Jump to content

Too Many 1024 Textures and the NEW (please hurry) Land Impact rules


Chic Aeon
 Share

You are about to reply to a thread that has been inactive for 2071 days.

Please take a moment to consider if this thread is worth bumping.

Recommended Posts

31 minutes ago, ChinRey said:

Cpu load is something everybody have ignored until now. We've all simply assumed that client side lag is all about gpu and bandwidth. Unfortunately it's beginning to look more and more like the client's cpu can be a significant limiting factor for SL prformance and that is soemthing that really should be examined closer.

The SL client has always been heavily CPU bound, hence the typical recommendation when shopping for a PC being to buy intel, That straight-line single core speed counts.

The viewer has an awful lot of processing to do, downloading & processing the interest list, individual assets, textures, rendering prims, openJPEG/KDU, moving data about, processing miles of XML to render the UI (that slight pause when you get a new IM is an example of how heavy that is) and finding time to hold the GPU's hand. It's a beast and a lot of people approach viewer issues misunderstanding that downloading, file IO and putting an image on the screen are free.

Most modern GPU's are under utilized by SL, we eat all the VRAM but the actual 3D processing is rarely maxed out. For all intents and purposes my 650 is on par with my 780 in terms of raw framerate, the 650 is pegged to the max, the 780 sits half idle. Newer 10 series cards do render SL faster, but this has more to do with overall architectural changes and straight-line speed than increased capacity.

 

If the cache changes from LL include storing textures uncompressed then that will go a long way to help to remove one of the larger bottlenecks, assuming you have the cache on an SSD.

Link to comment
Share on other sites

26 minutes ago, CoffeeDujour said:

The client is in a hurry to get stuff on your screen and in this case, that is not the best option to take.

That's true and it's why I used the word "meaningful". Rendering the fitted mesh as static, off scale, offset and attached to an arbitarily selected attachment point doesn't make any sense.

 

9 minutes ago, CoffeeDujour said:

Most modern GPU's are under utilized by SL, we eat all the VRAM but the actual 3D processing is rarely maxed out.

And yet SL's resource accounting system does not in an way consider cpu load. That was a serious mistake by the old Linden Lab and one that will be really difficult to fix.

Link to comment
Share on other sites

1 hour ago, Love Zhaoying said:

Different answers!

Not quite as different as it may seem at first. Mine is the correct answer to your question but CoffeDujour's is still very relevant to the topic as a whole.

It's easy enough to demonstrate that the transparent triangles don't affect the calculated render weight btw.: Switch "Show avatar complexity information" on, wear a mesh body, switch on and off various clothes layer and watch as the render weight display doesn't change.

The LoD bug is the only one of the factors I mentioned that affects the calculated render weight. When it comes to the transparent parts and the other factors it's the opposite problem: they should have been included in the calculation but aren't.

Edited by ChinRey
  • Thanks 1
Link to comment
Share on other sites

1 hour ago, CoffeeDujour said:

Transparent faces on a single mesh don't so much, whole transparent objects as part of a link set did the last time I checked

The example I used had alpha masking, not alpha blending. Sorry, I forgot to mention that. My fault.

 

My poor abused alt, Lynx, wearing a Maitreya body with all clothes layers visible:

5b514ea9b35e3_Skjermbilde(1273).thumb.png.2561fb661c1f2d3ec9b456d4acc9cb0a.png

Calculated render weight: 94052

With all clothes layers hidden:

5b514edb606e1_Skjermbilde(1274).thumb.png.94de2154a4686f98cc476255c151a152.png

Calculated render weight: 94052 - exactly the same.

Setting faces to 100% transparency with alpha masking does not affect the calculated render weight.

If I had used alpha blending for 100% transparency it would increased the calculated render weight though, you're right there. That is another flaw of course but I'm not sure if it's in the calculation or the code that handles alpha blended faces.

I can't imagine any fitmesh makers use alpha blending these days anyway. The penalty for it is way too high and so easy to avoid. My second fitmesh example would have had a calculated render weight of about 20,000 if it really had used alpha blending.

