Model issues

[PinkDot]

statue-hammer medium 2 has shadow, could probably be used in all the other hammer statues. they are 1,000 polys

They are not 1000 (and not polygons but tris - this also makes a diffrence). Big ones are +/- 500 tris. Smaller ones are less than 300. See Model Gallery for exact numbers. Only one hammer statue has more than 700 tris (because of rounded edges) - that's why it has shadow mesh. (BTW: I wasn't aware at that time that shadow mesh doesn't replace model's shadow but adds to it, I have to unsure that it doesn't cast double shadows atm).

 

Baddcoq - did you get those polycounts from DR? It gives you total number of polygons per model, including visible, collision and shadow mesh. Not really very useful, but I don't know if it could be fixed to ignore collision and shadow mesh.

 

While I agree that we need to optimize models, I would be careful with doing it too much. Going down from 2500 to 200 might result in shitty shadows really. I'd prefer to turn shadows off for some less significant models rather than having boxy shadows for all of them.

[Springheel]

demon head door knocker only casts shadows for horns and ring. Head missing shadow.

 

Are you sure? It does on my end.

[Baddcog]

@Pinkdot, you're right, I must've been looking at the higherpoly one that has a shadow when I stated that, it's around 900 altogether, I wasn't putting exact tris, but yeah most of those are pretty low.

 

Most of those statues were pulled from Ai mesh so they are very round, 200 might nbe low but it was kindof to make a point, we could still have good shadows with a huge difference in shadow polys.

 

I'd rather see somewhat blocky shadows than none personally. I think it's best to try and get shadows that give good performance, players on low end systems could run no shadows. But if we set stuff to no shadows I think it will start to look odd. Better blocky shadow than high poly objects not having shadows but low poly stuff having shadows, that would look very odd. worse than blocky shadows IMO.

 

@Spring, yeah, I had one and could only see horns and ring, a big gap of no shadow in the middle. I don't see why it would be different for both of us, I just saw it in game an hour ago. Maybe someone changed shader since you last looked? My SVN is current.

[Springheel]

Not sure. I checked the material and it is set for full shadows. It is acting as expected here.

[Springheel]

Going down from 2500 to 200 might result in shitty shadows really.

 

I've run a few tests and I can actually decrease the polycount for the shadowmesh to 25% of the original without losing any real noticeable quality from the shadow. The globe now has a shadowmesh of 600 polys rather than 2500.

[Baddcog]

The globe now has a shadowmesh of 600 polys rather than 2500.

Yeah, that's what I'm talking about!

 

I don't mind making some shadow meshes, I'm not trying to get anyone to drop what they're doing btw, just pointing out stuff that could be fixed.

If I do meshes for the lwo's though I'd just need to send someone a 3ds file and they could combine it into the lwo's. I don't want to start changing lwo's to ase's, that would be a mess.

 

tdm_knocker_demonhead
{
metal

diffusemap textures/darkmod/metal/flat/brush_steel_scratched01_dark
bumpmap		 models/darkmod/props/textures/knocker_demon_local
diffusemap	  textures/darkmod/metal/flat/rough_iron01_tiling_1d

{
	if ( parm11 > 0 )
	blend	   gl_dst_color, gl_one
	map		 _white.tga
	rgb		 0.40 * parm11
}
{
	if ( parm11 > 0 )
	blend	   add
	map		 textures/darkmod/metal/flat/rough_iron01_tiling_1d
	rgb		 0.15 * parm11
}
}

 

I think the spec map is the problem, you have 2 diffuse maps listed.

[Baddcog]

OK, all of the ase files inside vases now have shadow and collision meshes. In most cases the collision meshes are arounf 50 polys (compared to 500ish). exceptions are the stone bowls/cups. They should probably get simple collision meshes at some point so they can be moveable, but they are only 100 polys so they can wait for shadow meshes

 

Shadow meshes vary. I tried to used shadow instead of shadow 2 so they'll cast shadows on other objects and themselves. This required making some meshes higher than I wanted but still they range from 100-250 on vases that are around 500 polys. So they are still at least 1/2ed.

Some of the vases just have too much shape and are too think to hide a lower poly mesh inside.

 

vase2_small and mini have a minor shadow mesh issue, the mesh sticks out in 2 spots. Not too noticable (not noticable at all in max ), I'll get to em later, but they'll work for now.

 

-----

@Spring, I took care of the door knockers

[Springheel]

I think the spec map is the problem, you have 2 diffuse maps listed.

 

Yep, that certainly does look like the problem. Weird that it cast a shadow for me though, at least in D3ed (didn't test it in game). I'll have to doublecheck that I have the same material.

[Baddcog]

Well, if you SVn after last night you will have the same material. I mean you won't, cause I fixed it, another knocker was like that too.

