The Team Needs Ideas for tests for JGLMark Please Post Your Ideas Here.
The benchmark so far will contain at least 2-3 scenes much like 3Dmark03/05. Also we intend to have some GLSL and ASM shader tests.
Don’t underestimate the scenes in 3Dmark03/05 
Large development teams creates the engine, large art teams modeled the world. When designing, both devs and artists realized in which benchmark their work would run: “what are we benchmarking here”. Something to keep in mind when building JGLTest.
A massive particle-engine (z-sorted alpha-blended sprites) would be an indicator for truly dynamic content throughput. You might want to spice it up by using shader-based particle-engines using textures as data-containers. You should keep the sprites fairly small, because otherwise fillrate will be your bottleneck. Something like falling snow or waving grass would be ideal, no smoke.
A forest renderer would focus on static geom data combined with LOD, billboards and finally imposters (grouped billboards here).
HTH
[quote=“Riven,post:2,topic:24746”]
Exactly. I don’t want to see yet another lame-ass gay scene like the ones in 3DMark. They have no point really, there’s no indication of what they’re measuring, it’s like benchmarking Doom or Far Cry or whatever. And they become obsolete too fast (if you want to benchmark the latest GPU features).
I’d much rather see dozens of small tests, than only a few big ones. Build a clean API for tests (that the community can also use to contribute even more tests) that the main app can run and start implementing the tests, one by one, concentrating on specific features with simple scenes. Examples:
- Shadow Volumes
- Shadow Mapping (testing many different filters)
- HDR
- Perlin noise shaders
smacks self upside head
Good ideas Implement tests to benchmark features, rather then showing a scene.
Thanks Spasi!
But there is an advantage in a large scene… it is less of a micro-benchmark. Make a large scene with all of those features together in addition to having tests for them in isolation… heck just have checkboxes to turn each thing on and off in one big scene to see the overall effect as it might relate to a real game scene.
People also love to fire up the benchmark before and after changing GFX card, and not only see the perf difference, but see a “much cooler-looking, smoother animated, scene”.
So, don’t underestimate the eye candy.
For instance, the old 8-hardware-lights test with the mobius strip torus was a good test - visually attractive, but also stresisng the hardware.
Sure, a simple scene does not necessarilly have to be ugly looking. Eye candy should be there, but it doesn’t have to be massive (e.g. a room full of HDR reflecting spheres is cool enough). Keep it simple, focused and manageable. As long as the results are not vague and truly useful, additional eye candy could be easily added.
i’ve alway wanted to run a benchmark where you have a massive lord of the rings style battle scene, where two armies fight it out, the number of men being the graphic cards limit where it run smoothly at a certain fps. but then again thats just my imagination running of something simple to begin with would be more practical.
Post your thoughts and ideas here: http://jglmark.dev.java.net/servlets/ForumMessageList?forumID=1696
NOTE you must be signed into java.net
many thanks, I have sent them a mail.
Some TRUELY awesome shaders there, it would great to have one as the token JGLMark demoscene.