Kyrah Abattoir

The problem is the motivation.

The thing is that this memory limit is ONLY for textures, SL loads a lot of ohter things into vram (meshs, prims, etc etc...), I'm not sure why these don't have a limit either, but I assume there is a design reason somewhere that makes sense. Your vram is also used by everything else that runs on your computer and does even modest 2D/3D operations, so the vram is not for the SL client's exclusive usage. I love those people who always consider that problems that don't affect them are "other people's personal problems". Exactly. Limits have to exist because computers are not limitless. Anyone who does at least a little bit of programing knows this. Variable limites: they all have maximum values, (we just hope that we never hit those). Diminishing returns: The larger a data structure becomes, the slower it performs. Clock speed: There is only so much that can be processed per second. Parallele processing: double the cores is not the same as doubling the speed, you can only parallelise processes that do need to be executed in a specific order, otherwise you gain nothing, it's easy for a GPU to process each pixel in parallele, it's what they where made for. Not so much for CPU-type instructions that are typically executed in a sequential fashion (if X, then do Y and Z). Some games outsource things like physics calculation, AI pathing, or file loading on separate cores but that's typically the extent of it. I'm hoping that the new cache system they are working on will be intelligent enough to only load the requird texture resolution to obtain a consistent image. This would at least mitigate large texture usage in SL, by not loading the ram with more than the needs of the current scene (oh those 1024x1024 wedding rings...) I agree, the problem is to figure out what the right incentive is, the current mesh limitations encourage people to dump lods because they get a lower complexity & LI out of it, animesh attempted to solve the issue by making lods "free" as long as they are reasonable, but it doesn't solve the main issue: Lod models are important for a quality experience, but remain the submerged part of the iceberg, people don't realize they are looking at a purposely bad lod model, they just blame Linden Lab instead and crank up object details to "make it go away". Making Lod models is work and even if they come with a free triangle budget, unless customers begin demanding them, only a handful of creators will actually bother. In the end, Lod models can also cost prims, and that's another incentive for creators and their customers not to care. And yeah I enjoy working within limitations as a creative challenge too. But I also create because it's a neverending source of enjoyment and pride, the money is just an added bonus.

Skinny Dip

The "fix" is opening a Jira about it and asking Linden Lab to fix their shader, cause this is not normal.

It's the same as rigging anything else really.

Principled BSDF isn't too bad for what i've seen so far. Layers are gone, existing layers are converted into collections.

It's also broken and non commital.

To be honest they should require from creators to enter the following: Tri count used by each lod. Number of textures and their size at any given time. Number of scripts. I like your idea of star rating. But if an item has no rating just put a big exclanation mark "the creator of this item has not supplied rating informations, buyer beware." However I suspect showing the tricount of each lof won't help depending of the calculation method, showing the texture area might, but it also doesn't account for whether textures might be recycled across multiple items. Likewise, script count, while a useful indication, doesn't tell the full story.

I have nothing against "pretty", we all want that. I would praise everyone who can do "pretty" if it wasn't drowned by the inner voice screaming "WHAT WHERE THEY THINKING?". And yeah a good chunk of them don't care because it's not about 'good' it's about 'fast', and it's also about making the most money from the least amount of work. You're not new in this topic and despite all the explanations that where given by me, and people who are far better at this than me you are still blaming LindenLab? And what if they can't? What if there is no solution that they can implement because your fantasy of making SL look like a Pixart movie, and doing that at a reasonable framerate is simply not possible? Disney is still using render farms that grills top of the line hardware 24/7, to produce a single frame of movie, in a few HOURS. The only people whose content would become worse with sane limitations are those who refuse to get better at what they are supposed to be good at. Everyone else will do just fine.

Correct. I should have specified that.

Rigged meshs are scaled according to bone length, a shorter and narrower avatar will lead to a shorter and narrower mesh. Does mayastar allows you to work using your own body shape at all?

Ah yeah sometimes it's a good idea to think ahead in terms of unwrapping as it can save you time and also texture count/size. The typical workflow places unwrapping after modeling but you can absolutely do both at the same times. To make things like hair and fur it's actually recommended because sometimes the textures should adapt to the UVs, while in other cases the UVs should adapt to the textures.

That's how animesh handles costs. Most SL users aren't tech savy, might have never seen a texture map or don't even know that lod models exist. Most people want things that look pretty, the problem is how to showcase problems in the products they purchase, and how to make it clear that those problems are not bugs/errors brought up by SL itself, but mistakes and/or intentional shortcuts taken by the creator.

We have a situation where people can't even run ALM with shadows off, which every GPU of the last decade supports. But in SL clients already choke under texture and polygon bloat. Optimizing games, specifically games where many characters made of many different parts (your typical mmo) are visible at any given time is worth an entire book on the subject. Draw call reduction, texture sharing, shader optimization. SL Is a generic platform, there is no specific optimization that avatar attachments can be boiled down to that will just "work". Avatars in SL are absolute behemoths compared to what games typically have to handle. Games typically get around this by fusing meshs together on avatars and condensing materials and textures to reduce the number of passes required. But hey, SL avatars can change in many many ways, at will and many times per second if needed (I'm looking at you stop motion pets!). Avatars in SL most likely take a good hundred draw call each, (and that's being very conservative) Your optimization/caching system has to overall take less time than the time the stuff you are trying to optimize is taking. This is not one of those cases.

You grossly overestimate modern computers. Sure the viewer is probably not the finest piece of code out there, but the sourcecode is available for anyone who cares about fixing those performance issues. Improvements do get rolled out fairly reguarly, the viewer from 2004 would be completely unable to cope with the SL content of 2019. But there is only so much that the client can do in a world that has to be assumed 100% dynamic. Most games that aren't minecraft use prepared assets and careful ressource management to run properly. The N°1 killer with SL is ressource misuse, it doesn't matter if you have 32gigs and an I7, performance cost on data structures doesn't scale in a linear fashion. If you study next gen 3D models from games and compare them to SL ones you will see that there is a massive difference between the two, they typically achieve a much better aspect with only a fraction of the texture/mesh use. They also have multiple versions of models, cutscenes vs game sections for example (halflife 2 has this extremely detailed version of the gman for the opening cutscene that's used only for that)

Could you imagine the outrage? a vendor that only accept your payment if you are sitting next to it :3

We are all ears.

Time to either start my own grid, or work full-time on my game.

Phone Home

There was some serious thought put into this one at some point, what if instead of a copy, you where wired to a machine and your neural connections progressively moved over to the machine for example?

gift horse

Yeah I think the lod swap will be a lot less obvious if the final "tube" remains the size of the inner collumn.

It's a lot more than a list of "steps". How much experience do you have in 3D modeling (if any)

Manually degrading your mesh by removing edge loops usually gives you the best results, but you will always get a certain amount of distorsion after a point, but it doesn't matter if it's at a distance where you cannot notice it anymore.

If you want to make impostors you essentially need to "reserve" a material for that, hide it on a lone triangle on your high lod, and show it (and hide the others) on the low lod. But usually you can reuse your existing UVs/textures on your simplified lod models, it's just that in some cases, making an impostor texture can be more efficient.