Edited by ChinRey
Link to comment
Share on other sites

20 minutes ago, CoffeeDujour said:

Now try an object that is multiple linked parts with one set totally alpha, blended or masked.

Ummm, that's exactly what a mesh body with clothes layers is so I've already done that with illustrations and everything. (And just look at what my poor alt had to wear for those pictures - don't you think she's suffered enough already?)

Alpha masking did not exist back when the render weight formula was introduced so the formula doesn't know about it at all. As far as it is concerned, an alpha masked face is exactly the same as an opaque face.

Alpha blending did exist and as you said, it is seriously penalized. Add a single little face with alpha blending to an object and the base weight for the entire mesh (or prim or sculpt) is multiplied by 4. That is true even if it's set to full transparency which unless the render code is seriously bodged, actually should reduce, not increase the actual render load.

Edited by ChinRey
Link to comment
Share on other sites

On 7/18/2018 at 6:41 PM, Chic Aeon said:

I don't disagree with you in spirit but I have a hard time convincing myself that the majority (or even half - maybe a quarter) of content creators will redo and update as needed -- mostly because it seem logical that the folks that need to make the most changes are the ones that are purposefully making the oh so pretty but OMG heavy and difficult to render mesh.   They might -- if the edict was enforced -- change their methods in the FUTURE and that would be good. No argument there. 

I think you are misunderstanding me. I'm talking about hard caps here. Creators would have to become more mindful of optimization or no one would buy their stuff anymore. They'd be pushed aside in favour of creators who were willing to do so.

When I talking about a "grace period where the new caps are not enforced" that is a temporary period to let creators and consumers alike know what to expect. Creators who refused to take advantage of this grace period to update and redo and put out new, better made content would be a huge disadvantage against those who did take advantage of it once those new hard caps become enforced.

On 7/18/2018 at 6:41 PM, Chic Aeon said:

The others who I so hoped would make some changes, have not. These are NOT stupid people. They have made their choices and they are sticking to them.

They are, however, inexperienced and ignorant of good content creation habits. That's the thing, though. I firmly believe a part of SL's draw is that anyone can create content, you shouldn't NEED to be professionally skilled to jump into it. That's why SL needs better developed and thought out resource caps. An SL content creator doesn't need to know why a high-poly hat with 500MB worth of textures is bad if creating something like that results in an excessive point cost that would prevent anyone from actually being able to wear it.

On 7/19/2018 at 5:14 PM, ChinRey said:

Cpu load is something everybody have ignored until now. We've all simply assumed that client side lag is all about gpu and bandwidth. Unfortunately it's beginning to look more and more like the client's cpu can be a significant limiting factor for SL prformance and that is soemthing that really should be examined closer.

I don't think anyone is ignoring this, it's simply not something content creators can do anything about.

As an aside, I almost never see the problem with rigged mesh that you describe unless I'm in a sim stuffed to the gills with unoptimized content. Like everything else, rigged mesh loads faster when your viewer isn't trying to download and render gigs of data all at once. Hence the focus on working on the problems we can address. We might not be able to do anything about the CPU bottlenecks SL has, but we can ease up on creating content and environments that exacerbate the problem.

Link to comment
Share on other sites

On 7/19/2018 at 10:54 PM, ChinRey said:

Calculated render weight: 94052 - exactly the same.

Question, did you detach and reattach the body after changing the alpha states? If not then I believe that the viewer does not recalculate the draw weight.

I ran into this issue a while back when I was changing all my blended alpha attachment textures to masked and saw no change in my draw weight. I thought it was a bug until someone pointed out to me that I would not see the change until doing something to force SL to recalculate my draw weight (like detaching and reattaching the attachment in question).

Link to comment
Share on other sites

10 minutes ago, Penny Patton said:
On 7/18/2018 at 3:41 PM, Chic Aeon said:

I don't disagree with you in spirit but I have a hard time convincing myself that the majority (or even half - maybe a quarter) of content creators will redo and update as needed -- mostly because it seem logical that the folks that need to make the most changes are the ones that are purposefully making the oh so pretty but OMG heavy and difficult to render mesh.   They might -- if the edict was enforced -- change their methods in the FUTURE and that would be good. No argument there. 

