Jump to content
The Dark Mod Forums

Poll on Author awareness of Level Of Detail models in TDM


What is your take on LOD models  

10 members have voted

  1. 1. Authors, do you use, are you aware of LOD models in TDM?

    • wut is LOD?
    • I know TDM has them but I can't find them
      0
    • I use them, wish there were more.
    • I forget TDM has some when mapping.
    • I know they are there but don't use them.


Recommended Posts

Basically I'm curious who is aware of LOD models in TDM. If you guys realize we have them. If so, do you use them? Do you know where they are? Can we find a better way to manage them so authors can use them easily?

Dark is the sway that mows like a harvest

Link to post
Share on other sites

Shouldn't there be an option for knowing where they are but not using them?

Link to post
Share on other sites

I know they are there, but I have never used them. Transaction rain patches were auto-hidden when the player was far-away, but that isn't a LOD model, but rather a particle saving trick.

 

LOD models are probably useful for large outdoor areas we have not yet seen. Standard TDM locations (interiors, cramped streets/alleys) don't need them much.

 

Also there seemed to be so few LOD models, it is probably up to the mapper to create their own. For example, build a Dishonored style streets, with building facades that simplify a lot when the player is far away. I do not know if D3 engine can survive that kind of scenes, even with good LOD models.

 

Only one way to find out: we need that exterior area challenge contest. ;)

Clipper

-The mapper's best friend.

Link to post
Share on other sites

added...

 

True, they are great for high detail scenes. Large spaces, etc...

 

BUT, that is NOT the main point of them anyway.

 

The main point is that you can use the entity instead of a model. And when a user has their settings on HIGH, then they see the highest poly models up close no matter what. If they have the details set on medium then they see highest up close ( but the details will drop at distance). If they have settings on low they will never see the highest poly models for performance on lower end systems.

 

If you just use models you either use a medium detail and nobody ever sees how pretty TDM could be. Or you use the highest and low end users suffer in your map.

 

Tels did a great job of scaling detail based on system settings AND distance.

----------

 

But realistically it would be best if all models had them for performance. Not saying that anyone is ready for that task, it can be hard to recreate models in higher versions, takes time...

 

But if they don't get used what's the point in having them.

 

The pillars I'm making for example, takes a lot of extra time to do all the extra models and stages. Then if they just end up sitting in a folder somewhere and never used...

 

We can always keep upgrading and putting the highest details in the models folder, so TDM always looks best. But that's not best performance wise. (but TDM is performing pretty well these days)

---------

 

Authors can make their own LOD's, shadowmeshes for terrain , etc... for better performance. But that only goes so far in the editor. And that's for stuff they create. Imagine trying to make a lower poly version of a standard object in TDM just for your own map. That's a lot of work to add LOD for one object (and honestly in most cases just wouldn't look good with editor tools. To get good uv mapping you're almost required to use patches and you can get much better more optimized shapes in a modelling program).

----------

 

I see no reason why TDM can't handle large areas like in dishonored. The city map I am making has some pretty large long views. However I haven't added tons of details yet and have optimized them pretty darn well anyway. (remember too though that Dishonoured has load zones... In the street of the second map when you can go into the doctor or the place across the street, each location is in a new zone. in TDM you would open a portal and go through a door, possibly leaving it open and having an even larger space to render)

 

Doom3 handles polys pretty well (and open spaces for that matter), that's really not a question. It does come down great to number of overlapping lights and searching ai that bogs down performance.

 

Even with LOD's we have stages:

 

1:Hi poly : lots of tris but lower poly shadow mesh (maybe even the low model or the low models shadow)

2: medium poly : still, shadow mesh is lower than model.

3: low poly : very low shadow mesh

4: lowest : shadow can be turned off completely...

5: turn off model completely.

 

Not all the models conform to that necessarily but those are the types of stages that can be set. Even just have LOD's for regular models that turn off shadows at a long distance can help perf. (of course the mapper can override this if the shadow is obvious).

Dark is the sway that mows like a harvest

Link to post
Share on other sites

In my opinion the main issue is that they are no longer models, but entities.

