1-2 million triangles and textures, or avatars with 100k-400k triangles and optimized textures, all in the same place and with the same graphic settings? In some communities okay optimized avatars are the norm. How is it that the performance is so much better than in the presence of people who shop at stores that sell inferior, unoptimized items like Bluberry, Truth, Legacy and the like?

Please explain that to us pesky poor mortals who only have to buy a new $3000 gaming PC every 2 years to not notice the very obvious performance difference.

Sorry, Google translate.

[Luther Weymann]

I'm happy with SL. I import my own.dae and find even my most complicated stuff is under 5K VRAM. I just checked my avatar for the first time, and I'm wearing a very old mesh body and skin and with all mesh clothes on I'm at about 150K. Whatever that really means, I don't care much. If I leave my draw at about 200 when I jump from the store to another region with my boat docks, the rez time to 100% around me is about 15-20 seconds, but if I make that jump at 1024 draw rez takes longer. I'm on 100 Mbps by ethernet. I'm in no hurry in SL. If I wanted instant rez I'm be in a game made for a purpose. SL was not made for a purpose, it was made for us the Residences to have at it, and we are having at it!

[0xc0ffea]

@Lara Avatars are CPU expensive. The more rigged objects the worse it gets (remember, 1 attachment can be many objects).

You're seeing different performance in different communities, not because their avatars are lower tris or "better optimized", but because they are wearing less rigged stuff.

The rendering part takes about the same amount of time each frame. The processing part can vary wildly depending on what avatars are wearing. More processing, more time between frames, lower frames per second.

It's also really easy to apply personal bias to performance and blame others for the problems. Actually bench-marking SL is surprisingly difficult and involves specialist tools. There is a lot more going on that just "render frame rate". This difficulty is why people like to point fingers and blame "evil bad creators" for making their SL slow.

[seph]

@Penny "Most of the resource bloat plaguing SL content isn't contributing towards better looking content, so the entire argument that it has anything at all to do with "being competitive" goes right out the window. You're creating a straw man."

A straw man when I suggest creators use higher mesh densities for reasons other than "ignorance or laziness"? I'm not sure you know what a straw man is.

You've been accusing the worst of designers for years, loudly and echoed by this blog and haven't convinced anyone of harming their business to appease you. Meanwhile Linden Lab is adding PBR materials and better lighting and hopefully won't stop there.

I think Linden Lab is of my sentiment that if designers were given tools to put the detail they want in materials instead of mesh, they would. The old material system was faulty (ALM not reliably on for everyone, rarely any directional lights besides the sun). This new PBR based system is much better and if you really are more interested in performance than attempts to shame people that do for a living what you blog about, you should hope this straw man reasoning works out.

[Zidders Roofurry]

Nobody cares how competitive the market is. Designers having to compete is their problem. The fact is they're just using it as an excuse to not learn how to mesh properly. Most of them don't care about the issues their badly optimized clothing creates for everyone around them.

[chesse vyceratops]

if linden labs doubles the framerates it'll be like garrys mod when they added "fast download" and people will just double the amount of useless scripts and triangles on their avatars

[Adeon]

Hash uploads. Don't let them upload the exact same 1024x1024 100% black texture 11 times.

Or let them waste their money, give it a unique UUID and everything, but it just points to the same single download in the cache.

[Penny]

The comments here really drive home how important it is that LL becomes involved in the discussion. But until that happens there is one challenge I put towards anyone: Does optimization get results?

When I'm in one of the handful of sims I've built myself, I get much higher, and much more consistent, FPS than I do anywhere else in Second Life. I never see texture thrashing in my own sims, it simply doesn't exist in an optimized SL environment. I'm able to run SL at higher graphics settings in my sims. My sims load lightning fast, because there's less to download.

People can pound their chests and "debate" the topic all they like, but at the end of the day but at the end of the day the only thing that matters is who is offering suggestions that actually work.

[Lara]

