It depends what kinds of triangles. For example textured triangles alone gives you higher throughput since you can VBO the meshes and use few texture changes. Now if you venture into large scale outdoor terrains using varied and vivid texturing things get more stressful. Much more texture changes, LODding to cut down triangle count and so forth. Although… 3.2M tris @ 60fps on a 8800GT? What did you use? Your test scene duplicated multiple times? Furthermore does Blender use “forward” or “deferred” rendering? Since in the later case you can pump more triangles onto the screen ( I assume it uses deferred since it has this “node render system” which would be in favor of this technique ). I’m talking more about a setup where things are dynamically moving ( lights, objects ), where there are spot and point lights, where there is indirect rendering ( mirrors, portals ) and where there is transparent shadows. What costs speed is the number of render passes required ( transparency requires multiple passes, transparent shadows require multiple passes ) as well as the quality ( shadow map size ). These 3.2M are benchmark values which are not telling a lot except what your hardware could do in the most optimal case ( and most unlikely to happen in any sophisticated game ). We are though interested in how blender performs and this requires pushing it into places where the internal mechanisms can show muscles ( or break down ).
But with shadows and a large scene (that you can see for kilometers at at time) I’d say 800,000k is about what my testing got, with shadows and GLSL active.
What is this scene looking like? How many shadows are there? Spots or points? What shadow resolution? What screen resolution?
cryengine 2 doesnt’ have mirrors you know, and it only hadnles 3 light sources at a time without glitching last i’ve heard. So there isn’t really a single objective standard in which to judge all game engines as you suggest in your post. Even though I do respect your opinion and all that stuff. You do appear to know your stuff. I tip my hat to you!
I talked more about X-Ray right now instead of CryEngine. Handles ~100KTri/s according to specs ( that said specs are relative, only the real world performance really matters ). Doesn’t have mirrors though but does as many lights as you want ( until it gets slow ). On my Radeon 4870 test rig this one can churn out a full landscape with full post processing and dynamic shadows for 4-CSM sky shadows, multiple spot lights and multiple point lights at the same time with excellent performance.
I just described here a setup with which I compare my engine against others or other engines with each other. It contains typical performance neck breakers and if you can handle such test scenes well then you are well off for a real world project. Most important with these setups is to test the internal capabilities. Just cranking up triangles count on a good card doesn’t proof anything. What hardware is able you can see from the specs and all engines are bound to the same hardware limitations. Besides nowadays games are fill-rate limited ( or shader limited… not exactly the same but symptoms are similar ) and not vertex limited. What gives though is how the engine in question handles these limited resources internally. That’s about distribution render time to what requires it and optimizing rendering ( when to use costly shaders, when to down-tune, when to boost ). So even while this is no “single objective” to judge game engines by you can put something together which can push them to see what they are made off. That’s what I went for. After all I’m interested to see what the blender game engine “really” is capable off. That’s the interesting point… especially for people who play with the idea to use it. Show them what you can do… not in “universitarian” scenarios but real-world scenarios.
In the final Blender 2.48 you will be able to disable collisions between objects that share a constraint. Apply this to the shower curtain rings, to make it more stable.
Here is an updated file, but you need latest SVN or final 2.48: http://bulletphysics.com/bathroom_final_dof_noise_nocolconstraint.zip
In my test scene I just tested things I was likely to include in a game project. And my test results indicated 800k for my particular approach which was to use lots of polygons but to save the special features for when it counts. As opposed to just throwing mirrors in there for the heck of it. Some of the best games ever don’t have mirrors. And I hardly find myself thinking, “Gee wiz this game would be so much better if only it had mirrors in it”
It’s not even that I’m disagreeing with you. It’s just that my point is that 99% of the worlds greatest computer/video games have been played on less sophistcated engines than blender game engine.
which is how blender game engine is already - I dont know - WORTHY?
I don’t know where you take this attack from. I never said anything along this line so please don’t attack without a reason. What goes for mirrors I mentioned them to artificially multiply the stress on the test scene without having to manually push in more triangles since a mirror reuses geometry. I never said that mirrors are required to be supported or that games are bad without mirrors.
It’s not even that I’m disagreeing with you. It’s just that my point is that 99% of the worlds greatest computer/video games have been played on less sophistcated engines than blender game engine.
That is correct. But here the idea has been to show “how good blender is”. So this implies to show the most muscles to the audience to impress. For this it requires a setup tackling situations where even AAA game engines choke before giving it a stab. It’s about finding out how much the engine can be pushed. This is what is interesting to know about a game engine: where can it get me? I’m only interested in this point, neither in attacking you, neither the blender team nor anybody else. I’m just interested to see where the limits are, to see what is possible. And for this you have to push it to the edge until it drops :p.
which is how blender game engine is already - I dont know - WORTHY?
In this particular situation I have to get a bit “personal”. You here at BA ( or blender in general ) have a little “complex”. Is the original Doom engine “not worthy” because it can not do dynamic soft shadows or skeletal animations? Of course not. Even if not remotely as powerful as todays game engines it is one nevertheless, just for a different target hardware. So the blender game engine is a game engine as much as the original doom engine. The difference is just what performance profile it has. So please stop this bitching folks. Everybody who knows his trade knows that blender is a game engine and all others are morons anyways.
I downloaded this yesterday and it worked fluidly (until the floor was covered in ducks, of course), but now Blender runs incredibly slow no matter what file is open. I’ve tried deleting the file, updating Blender, virus scanning, and restarting my computer, but nothing seems to solve this.
Escuse me?
I didn’t attack anyone infact I felt like my tone was very respectful. Did you even bother to read my post before accusing me of attacking someone?
What did I say that sounded like an attack? I was discussing blender benchmarking, that was what i was talking about I’m not sure what anyone else is reading into to it. But yeah you got it Erwin. No more talking :eyebrowlift2:
Amazing! I couldn’t get glsl to work for me when I tried messing around on my own, but it work great (although slow) with your demo. Unfortunately blender crashed after a few seconds of the game engine.
I was also having crashing problems, so I tried your build. This build doesn’t seem to show the noise/grain effect and also there’s a new glitch which shows a small version of the scene in the middle of the screen. See the attached screenshot. I’m running an old Nvidia 6800 Ultra (mobile in a laptop).
Previously I was testing with the 2.48 apricot/no-python build from graphicall.org. The grain effect shows up, but it crashes (especially with DoF). The earlier bathroom demo files didn’t crash.
hmm?