 

So when decorating a scene who thinks of using an entiy? If it's a light, sure. But if it's a pillar... just a static object that doesn't appear to 'do' anything.

 

And seeing as how few of them there are it seems they'll only be used because someone knows that particular model is optimized.

Dark is the sway that mows like a harvest

Link to post
Share on other sites

I would use them for big outdoor area's if we have more of them. Eg tree's and bushes, we dont have nearly enough variety IMHO.

 

Yes but now you're talking variety. That's more models + lod's.

 

I was thinking about adding more branches to my pine tree models so they would be fuller and look better up close/on high settings.

But this harkens back to the original problem:

 

1: I either rename so the higher poly models would replace the existing ones (making existing maps look better, but taking a possible slight hit in perf ... they will have the same shadow meshes) and use the existing ones as low detail replacements (would only be seen in new maps that use an entity).

 

2: I do the work, stick them in a folder where they will never be seen. They MIGHT be used in future maps as entities.

-------------

 

It's not so much about adding more variety at this stage, as much as it is about making TDM more beautiful and optimized.

 

Of course any future models are more likely to come with lod's right off the bat anyway.

Dark is the sway that mows like a harvest

Link to post
Share on other sites

Sorry for spamming posts but I think these things need to be easier to find

============

 

So a question for mappers on finding them:

 

Is the current local bad.:

 

models/misc/lod/architecture/pillars (standard model directory under lod). IMO this hides them out of sight/out of mind.

 

whereas

 

models/architecture/pillars

models/architecture/pillars/lod

 

would show you when browsing for models that the pillar you are using MIGHT have an lod entity/models. At which point you can look and see (if there are no LOD models there won't be an lod folder in said directory). If it does you can create an entity instead.

Or, if you just refuse to use LOD (don't see why but for sake of argument) then you can choose to either force the highest or lowest detail version on the map. (and in some rare instances you might prefer a lower poly version [ say a 6 sided fountain as opposed to a very round one).

 

==============

 

I'm bringing this up now because we do have very few models. If we choose to fix the directory now we can leave the existing ones in place (to not break maps) and we can place copies in the better location. it will leave a tiny bit of bloat but not much, if we do it NOW.

 

it will also make LOD more obvious NOW, which will start to improve performance now not later or ever and help people use and understand the system we have available to them.

Dark is the sway that mows like a harvest

Link to post
Share on other sites

I skipped the posts here to being able to make a neutral post:

 

I know there are two things I heard of but never saw them / got used to them. One of them is placing objects randomly within an area and the other is placing an object in different shapes. So from reading the topic LOD means level of detail and this is the one that places a high poly model if you are near it and a low poly one if you're far, right? :)

 

Voted "wut is that" because I never saw this and never used this sadly.

"Einen giftigen Trank aus Kräutern und Wurzeln für die närrischen Städter wollen wir brauen." - Text aus einem verlassenen Heidenlager

Link to post
Share on other sites

I skipped the posts here to being able to make a neutral post:

 

I know there are two things I heard of but never saw them / got used to them. One of them is placing objects randomly within an area and the other is placing an object in different shapes. So from reading the topic LOD means level of detail and this is the one that places a high poly model if you are near it and a low poly one if you're far, right? :)

 

Voted "wut is that" because I never saw this and never used this sadly.

 

Well, random placement is SEED, which is tied into LOD.

 

Yes LOD is basically high level up close and low further away. But also changes due to menu detail settings (lower for lower end users)

 

I know they exist.

 

I don't know how to use them.

 

I don't know where to find them.

 

That's what I am afraid of. (my posts above explain), but we need it to be second nature for mappers IMO.

 

=======

And hopefully my posts are neutral enough. just trying to find the best option for TDM/mappers, personal bias aside.

Dark is the sway that mows like a harvest

Link to post
Share on other sites

It basically comes down to a cost/benefit analysis for me.

 

The cost is the amount of extra time it takes modellers to make LOD versions, and the issue of getting mappers to use them (and the potential cost of see Rage-style 'popping' in TDM).

