Jump to content
Chic Aeon

ARCTan -- 2020? What do you want to see?

Recommended Posts

According to current sources ARCTan, a project to update rendering cost, land impact, and related things like mesh upload costs, is in progress once again after some hiatus time.  I am posting this in general as it concerns everyone -- not just creators, but purchasers and even folks that never buy a thing but wander amidst the pretty regions. 

 

There were plenty of problems when mesh was new. Many of those have been solved or a reasonable work-around found.  But much like office furniture is often  made for 6 foot 5 men, the uploader defaults were designed to work well for simple medium sized buildings.   Nothing wrong with that -- except of course there are many other mesh items being made besides those middle sized rectangular abodes.  There are smart ways to work with SL mesh and not so smart ways, but there is no real incentive --- aside from having our SL world run smoother -- for creators to make the more "industry standard" choices -- or even get close.  

 

I went shopping today. I WANTED to buy some things.  I found some lovely decor items.   But?   

 

But a medium sized vase with simple flowers checked out at 92,000 triangles.  It could have looked as good at 24,000 triangle -- likely less. We won't even talk about the number of 1024 textures on that small item (sigh).  STILL, it was only 1 land impact.   So in many people's eyes a bargain. 

 

I TPed home even though I WOULD have purchased it had it been able to boast a lower poly count. 

 

 Two other decor items had the same high triangle and texture issues. Again, TP to home. 

 

If I take myself back to my few years at Inworldz, I can assure folks there is a more sane way to measure land impact.  Inworldz threw out our current  "small item" bonus that pretty much overlooks the complexity of an item if it is small. They made a different system even though the uploader looked the same. Many items that would be 1 LI in SL (these were my things so low tri count to start with) became 2 li there. BUT, large items weren't penalized as they are in SL.   It was based more on the triangle count than our current Linden uploader.   

 

More is not better so far as mesh goes.  A 125,000 triangle ashtray that was made to be a replacement for a photograph is not appropriate for our platform. And still we have a preponderance (it seems to be increasing) of heavy mesh  -- mesh that slows down computers, that of both the item's owner and any nearby residents. There needs to be SOME sort of incentive to stop making such heavy mesh.  Basing a big part of ARCTan on triangles (or vertices) count would be a good start.  

.   

I am hoping that ARCTan does indeed shake up the world a bit. We've been going down a bumpy road for a long while now. Time to make the road a little smoother. 

Share this post


Link to post
Share on other sites

Ideally I want to see a resource management system that reflects the actual load as closely as it's possible to estimate it. I know that's not going to happen but hopefully we'll get closer than we are today.

  • Like 1

Share this post


Link to post
Share on other sites

The viewer and GPU can draw some number of triangles per second. Divide that by the minimum allowed frame rate, at least 20, higher for gamer hardware, and you get the rough triangle budget for a frame. For textures, the general idea is to have an roughly equal number of texture pixels per screen pixel across all the textures. Those set the general objectives for the viewer.

The viewer can adjust the texture resolution and the mesh level of detail displayed. So it has the power to make that happen. Right now, the strategy for which level of detail to show is based on fixed distances, but it could be dynamic on a per object basis to achieve those goals.

So, if the viewer is set up like that, what does that mean for content?

In a crowded scene, you're only going to see the highest level of detail if you're really close. How close depends on how busy the scene is. If you're alone in a walled room with something elaborate, you're going to see it at full detail. Hair Fair, now going on, is set up that way, with each exhibit in its own little hut. If everything is out in the open, as at Fashion Week, almost everything will get kicked down a LOD or two unless you're really close.

If Arctan works like that, then levels of detail work like this:

  • High: as fancy as you want, but the more complex you make it, the closer people have to get to see it at full detail.
  • Medium: had better look good, because that's what people will usually be seeing. But bake everything you can into flat textures.
  • Low: should at least look OK.
  • Lowest: at least cover the entire area of the object. Triangles and holes means loser designer.

This should apply to avatars as well as fixed objects. It already applies to animesh, so switching avatars over can't be hard.

Share this post


Link to post
Share on other sites
1 hour ago, animats said:

If everything is out in the open, as at Fashion Week, almost everything will get kicked down a LOD or two unless you're really close.

That's another problem with the collapsed LoD models of course. No matter how high you set the viewer's LoD factor, if the scene is too busy rendering will be stalled at lower LoD levels no matter how your viewer settings are.