I think you are misunderstanding me. I'm talking about hard caps here. Creators would have to become more mindful of optimization or no one would buy their stuff anymore. They'd be pushed aside in favour of creators who were willing to do so.

When I talking about a "grace period where the new caps are not enforced" that is a temporary period to let creators and consumers alike know what to expect. Creators who refused to take advantage of this grace period to update and redo and put out new, better made content would be a huge disadvantage against those who did take advantage of it once those new hard caps become enforced.

I understood you. I am just not convinced that the group of creators I am referring to would change OR lose a bunch of their audience -- even with hard caps. And honestly while I would like to see that I suspect the "hard caps" thing really won't happen. 

Do you make mesh?   I am thinking not but really don't know.  Redoing a good portion of past work would take more time than it would be worth for most creators. While I don't think "I" would have a ton to update since I have been paying attention, others would probably just throw in the towel with some expletives.  We are talking MONTHS of full time work to redo just a portion of stock of MANY creators. I can't see them doing that. I wouldn't do that. I would just retire and I don't shy away from challenges easily. 

10 minutes ago, Penny Patton said:
On 7/18/2018 at 3:41 PM, Chic Aeon said:

The others who I so hoped would make some changes, have not. These are NOT stupid people. They have made their choices and they are sticking to them.

They are, however, inexperienced and ignorant of good content creation habits. That's the thing, though. I firmly believe a part of SL's draw is that anyone can create content, you shouldn't NEED to be professionally skilled to jump into it. That's why SL needs better developed and thought out resource caps. An SL content creator doesn't need to know why a high-poly hat with 500MB worth of textures is bad if creating something like that results in an excessive point cost that would prevent anyone from actually being able to wear it.

Unfortunately while I would like to believe they are ignorant, I know for a fact that they are NOT. Many know exactly what they are doing and for some it is working well. Others, not so much --  but not because of the heavy mesh and or textures but because they haven't moved along with the times creation wise. 

I completely understand the "hobbiest creators" (I was one for a long time and likely some folks still consider me that even if I do not) inability to make righteous choices due to lack of knowledge. I am not talking about THEM at all. I am talking about some BIG brands at the top of the food chain. They are very talented folks. They have just decided purposefully to make things look the clearest and sharpest and texture rich with tons of details == as they can. And right now, that is working for them.  I like their items too. They are impressive, just not practical for anyone without a very hefty computer. 

 

Link to comment
Share on other sites

On 7/19/2018 at 2:45 PM, CoffeeDujour said:

As I understand it, CPU is an indirect issue but not the direct cause. There is often a large queue of objects to be processed and that postpones fetching/processing the rigging. So the viewer knows its an attachment, it knows the point ... and it knows the scale/rotation etc.. which gets binned when the rigging loads and the object snaps onto an avatar.

I've wondered about the order of processing there. I have an overly complex leather jacket which often takes 30 seconds to position properly, and a pair of boots which sometimes don't appear at all without an avatar rebake. Both have way too many triangles. Someone actually modeled each strap and rivet in the boots.

Link to comment
Share on other sites

2 minutes ago, Chic Aeon said:

I understood you. I am just not convinced that the group of creators I am referring to would change OR lose a bunch of their audience -- even with hard caps.

If there were hard caps on what your avatar could wear, then people literally would not be able to wear unoptimized content. The group of creators you are referring to would either be forced to change, or would lose their customers because no one is going to waste money on content they can't wear.

3 minutes ago, Chic Aeon said:

Do you make mesh?   I am thinking not but really don't know.

I've fallen out of it in recent years, but I have a professional background in computer animation that goes back nearly 20 years now. I've worked with videogame developers and in television animation.

I know a thing or two about it.

6 minutes ago, Chic Aeon said:

Redoing a good portion of past work would take more time than it would be worth for most creators. While I don't think "I" would have a ton to update since I have been paying attention, others would probably just throw in the towel with some expletives.  We are talking MONTHS of full time work to redo just a portion of stock of MANY creators. I can't see them doing that.

