Whats up with this? More Bugs?

[TZeer]

Think I found the small boats edm files :) Renamed them and did a small test.

 

Here is before I renamed them. 72 FPS. Not bad, but add a cockpit and few other A-10C, each with 100 object models. And FPS quickly drops.

 

 

Now, here is after I renamed the boat files. 89 FPS

An increase of 17 FPS. Not bad :)

For a few boats you don't even see from this position.

 

[Lange_666]

One of the things that comes to mind why they are rendered even if you don't see them could be to prevent loading stutter?

You can see that happening if you use F11 to jump from airfield to airfield and on every airfield you do a 360° view. At some points while turning the camera around you'll see stutter that isn't there anymore on the next 360° camera sweep of the same area.

Edited February 17, 2013 by Lange_666

[159th_Falcon]

One thing that still bugs me is why the A-10C standalone version had a much higher FPS rate then DCS World even though all the objects mentioned above already existed in the stand alone version.

 

Cause DCS World has received quite some improvements in mainly the graphics department. Explaining a loss in FPS.

Ofourse a lot of code is also changed to make World happen in the first place and this probably is also related to the lower FPS your experiencing.

[locoidal]

Think I found the small boats edm files :) Renamed them and did a small test.

 

Could you post the list of files that you have found? (cranes, boats and so on).. this way we can test other objects to try to find more clues

THX!

Edited February 18, 2013 by locoidal

[TZeer]

Sure :)

 

DCS World/Bazar/Terrain/Structures/High/EDM

 

boat-1.edm

boat-1a.edm

baot-1angle.edm

boat-2.edm

boat-2a.edm

boat-2angle.edm

boat-3.edm

 

I don't know what file is for what boat :p

 

kran_port.edm

kran_stroi.edm

 

 

konteiners_brown.edm

konteiners_red.edm

konteiners_white.edm

[locoidal]

Sure :)

 

DCS World/Bazar/Terrain/Structures/High/EDM

 

boat-1.edm

boat-1a.edm

baot-1angle.edm

boat-2.edm

boat-2a.edm

boat-2angle.edm

boat-3.edm

 

I don't know what file is for what boat :p

 

kran_port.edm

kran_stroi.edm

 

 

konteiners_brown.edm

konteiners_red.edm

konteiners_white.edm

 

Gracias! amigo!

:thumbup:

Later, after work, i'll do some tests with the files... unless dark souls don't let me do them ;)

[locoidal]

Look at these thread .. it seems there is a little tool to open edm files "model viewer"

 

http://forums.eagle.ru/showthread.php?t=86205

http://forums.eagle.ru/showthread.php?t=78704

Edited February 18, 2013 by locoidal

[TZeer]

Small update:

 

Did some more testing today. It's not only the boats and cranes. The amount of containers is also a big contributor.

 

Here is the default setting. 2119 objects.

 

 

Here I have removed the small boats. 1858 objects. Almost 300 Objects.

 

 

Here I have removed the cranes and the boats. 1450 objects. Another 400 objects reducement.

 

 

And the last one, here I removed the containers as well. There are over 180 containers in the harbor!!! 665 Objects. Almost 800 objects less to deal with. So by removing small boats, containers and cranes I went from over 2000 objects to less then 700. And a doubling in framerates.

 

 

I guess the containers have about 4 objects for each stack. And I doubt there is much that can be done with them. Except reducing them, or making larger stacks of containers. Make it like there are more containers. But it's still only one model with 4 objects in the world.

 

Like this:

 

Edited June 2, 2013 by TZeer

[TZeer]

Looks like I found the reason for high object count on the train and it's cargo.

 

First picture here is the average number of objects for the rail cars. This one has 14 objects.

 

 

Now, look at this one. 93 objects!!

 

 

Here is a few other strange ones. This one has 15 objects.

 

 

While this one only has 1.

 

 

Here is one of the cranes. 40 Objects!!

 

 

And a radar. With 1 object.

 

 

 

Among the over 800 EDM files in the folder, it was about 40-50 of them who had a high number of objects compared to the rest. Most of the models where lower then 10 objects. The 40-50 other had an average of 15-16 objects. Where the highest was 93.