Chic asked what we want to see with ARCTan. What I hope to see, is a bit different. I've written about it at some legnth in two other active threads so I won't go into detail but there is no way we can get good LoD handling in SL without swap distances being adjsutable for individual objects, there is no way we can get a realistic load monitoring as long as we have the RenderVolumeLODFactor cheat and there is hardly any chance ARCTan will correct either of those old mistakes. Nor is there much chance of a proper fix for the fitmesh LoD bugs. So nothing will be done about three of the four major LoD/lag issues. But maybe they can do something about the fourth - the total lack of texture load management - and at least I expect a few minor tweaks to make the best of the situation.

Share this post


Link to post
Share on other sites
1 hour ago, ChinRey said:

the total lack of texture load management

Oh, there's texture load management. The trouble is, it's spread all over the texture code in the viewer, and you can't see what it's trying to do. Texture control policy needs to be centralized in the viewer so different things can be tried.

Share this post


Link to post
Share on other sites

Notes on LOD ratios.

Here's some info from the Unity manual for designing game objects: "Many LOD Groups use three levels, where LOD 1 is active when the GameObject fills between 25% and 49% of the screen height, and LOD 2 is active when the GameObject fills less than 25% of the screen height."

So where does that take us? That says LOD distances have ratio 1, 2, 4, 8.  Or, if we drop to Medium LOD at 5 meters, we drop to Lowest at 40 meters. OK.

At twice the distance, an object occupies a quarter of the screen space. So triangle ratios should drop by a factor of 4 from level to level. Animesh drops by a factor of 2, regular objects drop by something larger. Suppose we went with 4 for most objects. So we have 1, 1/4, 1/16, 1/64 triangle quotas. A 60,000 triangle shoe (sadly, quite common) drops to below 1000 at distance. 0.1% of the scene budget. (1 million tris per screen is about typical before the viewer on a midrange PC drops the frame rate below 30, so that's a reasonable number for planning purposes.)

This kind of makes sense. People want super-high detail at close range, and creators oblige by giving it to them. That implies big drops at distance.

Discuss.

 

  • Haha 1

Share this post


Link to post
Share on other sites
23 hours ago, animats said:

his kind of makes sense. People want super-high detail at close range, and creators oblige by giving it to them. That implies big drops at distance.

Yes they do and in my mind (depressingly so as I am a "shopper" now and not a creator LOL) they are giving us things that are TOO detailed. Well let's put it this way, it would look JUST as good with a much lower high level starting count.  I have been going to venues this last couple of weeks and inspecting. And yes, I DID WANT TO BUY!!!!   

 

But I cannot make myself (or let myself) buy a small decor item that is 270,000 K with 27 prims. It just isn't "THAT" pretty LOL.  So I am getting more depressed than usual over all this. Along with the LOD fix for the LI cheat, there really needs to be some sort of triangle count cost.   

 

For example. Last night I saw a nice looking -- let's say "outdoor decor item" at a popular new tourist stop.  It was one land impact with  over 22,000 tris.   I have the same type of item and mine is 2,000 tris. Now the 20,000 item is definitely prettier than mine LOL, I am not arguing that they are equal, but is it ELEVEN  TIMES NICER?   And I went back to check and they are both the same land impact which is ONE.  This isn't a small item at all. I cannot understand why they are the same land impact. Such a faulty plan.  There could easily be a cheat that I am not aware of -- not being a "cheater" :D at heart.  

 

Happily I have found a few creators that are thinking sanely (my version anyway) as well as some old timers who are making "higher tri count items than their historical goods but nothing like some folks.      So I can do a little shopping.  But I am looking forward to the change when it comes --- assuming it fixes some of these inequalities.    

Share this post


Link to post
Share on other sites

I don't want much from it, outside of a complete transparency on content.

I'm not getting the hard caps i've been vouching for anyway.

Share this post


Link to post
Share on other sites

What do I want to see? Well, as I am all for improving SL, even if it means redoing all my stuff and taking a hit on multi-scripts* as has been mentioned elsewhere, I will adapt to whatever comes down the pipe BUT I want full detailed lab-sourced documentation and guidelines, Lab sourced, not knocking the great advice given here in the forums by any means but I want the Lab to write it and sign ioff on it. Great recent example is the bacon mesh update - not my biz but detailed enough to work with. That was good to see.

And please can we have back the text records of the  user groups involved (much as I would like to attend  I can't) like there used to be. Again, Inara does a stellar job but I want it official. And yes - text.

*I try to stick to one script per where possible using the linked prim fast thingy but there are some missing bits - is why I had to resort to multi sound players for one.. Have tried to search through the jira to see if any movement but despite speaking several dialects of SQuirreL it is - frustrating.

Share this post


Link to post
Share on other sites

It desperately needs to take geometry into account, for one. According to the current ARC, a 500k polygon body is no worse than a 20k one, so I have to manually derender everyone wearing those.

 

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.


×
×
  • Create New...