And there would be plenty of creators to fill their absence. Yes, I'm aware of how much work it would be (although, in many cases it actually would not be quite that much work). And this is also why I proposed a grace period. I'd actually have such a grace period last at least a year. Plenty of time to update some popular items and produce some all new content to be ready in time for the new caps.

11 minutes ago, Chic Aeon said:

I completely understand the "hobbiest creators" (I was one for a long time and likely some folks still consider me that even if I do not) inability to make righteous choices due to lack of knowledge. I am not talking about THEM at all. I am talking about some BIG brands at the top of the food chain. They are very talented folks.

Most of the big brands at the top of the SL content creation food chain fall squarely in the "inexperienced/unskilled" category. Yes, they are very talented, but if they are not optimizing their work that shows right there a lack of skill and experience. Their experience is limited to SL where no one has ever explained to them why they need to optimize their work. This is not a knock against them, again I find the fact that anyone can create content in SL, regardless of skill and experience, to be a very good thing. LL just needs to implement some better tools to guide people towards better habits.

And if they were honing skills in optimizing (which isn't difficult to do if you already know how to create content in the first place) then they would quickly realize that optimization does not mean sacrificing detail and quality.

Link to comment
Share on other sites

Caps per item would lead to "fancy dress part1" and "fancy dress part2" .. combined would be worse than just a single item.

Caps per avatar wouldn't affect creators at all .. it's the customers problem to find items.

Newbie creators are not the always the problem .. half the time its people who know better but don't care or those who just recycle assets from other platforms.

 

  • Like 2
Link to comment
Share on other sites

6 minutes ago, CoffeeDujour said:

Caps per avatar wouldn't affect creators at all .. it's the customers problem to find items.

Try selling simple SL furniture that uses up hundreds of Land Impact per chair or couch. You'll have a hard time making many sales, no matter how good your content looks. Try selling a mesh house for a 512sq.m. parcel that uses 2,000LI. You'd run into the same problem, no one would want your houses.

This is because content creators are already affected by the hard caps we have on rezzing in-world content.

Let's say avatars get a 10 point resource limit.  No one would ever buy shoes that use 100 points. No one could even attach them. No one would buy shoes that used 10 points because it would be the only attachment they could wear. If you were trying to sell your shoes you would try to make them look as good as possible, while using as few points as you could manage.

Creators would be very much affected by a hard cap on avatar attachments.

Edited by Penny Patton
Link to comment
Share on other sites

In the case of buildings, typically the item is made and the uploaded .. and then you roll the dice and see what the Li is. If unacceptable, fiddle with the LOD meshes to get it under some arbitrary limit. Caps will nto affect this workflow at all, and if cutting out the lower 2 LODS is whats required to get under the cap that's the route creators will take. Pressure will then be placed on users to compensate by ramping up settings in the viewer.

This is exactly how we got into this mess with sculpt maps. Creators over cooked them to get under an artificial psychological cap (1 Li or bust) and then firestorm doubled the default LOD so the damn things would render.

Link to comment
Share on other sites

1 hour ago, CoffeeDujour said:

In the case of buildings, typically the item is made and the uploaded .. and then you roll the dice and see what the Li is. If unacceptable, fiddle with the LOD meshes to get it under some arbitrary limit. Caps will nto affect this workflow at all,

The important part is that it forces content creators to try and get their LI low. THAT IS THE EFFECT. Put a similar cap on avatars and it will achieve the same effect. Content creators will be forced to try and make content that looks as good as possible while still coming in at a low point cost.

1 hour ago, CoffeeDujour said:

and if cutting out the lower 2 LODS is whats required to get under the cap that's the route creators will take. Pressure will then be placed on users to compensate by ramping up settings in the viewer.

Let's not run in circles. I've already said that there are problems with how LI is calculated, problems that Linden Lab is working on right now to address. I've gone to great lengths in this thread to point out the obvious, that whether or not a resource cap is effective depends entirely on how well it's thought out and developed.

Just because LL did it poorly when they introduced Land Impact (and some of us tried to tell them this before mesh even reached the main grid) does not mean that the idea of a resource cap is useless.

Link to comment
Share on other sites

But your solution simply repeats the situation we had with sculpts.

