Performance drops in cities

[PopeRigby]

I'm getting pretty bad performance in some parts of cities. Testing in Balmora, some places drop into the 40s, while other places hover around 60. Same in Sadrith Mora.

OS: Arch Linux
GPU: RX 5700 XT
CPU: Ryzen 7 3700X
RAM: 16GB

Settings:
https://cloud.haddock.cc/s/TdYntPe6yarpPMs

Screenshot:
https://pixelfed.social/p/PopeRigby/266738511711965184

[LoneWolf]

You have enabled lots of things that increase the load on the system :

distant terrain , object paging , anisotropy=8 , antialiasing = 8 , water shader, shadows .
Also async num threads is set rather high . On my 12 core / 24 threads threadripper I don't get benefits with anything above 4.

Are you using https://aur.archlinux.org/packages/openmw-git with my bullet-multithreaded ?

[nowheredan27]

I'm sorry for bumping the thread without an answer for it, but is there a guide somewhere with the recommended changes to settings.cfg in order to boost stability and FPS? I'm using OpenMW pretty much as is, and I'd like to know more about these tweaks.
Didn't think this question was worth opening a new thread

[PopeRigby]

You have enabled lots of things that increase the load on the system :

distant terrain , object paging , anisotropy=8 , antialiasing = 8 , water shader, shadows .

Shouldn't my computer be able to handle all of that?

Also async num threads is set rather high. On my 12 core / 24 threads threadripper I don't get benefits with anything above 4.

Oh, well I have 8 cores so I just put that in. Is it a bad thing to go above 4?

Are you using https://aur.archlinux.org/packages/openmw-git with my bullet-multithreaded ?

Yep.

[psi29a]

This site https://modding-openmw.com/tips/performance/ might help you both.

Yes, there is a diminishing return when you keep throwing threads at a problem because the cost of overhead/administration starts to outweigh the benefits.

I would recommend the NumberOfPhysicalCores/2 approach for two reasons, what really counts are physical cores, not logical ones. So your 12 core threadripper would probably be most performant with async physics using 6 threads. However, the most 'gains' to be made would be 1 (just having physics in apart thread is a huge win), having 2 is nice but you start to notice a trend of diminishing returns.

Anyway, this is the general observation.

The real problem is that the original assets in the game are really horrible to even modern GPUs. If anyone ported as-is Morrowind assets to Skyrim, Skyrim would choke. Then put a few decades of mods on top where performance was never a priority and suddenly people start crying because they feel entitled to 60+ fps @ 1080P (or higher) with setting cranked to high, recast/navmesh AI, shadows, reflections, distant land and statics with 4K textures on their insanely priced GPUs.

Morrowind was made in a time where the OLD way of doing graphics programming was the only way and 15fps was fully playable, so it didn't matter if an Urn took up half of a scene's triangle/draw budget (for example, that "simple" static Urn had more triangles/vertices than most content in Skyrim). The more complicated (unoptimised) objects are, the more time is spent on the CPU telling the GPU what to do which hinders FPS.

OpenMW tries to help by merging distant objects together into larger objects so that there less CPU <-> GPU communication, thereby increasing performance. That is how we are able to implement distant statics. The same trick is done with Active Paging, meaning doing the same thing but in the active cells you're in which also gives a nice FPS boost in most situations like TR's Old Ebonheart. Programmatically simplifying these objects would increase FPS even more but that is hard to do right because most of the time the results will look like shit.

This is where a new crop of mods come along like Project Atlas https://www.nexusmods.com/morrowind/mods/45399

The goal of Project Atlas is to identify the most performance heavy areas of vanilla Morrowind and some popular mods and target high usage/strain meshes in those areas for atlasing. This effort involves reworking the UVs for those meshes and creating texture atlases to cover various sets.

This helps quite a bit by make Morrowind assets suck less.

Anyway... OpenMW continues forward one step at a time. All things get better with maturity.

[PopeRigby]

So the assets are shit and I should just deal with it for now? Fair enough. I already have Project Atlas, and I guess the cities aren't that bad.

[Mantar]

One thing that helped me, especially in Tamriel Rebuilt's Old Ebonheart, was

Code: Select all

object paging active grid = true

as that does paging of objects inside the local cell instead of just distant terrain, and there's a crazy huge number of objects in Old Ebonheart so bundling them together into fewer draw calls helped bump my minimum FPS there up from ~11-12 to ~15-16, which is a huge jump in playability.
Project Atlas and the Morrowind Optimization Patch are obligatory IMO.

[nowheredan27]

So I've installed Project Atlas manually and now all windows in Balmora at night look like this lol.

EDIT - OK so basically I copied some stuff from this Glow in the Dahrk mod that I don't use and that's what caused the textures to go missing.
Is there a way for me to download the vanilla version of just these textures again? I really don't wanna have to reinstall the game just for this

[PopeRigby]

One thing that helped me, especially in Tamriel Rebuilt's Old Ebonheart, was

Code: Select all

object paging active grid = true

as that does paging of objects inside the local cell instead of just distant terrain, and there's a crazy huge number of objects in Old Ebonheart so bundling them together into fewer draw calls helped bump my minimum FPS there up from ~11-12 to ~15-16, which is a huge jump in playability.
Project Atlas and the Morrowind Optimization Patch are obligatory IMO.

Looks like I already have that on.

[psi29a]

Might be worth mentioning... these are developer builds, not release builds. Meaning we've packed them full of debug information which might have an impact on performance.