LOD, more is better?

Hey, I have a game where I am showing a lot of gems for a player to collect. Previously this has not been an issue, but now a level is taking 15 seconds to load, and I beleive the issue is the gems. There are 760 of them on this one level (its big...).

Now each gem has 3 main components.

* Light
* Reflection Probe
* LOD GROUP

I have been exploring how best to do LOD, and I need some guidance here on those that have beaten this path and know the result. So the question is..

Is it better to have more LOD geo and levels, because it gives you a smoother transition between gradually degrading models. And those low poly LOD models really do give a big perf boost.

Or

Is it better to have at the other extreme, LESS or even NO LOD models, because LOD gives your scene a massive overhead in terms of initialisation and extra objects.

I ask this, as using the profile shows I have 13000 objects in my scene!!! Im not sure exactly what it means by that, Im assuming it means discreet classes and components on GameObjects. But it raises a point. Instantiating each Gem does now seem to take a decent amount of time. I have 4 LOD levels, and that means 4x 760 gem models, plus 760 lights, and 760 reflection probes. thats a lot of object to create on load, and I guess thats where the majority if the 15 seconds load time is now spent on.

BTW, I know 760 may seems excessive, and yes I can pull back from this, and probably get it down to say 300, without the level feeling too sparse. But its more the principle here. If I CAN make my gems more efficient in terms of loading, and rendering, then I can just concentrate on making a good level, on not on keeping under X gem budget.

 

Has anyone noticed that there is VERY little actual help here on the forums. I'll admit, Im not paying for Unity yet, as I am not commercial. But most of the time, when I ask a question. There is no reply, and if it is, its often from another public forum user, with a "I have that problem as well" kind of post.

I looked on this thread, and there was one post with 55 views, and one reply. And that one reply was from the OP clarifying a detail on his post.

I would like to take action on the issue posted about tonight. But Im not sure at this stage if I will get an answer that is of any use. Im beginning to think that maybe I will have to do some kind of R&D tests on this issue.

 

Probably because no one knows the answer to your question. Any question of "Should I use this optimization or not?" is use-case specific. Since we don't have access to your full project, anyone who would say "yes" or "no", would only be guessing. Have you tried using the profiler to see a break-down of what is taking so long?

What you mentioned about R&D tests is typically what you would need to do in order to optimize a scene, but the profiler is meant to help you with that.

 

In most cases, about 4 LOD's, are good for characters in an FPS game.
Eg. Halo etc.

The lowest poly LOD, is normally for distant camera shots, or panoramic views.
The mid to high poly LOD's, are normally for close up camera views.
And objects in cut scenes, have LOD's that have, super high polycounts.

If the in game gems and objects in the scene, are all too high poly, and there
are like a ton of them present in the stage, then that can easily cause, really
bad performance issues.

Also, 760 lights sounds like waaaay to much.

If possible:
Can you post the polycounts of the objects, and enviroment, in your scene?
And can you post a few screenshots, of the enviroment and objects etc,
so we can get a better idea, of what you're doing?

 

I did see your post and typed some stuff out then it hit me - I felt a bit depressed repeating the same advice to you, that I've given so many other times. I thought to myself - perhaps this person will search and find one of those posts. I don't mind - but - unpaid volunteer here

Er, that's why. When anything is perf based it *has* to be R&D because someone's watch game is really not gonna fly on an nvidia RTX FPS you know?
And we don't know your shader, pipeline, anything. All the data is here, and it begins with the profiler and the unity manual which has a ton of advice.

The key is to first identify what is slow. It's usually the shaders followed by draw calls and maybe scripts and post fx last.