 

However, the benefit is unclear to me. We know shadow-casting polys can cause performance drains, but everyone always says that D3 can handle non-shadow polys with ease. Is there any evidence that reasonably high poly models with low poly shadowmeshes actually do cause significant slowdowns to performance? Especially if those shadows can be turned off if they're far enough away?

Link to post
Share on other sites

I'm also not sure of the benefit gained by this approach. In my last FM the grass outside consists out of roughly 2000+ grass entities. I mean, each single one may not have a very high poly count, but in the sum ... . I just set them to noshadows and disabled clipping IIRC and there never occured any performance problems to me.

 

Personally I think that the highest performance impact is caused by lots of shadow casting lights (in combination with a high poly count especialy) and particels. I don't want to say that this approach is a dead end in general, but especially when thinking of the time needed to create all those models it may be a waste of time.

 

In addition I don't see any reason for FM's with large outdoor areas as mentioned always. This is not Skyrim. And even if they are, they may not need to be high-detailed in every corner.

 

Regarding the look vs. performance argument: You got a point there.

 

In the end it comes down to the modellers to create all those models before it will make sense for the mappers to actually use them. If there is only a LOD entity for half a dozen different models out of 100+ that may be used in a map, using them won't give a tremendous performance boost IMHO.

FM's: Builder Roads, Old Habits, Old Habits Rebuild

WIP's: Several. Although after playing Thief 4 I really wanna make a city mission.

Mapping and Scripting: Apples and Peaches

Sculptris Models and Tutorials: Obsttortes Models

My wiki articles: Obstipedia

Texture Blending in DR: DR ASE Blend Exporter

Link to post
Share on other sites

I cant see why anyone would argue against LOD. All modern engines support it. It was even added to Id Tech 4 commercially in ETQW.

 

It's a design concept rooted in future proofing the mod.

 

If other engine performance issues are ever surmounted, wasting performance on full detail geometry at vast distances will be a problem

that folks would want to tackle anyway.

 

The engine is heavily fillrate limited so every triangle you draw is also costly from that pov. Every light overlap multiplies those triangles.

If you start with a lower triangle count to begin with, that exponential curve is starting at a lower point.

 

I think it's a worthy effort not only for the above reasons but it also singles out The Dark Mod as a unique technical achievement compared to

other Doom 3 mods. Having that as part of the pedigree means that possible new coder contributors know that this mod

sets the bar higher. I don't know if coders feel any general attraction to more technically impressive projects but it seems logical to me...

Please visit TDM's IndieDB site and help promote the mod:

 

http://www.indiedb.com/mods/the-dark-mod

 

(Yeah, shameless promotion... but traffic is traffic folks...)

Link to post
Share on other sites

(i was writing this at the same time nbohr1more was writing his post so...)

 

true the highest hit to performance is shadowcasting tris/multiple lights per tri.

 

but again, the more detail you put in the more impact it has. So where ever we can save is good.

 

This isn't only about Skyrim vistas though. it's about performance scaling for all users to give them the best visuals on their machine. it really is more about the system detail settings than distance imo. Of course they are tied into the same system.

-------------

 

Not really scientific but I did a test here...

 

I made a map with 1 very large room, approx 4,000 units by 6,000 units. (this is a VERY large open area). It was just a cube so I sliced the brushes up into a bunch of smaller brushes and put alternating textures so they won't optimize back down into 6 giant brushes (keep terrain light count out of the study).

 

I randomly placed a bunch of lights and a world ambient just for 'random light count'.

 

i cloned this room for a total of 3.

 

One i left empty. one i placed 96 of my 'low poly' pillars. and the third had 96 high poly pillars (same exact placement as #2).

 

On my laptop (ati 420 mobility card (pretty weak) and win 7.

 

Room 1 i was getting 60+ fps. Size of area doesn't seem to really have any effect.

 

Room 2 i was getting approx 45 fps. (96 lowpoly pillars with low poly shadow mesh). Pillars have a total (with shadow) of 568 tris each.