@0xc0ffea
You really do come up with more and more excuses every time, right? Average avatars tend to be looking boring. They wear a body, head, hair of which up to 10+ copies are invisible, shoes, upper and lower garment or a dress or most commonly nothing but underwear, maybe earrings and a necklace. Look at them in wiremode, almost solid. Roleplay, Anime or Furry avatars wear way more attachment the humanoid often use the same heads and bodies which still make them more complex until alternative arrive. Clothing, hair and accessiories is 2-5 times lower in triangle and texture, their own creators make head and bodies are waaaay way lower. It seems in your view SL HAS to run like crap so non rich people and non gamers can't accsess SL, mesh content HAS to be the exact same an intelligent person would only use to render pretty pictures in a 3D programm. Why are you so against the tiniest amounts of optimization? Do you like zooming in on weird looking genitals and no rough shapes have to be seen when 1cm of SL covers your entire 300 inch 8k screen or what? Why are game developers still not using this kind of content despite having better foundation to build on? You seem aware that people squeeze more out of SL than SL is build to handle but you play apologist. And I know you will come with your next excuse "muh aMaTeUrS". All creators who optimize best to their ability are amateurs, one always cries in group chat about how it takes soo long to create. Everyone who learns to do mesh learns you HAVE TO do optimization when content is made for games. How you don't come across this basic information? You use a dense model for textures you don't upload said dense model like popluar stores. You make LOD model each cut roughly in half, you don't upload dense model and set mid and low to 2 triangles. All vehicles in SL are Spongebob invisible boat mobile at 10m distance lol, will you tell us max out LOD factor or deal with it, hm? Bluberry, Maitreya, LeLutka, Truth and so on are lazy greedy f*cks who don't give a crap. LL should not care about licking their asses and make a AR category under griefing for this content and force them to do better, period.
I'm not the only one who tells you optimize regions and avatar run better and yet you don't acknowledge and make excuses.

[sirhc desantis]

Fabulous to see ser Penny back on their crusade. What was the long gone SLU response..

[Wagner James Au]

> Does optimization get results?

How about someone shoot video of themselves teleporting into a sim wearing very poorly optimized mesh, and then teleport into the same sim with very well optimized mesh, and compare the results?

[seph]

This blog and Penny's fail Google's PageSpeed insight badly for performance. Does that make either of you ignorant or lazy about running blogs? You do what works, and you aren't inclined to change things if your customers don't complain and keep being customers.

Imagine someone that doesn't do what you do for a living insulting your know-how or work effort because they're able to make a 100/100 PageSpeed scoring blog that has no real business associated with it.

Just as NWN would probably drop all ads and etc. for better performance if it was good business to do so, SL designers would prioritize performance if it was good business to do so.

Make optimal performance the same thing as best graphics; which PBR will help with and they shouldn't stop there. 'til then, designers will prefer good business over best performance.

And it should go without saying most SL designers aren't making 50k black t-shirts with 12 materials. Tons of designers put out great low-poly content that's still beautiful and still favored. It just all could be better with more modern workflows like PBR.

[pointer]

Meh, whataboutism and false equivalence. Aside the fact that Penny's blog is hosted on blogspot.com which gives only a limited control and speed is not just about ADs, the validity of her argument doesn't change.

Want to know what's alike, instead? Penny uses JPEG images in her blog, that should be the norm and obvious (or webp now), but many SLers always uploaded PNG images, even when slow A/DSL was the norm. Webpages that take longer to open than downloading the SL Viewer. That's the same attitude of the content creators in SL that don't know or don't care.

That way, they create SL content that adds nothing to "quality" and technically is qualitatively rather poor, instead. Essentially it looks the same, but it is terrible to download and to render. That essentially that's what changes. "optimal" doesn't mean "absolute best", it means optimal. That stuff isn't optimal at all.
Not just that T-shirt. I always check the demos before buying and I facepalm most of the times. Insane amount of triangles, textures, poorly scripted HUDs that use a script per button and even fail...

Letting people know and encouraging good practices is a good idea, that makes the virtual world better to everyone. Grasping at straw doesn't.

[poin]

@ Lara
You have some good point. However, please note that invisible parts are ignored by the viewer now, so invisible copies are no longer an issue nowadays, except perhaps taking a little longer to download and rez.

[seph]

@pointer "That way, they create SL content that adds nothing to "quality" and technically is qualitatively rather poor, instead."

Money talks. Those bodies, heads and clothes you call qualitatively poor make millions USD and thousands of customers happy.

Also if you can understand how browsers offering modern features can help and encourage these blogs look better as well as perform better, its the same with SL. The asset capabilities are stuck in the 00s right now. PBR will be a step forward. Further step forwards will help. Yelling at clouds won't.

[AlexSm]

!!! The Kratos for God of War (PS4) has 80,000 polygons in total, with 32,000 for the face – 30x what we had for the PS2 models. Kratos (PS4) uses over 140 textures. !!!!
========
Read the source carefully!!! Polygons, not triangles. 80,000 polygons, not 32 000 triangles. And 140 texture maps.
===
My comment does not cancel the fact that we need to optimize the grid for clothes in SL. And believe me the professionals are doing it and strive for it. But ... The shirt you bought is bad. Take the demo version before your next purchase. The limitations from LL are already there - go to the SL wikopedia and read them - http://wiki.secondlife.com/wiki/Limits 90% of the craters in SL are not professionals who learn using SL.