Edited February 18, 2013 by TZeer

[TZeer]

I don't know why, but when you put the model into the game, the object count on them doubles.

 

2 examples:

 

M1A1

 

In modelviewer, 3 objects.

 

 

In game, 6 objects. Still not that bad.

 

 

Now, a badly optimized model. The 1c91. 38 objects.

 

 

In game. 76 Objects!!!

 

 

 

That means the single rail car earlier actually gives you 186 objects. While the little crane will give you 80.

[Ranger79]

TZeer, ED should hire you man, would surely help everyone who loves the game!

[SkateZilla]

it's not the number of OBJs that important, it's the Number of TRIANGLES/POLYGONS.

 

 

Number Of Objects is the Number of Individual Parent Mesh Pieces there are in the EDM (or something along those Lines)

(ie, An Aircraft would have separate pieces for all moving control Surfaces etc),

SO A Tank would be a Rotating Turret, Base, Tracks.

Now, You have to add the Shadow, which is the same model just rendered differently. So you double the Poly and Object count.

 

 

Also, This is ONE Model, but separate Pieces: Note it says 14 OBJs.

Edited February 19, 2013 by SkateZilla

[TZeer]

Here you can see the difference between optimized models and models who has not gotten the same treatment.

 

In all pictures I have 100 units.

 

First one is the M1A1. Roughly 140 FPS.

 

 

Second set of pictures is the 1c91. 7980 objects!!! and 33 FPS.

 

 

What was that SkateZilla?

[Hamblue]

Does your last set of objects have moving radar on top? Also if they are sams they probably have some pretty

hefty AI code involved. and each with moving parts. Big difference.

Wonder what it would look like flying over them when they all launch if they are SAMS

[SkateZilla]

Ok, well, if the poly count isnt the problem, then it's prolly shaders applied to the textures.

[SkateZilla]

turn off shadows and see if that changes anything, as teh 1c casts 38 shadows while rendering another 34 onto itself, the M1 casts 3 shadows, while rendering 1 on itself.

 

 

Even the F-100 Module EDM, at 118165 Polygons only casts 6 shadows and renders 4 onto itself

[TZeer]

All tests I have done, shows that when I'm reducing the number of objects in different settings, my performance increase.

 

Does your last set of objects have moving radar on top? Also if they are sams they probably have some pretty

hefty AI code involved. Big difference.

 

There are models with moving parts that consists of 1 objects in the modelviewer. Also, the rail car I showed earlier with 93 (186) objects is a static model in the world. And AI code has nothing to do with the pure visual.

 

Ok, well, if the poly count isnt the problem, then it's prolly shaders applied to the textures.

 

Based on what? All test's I have done points toward that the system is getting clogged by the number of objects. Increase number of objects and GPU usage goes down. Decrease number of objects, and the system get's breathing room to utilize the GPU and CPU properly.

 

Number of triangulations or whatever with the high object model, was just under 400000.

 

With the low object model, it was almost 2 mil. And still pushing 110 FPS more.

Edited February 20, 2013 by TZeer

[TZeer]

Reduced shadows to low. No difference. Still ~30-40 FPS with the high object model.

[speckfire]

Wow what a difference at Novorossisk !!!!! Thanks a lot TZeer, you're on to something biiig

[Quirkitized]

keep up the good work, interesting to see all these results.

[Hamblue]

All tests I have done, shows that when I'm reducing the number of objects in different settings, my performance increase.

 

There are models with moving parts that consists of 1 objects in the modelviewer. Also, the rail car I showed earlier with 93 (186) objects is a static model in the world. And AI code has nothing to do with the pure visual.

 

Your data shows 1 object. have you tried to see if each object has the same effect on the game? For instance it would make sense that a SAM targeting a plane in-game would have a different effect that one sitting idle. If you fly an aircraft towards the same group does it change things?

