Quick map request...

[fromhell]

Can anyone make a simple box room with rows of ammo items, with 10 per weapon (and not the weapons themselves, just the ammo)? Non-TA weapons only.

Why, you ask? Performance testing, as well as comparing with Q3A's performance in the same room.


I'd do it myself... but Quark on Windows 7 is extremely unstable for me


Basically the ammo boxes are a large source of texture switching, and I'd want to reduce that as much as I can.  My idea is making the ammo prisms very simple with all the colors and icons all on one texture page in which the only difference between the ammo boxes will be their UV maps.

[Akom74]

I'm not sure i've understand well, but here is my fast job 

In the attachment you will find the .map and the .bsp 

[fromhell]

Thanks

This is roughly the idea I had for the ammo model. It's the same design, just heavily simplified and with a rough texture.


The new model is 33 triangles, the old was 62 triangles and used 3 textures (1 per ammo type)


For reference, Q3A's ammo boxes were 80 and 56 triangles.

[Suicizer]

That yellow triangle looks pretty odd compared to the rest of the model.
I like the old one way more. The side trimming looks cool though.

[Neon_Knight]

I guess that's fixable.
Low-poly is always welcome when it comes to performance, one of the most criticized things about OA.

[fromhell]

fine, i'll improve the skin

[Neon_Knight]

Hmmmm... is there anything with those empty triangles at the bottom of the model?

[pelya]

Well, I've picked OA as my Android 3-D shooter of choice, because it's low-poly, so it has acceptable FPS even on low-end devices.
On the other hand, I don't think cutting polygons for ammo boxes will improve performance noticeably, the player models and the level itself is usually lot more polys than ammo-boxes.

[Gig]

Fromhell, could you please do some "timedemo" benchmarking with both the standard and the "low poly" ammo boxes, and compare results?

PS: is it possible to have a screenshot of the editor with the two kinds of boxes one next to the another, from the same view?

[Gig]

The low poly version may be used for LOD maybe?

[fromhell]

Initial benchmark with r_primitives 1 (it's slow) with a flyaround spectator demo:

OA0 Ammo Prism: 57fps
OA3 Ammo Prism: 202fps


Have to do further testing on older hardware (PCX2, Rage, etc).  So far i'm liking the results

The main reason is to have less texture switching. Just making this new model the LOD for the existing one won't bring much of a speed advantage and would eat more VRAM.

[Gig]

Is it possible to apply che colored "border" also to the new version? (Check the difference between the new and the old version... the old one is more colorful in the dark!)

[fromhell]

Intel Pentium II Deschutes 300MHz  + Matrox G100A results, 640x480x16, lightmaps, r_picmip 0

OA 0.x Ammo Prism

528 frames 304.2 seconds 1.7 fps 261.0/576.2/907.0/0.0 ms

OA3 Ammo Prism

528 frames 46.1 seconds 11.5 fps 62.0/87.3/169.0/6.7 ms

^ + r_picmip 3

528 frames 25.0 seconds 21.1 fps 28.0/47.4/246.0/10.4 ms

^ Simple Items

528 frames 14.4 seconds 36.7 fps 24.0/27.2/38.0/2.4 ms

^ Simple Items r_picmip 0

528 frames 15.0 seconds 35.3 fps 25.0/28.4/63.0/1.9 ms


The performance gain's not so bad on something that doesn't support OpenGL natively at all!

[Suicizer]

Intel Pentium II Deschutes 300MHz  + Matrox G100A results, 640x480x16, lightmaps, r_picmip 0

OA 0.x Ammo Prism

528 frames 304.2 seconds 1.7 fps 261.0/576.2/907.0/0.0 ms

OA3 Ammo Prism

528 frames 46.1 seconds 11.5 fps 62.0/87.3/169.0/6.7 ms

^ + r_picmip 3

528 frames 25.0 seconds 21.1 fps 28.0/47.4/246.0/10.4 ms

^ Simple Items

528 frames 14.4 seconds 36.7 fps 24.0/27.2/38.0/2.4 ms

^ Simple Items r_picmip 0

528 frames 15.0 seconds 35.3 fps 25.0/28.4/63.0/1.9 ms


The performance gain's not so bad on something that doesn't support OpenGL natively at all!

I still wouldn't call it playable as most players get 200fps.

[pelya]

35 FPS is very smooth for Android 

[fromhell]

I still wouldn't call it playable as most players get 200fps.

Little things like these will help most players achieve closer to 500fps. Testing new stuff on a Pentium II system is a big deal.


Furthermore, it's faster than Q3A at the ammobox performance contest, which has me optimistic for driving OA3 further

[Gig]

I'm quite curious to know the results of the same benchmark demo you mentioned above, played on better hardware...

[son_goku2]

I guess that's fixable.
Low-poly is always welcome when it comes to performance, one of the most criticized things about OA.

For map's performance, hint brushes. And another thing is texture atlas. Either change render to merge lightmaps in larger texture pages or change engine to support high res lightmaps natively.

For really low poly, go for players rendered as simple boxes looollll

[Suicizer]

I guess that's fixable.
Low-poly is always welcome when it comes to performance, one of the most criticized things about OA.

For map's performance, hint brushes. And another thing is texture atlas. Either change render to merge lightmaps in larger texture pages or change engine to support high res lightmaps natively.

For really low poly, go for players rendered as simple boxes looollll

Well, is occlusion culling actually a feature in Open Arena? It greatly increases detail over something with the same performance.