We are working on a project using mental ray's Final Gather. Rendering time is a huge issue for us (as it is for everyone). In the past we tested Mental Ray vs. Turtle and found turtle 6 to 10 times faster. The test was using a Caustics and Global Illumination render of a single piece of geometry. We are now doing a landscape animation with many trees that rotate with the camera.
Is turtle really that much faster, if so why is there no "Turtle Forum". Or am I using Mental Ray in an extremely inefficient way?
Turtle is one of thes new renderreds like vray and brazil that have obly ONE basic requirement about them and thats doing fast FG. What they win in faster fg lookup they loose on HEAPS of other things. (ok so maybe brazil actualy is a bit more rboust)
Theres o turtle forum sice i cant imagine all that many turtle users out there. But basicaly we cant jsut split of a forum for every renderer out there thet would just spread tha data too thin.
Howvere in my experience the dofferemce is more like 1.5-4 times. BUt thenagain i do quite a lot of optimizing on my fg . see just slapping it on and putting flickerfree values for the steps tends to make for BAD fg. You can do the same thing with considerably less rendertime by doing a ahead solve first. Even more with a half bake!
Often the avarage fg scene can be speeded up by 5-6 times by well tought design.
Anyway fg is not realy production ready fer anything beyond the HUGE farms and baking.
Ahead Solve, and Half Bake. What do you mean by these terms?
My scene is a landscape animation with a moving camera and basically static geometry. The trees rotate to face the camera, thats it. I have turned off final gather rebuild to keep from re-emitting the photons on every frame. I have also enabled pre-compute photon lookup. This has cut my render time by about 40 percent in Mental Ray.
OKay now the planes are actualy bad for the fg of the surroiunding not the plane itself. i usualy use spindles a sproxys in fg for the surroundngs for the planes. And mak sure they are onesided so teh plabne can se the rest trough.
Now because, fg works by taking the camera ad looking out of it to sample some appropriate points for the solve they differ from frame to frame. NOW the most expensive part of fg is shooting many rays or taking many samples. SO the idea is to walkj trough the scene and render ounly out fg points ahead of time for the entire sequnce. Luckily you dont need to sample ALL frames just a few here and there and write theese to the cache. NOw what MR does is it looks up cached file s first and if they fall within the tolerace of sampling desity, it uses a cached point instead.
NOw what this does is:
1) you need smaller step sizes because the steps are evaluated for optimal positions trough out the animation 2) You need less rays bacuse the points stay more or less pretty stable under the whole animation Saving HEAPS of calc time.
Partial baking. Sometimes the scene is HUGE and contains a lot of animation. Now this means that not all objects need to take part of the fg. Secondly such things as walls dont need to chek eachother in teh fg solve becaus ethey stay put so you can BAKE out teht part to a file texture wile adding this to teh obtained value of moving pbjects only.
Offocurse if the scene is static and the rendertime is long you might want to speed it up by baking the entire solution to textures, Offourse after this your scene probably renders at a modest pace of 2 secs per image. So even if teh bake takes allmost as log as the whole renderit can be worth it.
OH and PS dont compare equal settings but rather settings that produce comparable imageryy since teh intteligence of the sampling algo plays role here. for exmple if yoiud compare newest mr with a few incarnations oler ver your dget the new ones SLOWER on equal ray amounts. But thing is it needs approximately 1/3 of all the rays, winning hands down.