[Hetzer]

Some more models:

 

platforma_akacya_.edm

platforma_btr_.edm

platforma_crush_.edm

platforma_ikarus_.edm

platforma_msta_.edm

platforma_tank_.edm

platforma_track_.edm

platforma_tunguska_.edm

dok_s_.edm

dok_s_crush_.edm

 

DCS World\Bazar\World\Shapes

zwezdny_.edm

zwezdny_.lods

[Keyser]

It would indeed appear that the bottleneck is an internal engine pipeline that can't process a larger number of "objects" (whatever those actually are - it seems they have nothing to do with polys, animation or AI). That explains why the loss in performance occurs while GPU and CPU usage go down. The engine simply isn't able to output the scene quickly enough - CPU and GPU have nothing to work with.

 

TZeer does of course have a very powerful system, so I would be interested to see some data of performance on weaker systems. The question is, is the bottleneck in the engine an absolute one which only becomes relevant on fast systems, or is it a relative one that acts as a bottleneck even in slow systems.

However that may be, this is still excellent work, and I would vote that at this stage, the findings should be replicated by a Tester, and then put on the fast-track (I read that there is such an option in the bug reporting process) for developer attention. Currently, the seemingly random FPS drops are one of the few still remaining major problems that negatively affect the gameplay experience in DCS.

 

If the objects are indeed the culprit, as now seems to be almost inarguable thanks to TZeers excellent work, I would expect updated models or an equivalent fix to the rendering engine quite soon (1.2.3.1?). That is to say I would be quite disappointed and annoyed if such a fix was not forthcoming, as my game suffers from severe FPS drops which are sometimes testing the limits of my patience.

[TZeer]

Your data shows 1 object. have you tried to see if each object has the same effect on the game? For instance it would make sense that a SAM targeting a plane in-game would have a different effect that one sitting idle. If you fly an aircraft towards the same group does it change things?

 

If you had read the entire thread you would have known that I have tested multiple objects.

 

And for the last part about a SAM targeting a plane. It has nothing to do with this test. And even if it was, the AI would just add on top of the already reduced performance.

 

All my testing as far as its possible I'm doing far far far away from any other objects on the map, in a clean mission editor. This is to prevent other factors to mess up the data.

 

Where that's not possible, like Novorossiysk, I do adjustments, and observe the results I get. And just to use that scenario as an example. By removing the cranes, containers and smallboats I improved my FPS by 100%, while reducing the number of objects from over 2000 to under 700.

 

Does your last set of objects have moving radar on top?

 

Forgot to give a proper answer to this one, was late last night :)

 

No, the model had no animated parts moving while I tested.

 

 

 

TZeer does of course have a very powerful system, so I would be interested to see some data of performance on weaker systems. The question is, is the bottleneck in the engine an absolute one which only becomes relevant on fast systems, or is it a relative one that acts as a bottleneck even in slow systems.

 

This is something I'm wondering about myself. What I suspect, is that weaker systems have the same performance drops, but on a much smaller level. Simply because they have other bottlenecks in their system. Or they are already experiencing other performance issues like the heat blur bug or FPS loss due to internal flood light in cockpit.

 

With the heat blur bug, my FPS would go down to around 50-60 FPS stable. Same with the flood lights in the cockpit. But when I combined them both, I still had between 50-60 FPS??

 

Since I have been able to remove that factor and get a very good performance, my system is very sensitive to changes. 20% change for someone running 30 FPS, is only 6 FPS, If I'm running 140-200 in a test environment that 20% is 28-40 FPS difference, and quite easy to pick up.

6 FPS is very easy to to just brush off as "normal" or probably something like "oh, your looking in that direction"

 

 

Now, it's not much we can do except highlighting the models with abnormal amount of objects in them. It's up to ED to take action on it. Hopefully they will :)

[ZaltysZ]

There are models with moving parts that consists of 1 objects in the modelviewer. Also, the rail car I showed earlier with 93 (186) objects is a static model in the world. And AI code has nothing to do with the pure visual.

 

I wonder what does "object" really mean? Quickly browsing through EDM docs, I see that EDM can contain multiple objects: renderable geometry, bounding box, collision shell and etc. That could explain variation in object count.

 

Can these 1 object models be destroyed in game? Or are they only for rendering?