[Springheel]

Ah, I know one of the knockers was affected by a texture being renamed a while back; this might have cropped up during it being fixed.

[Tels]

I've run a few tests and I can actually decrease the polycount for the shadowmesh to 25% of the original without losing any real noticeable quality from the shadow. The globe now has a shadowmesh of 600 polys rather than 2500.

 

Very good! I can confirm that shadow-casting polygons are *the* performance killer in D3.

 

Cutting down a scene from 140000 to 120000 tris (but removing 10000 of 40000 shadow tris) can boost your performance much more than you would expect from the tris alone.

 

And I think having slightly simpler shadows but better performance is better than having no shadows, or lagging FPS.

[Springheel]

Another side benefit of this is that many objects that are currently set to noselfshadow will be able to receive shadows again. For example, most vases are set to noselfshadow, so one vase will not cast a shadow on another vase, which can look out of place. By using a shadowmesh, the vase surface can be set to noshadows instead of noselfshadows, meaning both vases will cast a shadow on each other, as you would expect (if you can follow that).

[Baddcog]

did a little more work.

The rippled pillar was about 950 tris. shadow is now 54 and collision is 38

 

The Stoup (birdbath) was 1148, the shadow is now 486 and the coll is 30. Checked the shadows in game too, still looks pretty good, I don't think it's noticeable really, I kept all the fancy shape at the bottom, that took quite a bit of work.

[Springheel]

This is the perfect, fairly mindless, task for me at the moment. I was feeling a bit burned out.

[PinkDot]

The globe now has a shadowmesh of 600 polys rather than 2500.

600 and 200 makes a significant diffrence and that's more the way I'd see it. It's a lot of shadow polygons saved but still can give some good looking shadows.

[Baddcog]

I'm feeling the same way

 

Just did the statue_folded_hands med and large. mesh about 1300, shadow and collision now 130

 

@Spring, The decor hammers could be cut alot too. You can open/save ase correct? The hammers crash Max when I try to import so I'll be skipping them. I'm done for awhile.

 

@PinkDot. Yeah, that's fine by me too, I wasn't trying to establish a certian number by any means. That really depends on the obj. I do think we should keep the shadows as a good representation of the model, not just a blob.

For example, that statue I just did, the shadow actually looks like it comes from that model (to a large degree, if someone really studied it they probably say, "wait a minute..." and I got that WAY down.

But that birdbath, I might've been able to cut out a few more polys but didn't want to spend hours on it, I still halved it and the shadow looks like it comes from that object still. Noway I could've got it down to 200.

 

Also, on some of the pots, the very large ones used in the bonehoard have shadow mesh for the handles and it looks good. But the small and mini handles were so small it was easier to seperate them from the main mesh. So the main part of the pot casts no shadows but has a shadow mesh. The handles use another material that does cast shadows. They are a little high poly but the shadow looks good and I still cut at least 100 polys out of shadows. I guess some things will just require a bit of creativity.

[Baddcog]

Another side benefit of this is that many objects that are currently set to noselfshadow will be able to receive shadows again. For example, most vases are set to noselfshadow, so one vase will not cast a shadow on another vase, which can look out of place. By using a shadowmesh, the vase surface can be set to noshadows instead of noselfshadows, meaning both vases will cast a shadow on each other, as you would expect (if you can follow that).

yeah, they actually cast a shadow on themselves too if you use shadow material. which is good because they now have a shadow inside.

 

an exception I made if my pots with the metal rings. I didn't make a shadow for the rings, so the ring won't cast a shadow on the side of pot. It might not look quite as good but in that case I didn't feel it was necessary.

 

On my facny metal chest I did include shadows for the rings, mainly because it's a fancier object and looks good that way, also less chance of it being used in alot of places like common decor pieces.

[Springheel]

You can open/save ase correct?

 

Nope, just .lwo.

[Springheel]

Hmm, many of the statues just reference a generic stone texture. I can't set that to noshadows or it will mess up brushwork using it. And if I make an alternate material then I'd have to fix all the skin entries, which I don't feel like tackling right now. I guess I'll leave the statues alone for the time being.

[Springheel]

Wow. Now *that's* optimization. The rope on the right? 2040 polys. The one on the left? 316.

 

[Springheel]

Hmmm. I ran some tests that are pouring some cold water on this shadowmesh idea.

 

I took the rope above, 2040 polys, and made a 500 poly shadowmesh. Then I made a map with 2 lights and stuck about 40 ropes in it. When the ropes weren't in my FOV, I got 30 fps. Here are the fps when all 40 ropes were in my fov:

 

highpoly model set to noshadows: 30 fps (no change at all)

highpoly model with full shadows: 20 fps

highpoly model set to noselfshadow: 20 fps

highpoly model with 500 poly shadowmesh: 20 fps

 

The good news? Having lots of high poly objects in a map doesn't seem to have any significant impact if they don't cast shadows.

