Submesh test on performance (original test)

[Springheel]

For those that don't necessarily read the bug report forum:

 

 

Ish: Create two sets of AI, one with a uniform material over them and one split into 3-4 submeshes. Then we could place tons of these AI in a room and compare the performance, to see if the performance hit is large enough that we really have to write a surface map system.

 

Orb: You can do this test with static meshes. I don't see any reason why the renderer should treat animated AI and static meshes any differently from the perspective of draw calls and batching, since the animation has already been "done" by the time the renderer gets the data.

 

Just get some moderately complicated meshes, make a multi-material version and a single-material version (but make sure the actual UVs and images used are identical) and place several hundred of them in a room to test performance.

 

Ish: Okay, good point, we should do this then. Do any modelers have time to do this?

[Ishtvan]

Yeah, thanks for reposting that here. Just wanted to add this is rather high priority, since it will effect how we plan to texture any new AI from now on.

[_Atti_]

ok, i am testing this..

[_Atti_]

Well, i made a maps with 81 balls with 2500 poly.

 

With the multimat model i got 43-44 fps when all the balls was in view

after replacing it with the lwo containing the single mat model this fps was 45-46

 

i repeated it with 135 balls and i got 29-30 for single, and 28-29 for multi

 

do you think it would worth redoing with even more objects/less poly? i didnt notice you wrote 'several hundred'

[Ishtvan]

Thanks a lot for doing this test!

 

I dunno if low poly / more objects is needed. What you did sounds like a decent test because the AI are going to be pretty hipoly, and I doubt we'll have more than 135 AI visible at any one time.

 

How many materials per ball are you using for the "multi-material" balls? We were thinking the AI might have something like 4 in the worst case. Also, I don't know anything about modeling, but we should make sure the modeler isn't making any optimizations that would auto-merge them into one material (I doubt that happens, but just in case).

 

Also, I guess for a good test, the materials used should have normal and spec maps too. They don't have to be accurate or anything, just as long as they're there.

[_Atti_]

The materials didnt merge ( arrow reacted properly, i guess its alright)

Spec and bump maps were used.

It was only 2 materials but i will test it with 4

[Ishtvan]

Cool, thanks! Would it be possible to also either put it up on CVS or make it available on the ftp so that people with other systems can test it too?

[OrbWeaver]

Looking good so far, I'm reading that performance impact as "utterly negligible, possibly even statistically insignificant".

[Ishtvan]

Also just a random question, but when you did the test, did you stand somewhere where both of the materials would be visible from the place you're standing? You probably did, but just wanted to make sure.

[_Atti_]

tested it with 4 materials and fps was now decreased to 26-27

 

yes i will put it on cvs soon.. there is only one map you have to rename the model,i couldnt think of better method to place all those balls in the exact same situation.

[_Atti_]

and yes, i could see both(and while latest test, all four) materials:)

[_Atti_]

ugh just noticed..one ball is not 2500 polys but 5000..misread something.

 

Its up on CVS.. mapname is perftestmulti

 

You can find the models in models/darkmod. you have to move them to props/misc (sorry forgot the map uses it from there and i dont know how to delete or move files on CVS..

i also know props is not the best place for it but i didnt know it would eventually go on cvs)

 

perftestmulti0.lwo is the one with single material and perftestmulti.lwo is with four materials..

the map uses the perftestmulti.lwo model. you got to manually switch filenames unless you know a better way of it

[Ishtvan]

Okay, thanks a lot!

 

5000 polies is probably okay since I think that's roughly equivalent to an AI model. One way to switch between would be to create two copies of the map, one that has the names changed to perftestmulti and one with the single model perftest. I can do this if nobody beats me to it.

 

[EDIT: Btw just to clarify, if I'm understanding correctly, you're saying FPS decreased from 29-30 to 26-27? Or is it from ~40 to 29-27?]

[Ishtvan]

Okay, I'm putting perftestsingle.map on CVS, which only uses the single material object. Also, I noticed that the models PERFTESTMULTI were accidentally in /models/darkmod instead of /models/darkmod/props/misc like the maps pointed to, and the maps didn't initally work until I moved them there, so I'm moving them there on CVS.

 

The textures on the map also seem messed up, but I didn't know how to fix that and not sure we need to for this purpose.

 

I tested it out from the same spot (backed all the way against the law, looking at all the spheres but not the sky, because something seems weird with the sky). FPS varies a bit depending on how you look at it, so make sure you test from the same spot same view,s ame lightgem settings, etc, both times. (I guess we could also move the player spawn back so it's in that spot anyway and just record FPS after spawning without changing the view.)

 

Single material (perftestsingle.map) : FPS = 29

Multi materials (perftestmulti.map) : FPS = 25

[_Atti_]

"Also, I noticed that the models PERFTESTMULTI were accidentally in /models/darkmod instead of /models/darkmod/props/misc like the maps pointed to"

 

yea, mentioned it in my prev post.

 

"Btw just to clarify, if I'm understanding correctly, you're saying FPS decreased from 29-30 to 26-27? Or is it from ~40 to 29-27?"

 

it was 29-30 with single

28-29 with 2 materials

and 26-27 with 4 materials

[Ishtvan]

Okay cool, hopefully more people will go and test it out.

 

Again to test it out, go to perftestsingle.map, pick a spot to stand in and look at the spheres, note your FPS.

Go to perftestmulti.map, stand in the same spot and look at the spheres in the same way, and note your FPS.

[sparhawk]

Does this map look normal? I have some strange effect on the textures and I thought it is due to the texturereorg in my map, but now this map has the same issue. The texture looks as if it is located in a single spot, and then smeared over the wall in all other places. I can take a screenshot later.

 

FPS: 9 and 14 when I move to a place that is farthest away so that everything is visible.

 

Since the FPS vary from where you look, it only makes sense to compare it when you know where the player stands. Best way is to put the player in a certain position and don't move him.

[Dram]

Does this map look normal? I have some strange effect on the textures and I thought it is due to the texturereorg in my map, but now this map has the same issue. The texture looks as if it is located in a single spot, and then smeared over the wall in all other places. I can take a screenshot later.

 

Yeah I noticed this with a few maps, namely inventory_test. Very strange indeed.

[Springheel]

The texture looks as if it is located in a single spot, and then smeared over the wall in all other places.

 

Any map that points to a moved texture looks like that.

[sparhawk]

But the perftest maps are new ones, right? And Ishtvan didn't mention it, so I thought that it looks normal to him.

[Springheel]

He did mention it:

 

The textures on the map also seem messed up

 

Probably the textures just point to the wrong folder.

[sparhawk]

Aha! Well, I didn't think it is THIS messed up.

[Ishtvan]

Less talk, more numbers!

[sparhawk]

I fixed my own map now, so it was really only the path. Looked quite strange though, because the texture was there, it was just not correctly displayed. If the path is wrong I rather would have expected to no texture showing up at all. Strange.

[Ishtvan]

So can we draw any conclusions from this test? What does the FPS drop in this stress case with 135 "AI" indicate about what it would be in normal cases? Is the drop tolerable? What happens when we put in a shadow casting light?