 

Room 3 I was getting 15 FPS. ( 96 hi poly pillars, 2040 tris each including the low poly shadow mesh).

--------

 

So there's definitely an impact. in this case it could be more about the high amount of tris receiving shadows, not about the low amount of shadows being cast.

 

So it does show that no matter what, tris casting OR receiving lights have an impact.

 

I'm pretty sure on my desktop room #2 would still be at 60 and the 3rd would be around 30-40.

========

 

Further testing:

 

I replaced the shadow mesh with caulk on the pillars in room #3. So NO shadow casting, ONLY light receiving. An no more than 3 lights on any tris.

 

Still sitting at 15 FPS. IF I look away so only half the pillars are in view i get 25 FPS, only half again an I get 45 FPS.

 

So yes, just lighting a bunch of tris does have a pretty good impact on a lower end system.

 

=======

But we are talking about a LOT of tris. 96x 2040 = 196,000. Definitely more in view than T2's limit of 1024.

 

=======

moar test:

 

On the other end of the spectrum. i made two new rooms.

 

1: 3 low poly pillars

2: 3 high poly pillars (same shadow mesh)

 

3 lights in each.

 

in this case my FPS is sitting at 60 in each room. So there is definitely a cutoff somewhere where performance starts to 'tank'.

-------

 

So...

 

The 2000 - 6000 tri range with reasonable lighting is basically no performance hit.

 

once I hit 50,000 tris (casting/receiving shadows - reasonable lighting - around 3 lights per tri max) I started taking a pretty decent hit of 15 FPS. Still very playable at 45 IMO.

 

100,000 tris (casting/receiving shadows - reasonable lighting - around 3 lights per tri max) I was hitting what I think is about as low as you want to go, 30 FPS. i think at 30 FPS you can still have reasonably good gameplay. but this is ONLY rendering. No ai thinking involved, so 100,000 tris per scene might be pushing it for good performance/gameplay.

 

At 200,000 tris (casting/receiving shadows - reasonable lighting - around 3 lights per tri max) my laptop is pretty much tanking at 15 FPS.

------------

 

What does it all mean regarding LOD models?

 

Well, that depends on several factors, but we new that.

 

View distance, map detail, visportalling, system map is run on, if the author uses reasonable lighting or not...

 

obviously we can push pretty good detail, but it does scale to the system. LOD is a way to push as much as possible on the users system.

 

(i didn't test THIS map on both of my systems, but i get 25 FPS on this system in my city map in spots i get 60 FPS on my PC, so using that as a general guide to what i would expect).

 

------

But keep in mind, I am not pushing to start making LOD of every model in game. obviously to have an lod of a goblet that is less than 150 tris isn't going to do much for us at all. you'd have to have 20,000 of them in a room to make it worth while.

 

Mostly i'm concerned about future scalability/ease of use/folder structure. The 'popping' is more of a setting value. We can set the distances pretty high to minimize the noticablilty, but that still leaves LOD as a system setting for low enders.

 

Most of the newer models I have made I made higher versions of because of the LOD system. For the most part if you are baking a normal you are already well on your way to easily making a few detail meshes.

 

Either way:

IF we just put an LOD folder in the models forlders so they can be seen it really doesn't clutter the editor, but it helps people find/be aware of them. (or just decide if they want more or less detail default in their map) (ie: i want big pillars but this scene is already so complex, i'll go for low poly version- or i really want this place to look good, I go for high)

 

So I guess the best thing we can do IMO is to set up a folder structure like this and let mappers decide. then the player also has detail settings to choose IF authors wish to use lod entities.

 

Whether or not new models are made with lod, or old ones are retrofitted will really be up to each modeler to decide for themselves.

  • Like 1

Dark is the sway that mows like a harvest

Link to post
Share on other sites
But we are talking about a LOT of tris. 96x 2040 = 196,000.

 

Right. So you'd have to be using a LOT of LOD models to have any real impact...at least 30-40 in a single view, from my reading of the data. I can see it being useful for foliage, or large architectural facades or something, but not much else. That said, as long as the folders don't get in the way of people who aren't looking for LOD models, I don't have strong feelings about where they go.