 

The bad news? There was absolutely no difference between 40 x 2040 shadowcasting polys and 40 x 500 shadowcasting polys.

 

I don't fully understand how there could be no difference. Could someone else try a similar test and see what kind of results they get? If this holds up, then there is no point in making shadowmeshes for static objects.

[Baddcog]

Did you rememder to set the material to no shadows for the rope when you tried the shadow mesh?

 

I know, kindof a silly Q but I did check a few shadows in game and had forgot to fix the material. I was like 'Wow, I did awesome work on that shadow, Doh! forgot to fix material..." They still looked good but I could tell after I fixed. However I've been looking at em windowed and my framerate is always crap windowed.

 

I'll run some tests myself.

 

The gargoyle (2500) now has a shadow mesh around 500 and a collision around 50. I decided to 'cheat', didn't want to mess with the shadow inside 5 fingers so I used shadow2 on a seperate hand mesh. It's sloppy and sticks out of hands, but doesn't cast shadows on itself.

works good since it the hands will most likely only cast shadow on a roof, which will most likely be func_static or terrain, it still casts shadows on them.

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

OK, I just did a test with the above gargoyle. So that's where the numbers come from.

Simple room, four walls, ceiling,floor. 5 lights, one per corner, one in middle.

saved under 2 names so the only thing that changes at all is the models.

standing in corner looking at wall = 60 FPS. (full screen)

 

created 40 gargoyles in one map. original gargoyles with No shadow/collision meshes. 2500 polys each.

Draws = 315

Tris = 365,000 (hard to tell exactly they jump around too fast but I did a fair estimate of both rooms)

Shadows Tris = 139,500

FPS = 19-24 (standing in corner looking at room, moving view around)

img = 1.6 MB

 

next map

created 40 gargoyles that I added the shadow and collision mesh to.

Draws = 400

Tris = 435,000

Shadows Tris = 92,000

FPS = 26-30

img = 1.1 MB

 

vbo = 0 on both, I have no idea what that means.

 

What that tells me is there are a substanstially larger amount of tris with that shadow mesh. not exactly small at 500 but it gives good details.

Like the globe, some models just can't get too low.

But that amount of polys doesn't matter. It's about 70,000 polys more than no shadow/collision meshes.

 

But the shadow tris are cut by about 50,000 polys. that gave me 6 to 7 FPS.

also note that it cut the img MB useage by 1/3. I'm not sure exactly how that plays into it, probably the MB's used to render?

 

This is with 2 interent windows open (one web radio), Dark Radiant open, a couple tex pad files and a few folders.

 

---------

Yeah, that rope is probably a picture perfect example of what a true nex gen low poly object is. (I say nex gen in comparision to T2)

[Springheel]

Did you rememder to set the material to no shadows for the rope when you tried the shadow mesh?

 

Yep, I did. You had five lights and I had two, but otherwise our test was similar. I can try adding more lights to mine and see if that changes anything. I wonder also if it matters whether the shadows fall on other models? In my test, the rope shadows just fell on the ground, but I bet the gargoyles cast shadows on each other.

[Baddcog]

Yeah, the gargoyles were in bunches so they deffinately cast shadows on each other. And I didn't do any test with Ai/collision models, I suppose it helps a bit though going from 2500 to 50 there too.

 

--------

hmm. waiting for dindin to cook so I figured I'd try the rope test and see why we were getting such different results.

fewer lights probably helped but...

 

I duplicated your test Springheel.

one room, 2 lights, 40 of each rope type and I got quite different results each time.

 

40 lowpoly ropes with shadow mesh = 60 FPS whether or not I looked at wall

 

40 high poly ropes (with untouched tex-has noselfshadow dec) = 25 FPS

40 high poly ropes (edited mtr, NoShadows, NoSelfShadow) = 45 FPS

 

and oddly enough:

 

40 high poly ropes (edited mtr, commented out //noselfshadow ) = 30 FPS (so I got better performance with full shadow cast than without noselfshadow

 

Still, I got much better performance on the lowpoly ropes with a shadow mesh.

 

I still think the gargoyles was a better test as it was the exact same mesh, the difference only bing self shadow or shadowmesh

[SneaksieDave]

Just to chime in from the low-end-PC side of the pool. While unlit/unshadowed polys may not seem to have much effect, they definitely do on the low end. My 200k polys of 40 AI (frozen, not animating) gives me a significant dip in framerate. Not nearly as much as lit or animated, but it's still there.