Vendors will skimp and hack to get their work into SL, that then gets passed onto the end user who in turn will end up further crippling the viewers LOD handling.

Trimming the complexity of a model down presumes the person uploading the item actually made it themselves and has the required skill.

Link to comment
Share on other sites

9 minutes ago, CoffeeDujour said:

But your solution simply repeats the situation we had with sculpts.

I'm not even sure what you mean by this. The problem with sculpts was twofold:

  • First, they only cost 1LI despite being quite a bit more resource intensive than that to render.
  • Second, the issue of people cranking up object detail to prevent them from breaking down over short distances.

Neither of these are an issue with what I suggested.

Link to comment
Share on other sites

Sculpts were never intended for the kinds of complex geometry that they ended up being used for, so where someone should have used 2 sculpts they would use 1 and push the amount of vertices that the format allowed to the absolute maximum. Obviously these tortured sculpts didn't render unless you were sat on them and degraded to a screwed up ball. The solution was to cripple the LOD handling in the viewer to prevent them from degrading and allow them to be seen at reasonable distances, when this failed it was combined with bounding box hacks to make the viewer think they were double their actual size on screen.

This is a direct result of creators needing to get their work in at 1 prim. The perception being that no one would buy a 2 or 3 prim object when it could be made for less. 

A hard cap on object detail will result in the uploaded mesh being over cooked in the uploader. So rather than ditching the lowest LOD we will see things ditching the bottom two. Hard cap achieved!

 

Q: How do I determine good content from bad content ?

A: Inspect the object and apply a number of subjective judgments about its texture use, LOD mesh quality and so on.

A: Does the box have a pretty picture and is it for sale at an event by a fancy brand.

 

Q: But the stuff I just paid for doesn't rez!

A: Don't buy junk over cooked objects, not rewarding bad content will force creators to make better content.

A: You need to edit a debug setting and it will be fine.

 

Q: Why does SL run so badly?

A: Because it's old and junk, LL are quite evil you know. If only they would use a real game engine.

Edited by CoffeeDujour
  • Thanks 1
Link to comment
Share on other sites

43 minutes ago, CoffeeDujour said:

Vendors will skimp and hack to get their work into SL, that then gets passed onto the end user who in turn will end up further crippling the viewers LOD handling.

i mentioned somewhere way up earlier and I will mention it again as its relevant to this topic

when we are min. charged for LOD levels whether we use them or not, then we tend to use them

is different formulae that can be used for this. One method (mentioned earlier) using 6ths being:

LOD1:  calculate
LOD2:  min. cost 3/6 of LOD1
LOD3:  min. cost 2/6 of LOD1
LOD4:  min. cost 1/6 of LOD1

calculate LOD1 as 0.5 say, then LOD2 = 0.25, LOD3 = .167, LOD4 = 0.083. Total = 1.0

if not 6ths then:
7ths:  4/7 + 2/7 + 1/7
8ths: 5/8 + 2/8 + 1/8
10ths: 6/10 + 3/10 + 1/10
etc

when a LOD2->LOD4 model is calculated to be more than its min. fractional cost then we are charged for the overage
 
however, when LOD4 costs at least 1/x of LOD1 then not using the 1/x gains us nothing over those who do

i dunno if LL have considered min. costing in their current deliberations. I would like to think tho that they have at least given it some thought  

Link to comment
Share on other sites

3 hours ago, Penny Patton said:

Question, did you detach and reattach the body after changing the alpha states? If not then I believe that the viewer does not recalculate the draw weight.

No, I didn't think of that but I tried again now and get the same result: faces set to 100% transparency have exactly the same render weight as opaque ones.

But of course, the texture will have to have an alpha channel to start with for it to be alpha masked so I did a dfferent quick test jsut to make absolutely sure.

Here are a couple of cubes textured with two different 4x4 pixel textures, one with and one without alpha channel. The ones using alpha blending or transparency added in the edit window, have 337 RW, the non-transparent ones and the ones with alpha masking (and no transparency added) all have 277.