Link to post
Share on other sites

Well, the impact comes into play more with the menu settings imo.

 

the model has to switch at a fairly good distance to not notice the 'pop' anyway. So most likely the LOD useage is more because someone has a slow system and chooses 'low' in the menu, or they have a really good system and they choose 'highest' which would never switch to a lower model.

----

 

But that said, in my test between my machines lod would be thus:

 

High End PC : i would always be 'in room #3' and still be getting 45-60 FPS and best details.

On my laptop i would be in room #2 and be seeing the medium detail models and getting 30-45 fps.

 

not based on distance, just on user settings.

 

-----

Of course it's an extreme example, especially with the low amount of lod models we have. And considering much of the detail will be terrain brushes and func_statics that probably won't be LOD anyway.

 

i'm not expecting that we'll see any huge gains in performance or anything. just trying to clean up/future proof the system. Which right now is more about establishing a system to use in the future.

======

I guess tels isn't dropping by much more anymore these days but from looking into this my opinion is swaying more towards using really high distances for the LOD swap than we previously looked at.

 

Which would basically just force the LOD to almost never change models, so the only model changes would be menu settings. Authors can always quickly over rie on an entity basis if they need to get more perf.

 

this would also fix the hassle i am dealing with now, which is getting the low poly/normals map model to look very similar to the highest poly model. They just don't render close enough to make the swap easy and clean.

Dark is the sway that mows like a harvest

Link to post
Share on other sites

At a guess, IF huge distances ever become possible due to renderer changes (as in, don't render past some distance, fog, and backface elimination), the most common meshes would be critical to have LOD (like grass, trees, AI, intricate buildings, rocks etc). But the real cpu waste is the AI. Gothic 2 has a trick, it uses with the AI. When they are on a waypoint route, pathfinding towards a destination, and the pc can't see them, because they're slightly outside the draw distance... they just teleport. This has side effects sometimes due to buggy 'escort quest' waypoints that don't disable this (they can teleport to the next waypoint and leave enemies alive behind them), but it seems a good timesaving trick for pathfinding (and i only noticed it very much later). Gothic is very good in not having zone loading screens (there are one or two for major expansions or the ending - for obvious reasons).

Edited by i30817
Link to post
Share on other sites
the most common meshes would be critical to have LOD...AI

 

You'd have to have thirty or forty AI in one place to have any potential benefit from LOD AI.

 

and the pc can't see them, because they're slightly outside the draw distance... they just teleport.

 

That would mean AI would not have a chance to discover bodies, stolen objects, open doors, or other suspicious stimuli unless they were in the player's POV.

Link to post
Share on other sites

That's true; it's the kind of optimization that is linked to certain game mechanics (non-persistent bodies, unsophisticated ownership system - gothic had one thou, but only if the player did it on while they're in the room).

 

There are two options i can see for this kind of trick;

do away with something particularly costly; pathfinding, AI event-notifications, AI checking out for the player (which they shouldn't do anyway), AI inspecting the environment.

 

Or more radical, just pause them. Gothic might have made both actually. Teleport AI transit to a destination and pause them once there.

 

If you want to do something crazy, like say, a 'city' with 60 npcs and 40 houses, like gothic 2 city, this seems like something that has to be compromised at least partially. AI does this game mechanic stuff, EXCEPT, in areas marked 'no-thinking' or something, that the designer tried to ensure are not critical for persistence. Or AI does one more complete patrol route after the player goes away and then pauses (after all, if they didn't find anything on that 'last' one...)

Edited by i30817
Link to post
Share on other sites

they just teleport

 

This only reduces cpu use if they then stand where they were teleported, doing nothing, for the amount of time it would have taken for them to walk there. If they arrive at the teleport spot and continue patrolling, nothing is saved.

Link to post
Share on other sites

That would mean AI would not have a chance to discover bodies, stolen objects, open doors, or other suspicious stimuli unless they were in the player's POV.

 

Several good reasons why not to do it.

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...