For those who can't get enough details, from left to right:

  • Alpha texture, alpha mode none, RW 277
  • Alpha texture, blending, RW 337
  • Alpha texture, masking cutoff 50, RW 277
  • Alpha texture, masking cutoff 0, RW 277
  • Non-alpha texture, transparency 50%, RW 337
  • Non-alpha texture, no transparency added, RW 277
  • Alpha texture, masking cutoff 50 but with 50% transaprency added, forcing a switch to blending, RW 337

5b52f0a2d0468_Skjermbilde(1275).thumb.png.2a63c1b4a385b56dc0692d4888f22e35.png

Link to comment
Share on other sites

How come every time somebody starts a thread about texture lag we end up discussing mesh geometry and every time somebody starts a thread about mesh geometry lag, we end up discussing textures? ;)

Never mind, that was a rhetorical question. Here's some food for thought.

This is a cylinder prim, 10 m long and 0.2 m diameter. It looks like a hexagon because cylinder prims have lower LoD than any other objects. which means it's rendered with 24 rather than 96 triangles at 5 m distance (5 m from its center that is, so at point blank range from the end). According to Grumpity Linden, this done to reduce the gpu load and he says it's amazing how much we save that way and that it's well worth the fairly minor reduction in visual quality.

5b52f826a7c58_Skjermbilde(1278).thumb.png.7b310a3336b221c81d8564710e92a9c7.png

 

This is a skybox. It is made with ultra-strengthened LoD models which means the viewer of the neighbor in a skybox 100 m away will handle all of its 10,612 triangles and not jsut the ten or less (depending on view angle) that are actually drawn on the screen. According to Patch Linden, this is to reduce the gpu laod since it means nobody will have to increase their LoD factor.

5b52fa0e78e0a_Skjermbilde(1277).thumb.png.5050b655a1a7a2ec61111b1477abaa8d.png

Now, we can of course use this to make fun of some Linden since when two of them so obviously contradict each other, one of them must be dead wrong. ;)

But there is a more important point here: Good LoD is not about strengthening the LoD models as much as possible, if it was that easy, the wouldn't need LoD in the first place. Making good LoD is a balancing act: you want to reduce the complexity for each level as much as possible with no changes of the actual look - or at least without any noitceable changes. No landimpact calculation formula can catch that and besides, much of the poor LoD we see today are not caused by people trying to save LI, it's caused by people trusting the default LoD models of the uploader. It is possible to get a more realistic balance between the different models' significance than we have today but it will never be precise enough, we need more. And that's where it gets complicated.

 

Link to comment
Share on other sites

On 7/19/2018 at 12:50 AM, Theresa Tennyson said:

The reason that the server asks multiple viewers now is to make spoofing the number largely impractical - you can turn off yours, but the avatar next to you will send your numbers whether you like it or not.

"Gee, all these people around me have an ARC of 9'999'999!"

Link to comment
Share on other sites

7 hours ago, CoffeeDujour said:

This is a direct result of creators needing to get their work in at 1 prim. The perception being that no one would buy a 2 or 3 prim object when it could be made for less. 

A hard cap on object detail will result in the uploaded mesh being over cooked in the uploader. So rather than ditching the lowest LOD we will see things ditching the bottom two. Hard cap achieved!

You're still running in circles. The problem here is that Linden Lab made it so people could get lower LI costs by nerfing the lowest LOD levels and people exploited it. Don't take a mistake LL made and just assume it has to be that way. Because it doesn't, and as long as you make this incorrect assumption the very core of your reasoning, every conclusion you're going to come to is going to be flawed because of it.

7 hours ago, ChinRey said:

No, I didn't think of that but I tried again now and get the same result: faces set to 100% transparency have exactly the same render weight as opaque ones.

Ok, I think I was overlooking something last night. I believe, and anyone who knows better can feel free to correct me on this, that LL did something in recent months so that any object set to 100% opacity does not get rendered at all (until you highlight transparency, which is why 100% transparent objects now pop in with a delay when you do this). If this is correct, it would make sense that they do not add to your render weight as you would assume blended textures should.

 

Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 2071 days.

Please take a moment to consider if this thread is worth bumping.

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share

×
×
  • Create New...