demostenes Posted April 1, 2013 Share Posted April 1, 2013 "Increased the number of digits for Density field in ObjectMeshClutter and WorldClutter. Now it's possible to set such values as 0.0001 for it." Does not work (objects disappears, latest version of Unigine), lowest number I can put is 0.003. I was trying it for object grass (billboards), ObjectMeshClutter and WorldClutter. Link to comment
manguste Posted April 2, 2013 Share Posted April 2, 2013 It depends on the Step. Increase the Step in order to render anything with such small values. Link to comment
demostenes Posted April 2, 2013 Author Share Posted April 2, 2013 It depends on the Step. Increase the Step in order to render anything with such small values. Now I have step according to your recommendation and it seems quite optimal (= what is best for performance). Increasing/decreasing this value affects performance. Value is indeed really small, but I have still too many objects (trees). I cant imagine using value like 1, it would lock any computer. I will try to modify clutter mask somehow, we will see, if this helps. Link to comment
demostenes Posted July 29, 2013 Author Share Posted July 29, 2013 Even if i increse step to 200, it is still not enough and performance impact is big. I am preparing quite rich forest (lots of kinds of trees), so I need very low density for each tree, but it would be memory waste to have indivual mask for almost each tree...Is there any other possibility, how to handle this? Link to comment
necris Posted July 29, 2013 Share Posted July 29, 2013 Demostenes,Step in objectClutter affects on performance only on trees generation stagePerformance drop depends on density of trees, visible distance, LOD's and distance of switching to impostorsAt what distance you switch trees to impostors?May be we can say something more specifically if you provide us a test scene with several kinds of trees. Link to comment
demostenes Posted July 29, 2013 Author Share Posted July 29, 2013 Demostenes, Step in objectClutter affects on performance only on trees generation stage Performance drop depends on density of trees, visible distance, LOD's and distance of switching to impostors At what distance you switch trees to impostors? May be we can say something more specifically if you provide us a test scene with several kinds of trees. I am switching to impostor cca 80-100m (depends on tree). I have experience, that for very low poly trees (to cca 1k poly), LOD does not make sense, because with LOD it is far slower than without LOD. I was toying with step, for mesh trees I am using 20, for billboards 200 and thes settings are giving me the best performance. Actually it is even very close tu numbers recommended by Manguste (she told me, that step should be cca 4-5x lower than visibility). My scene is nothing special, just cca 15 kinds of trees from 500 to cca 2k poly. http://imageshack.com/a/img29/8890/5o7g.png Link to comment
necris Posted July 30, 2013 Share Posted July 30, 2013 Beautiful forest :) May be you can try to switch small trees earlier. I'm sorry, but it's really difficult to give you specific advice about optimization without a scene. Link to comment
demostenes Posted July 30, 2013 Author Share Posted July 30, 2013 Beautiful forest :) May be you can try to switch small trees earlier. I'm sorry, but it's really difficult to give you specific advice about optimization without a scene. I have no problems with performance, I just need to have billboards for these trees to have visibility of this forest from bigger distance....for example Kapok is tree 70m+, so it is really important to be able to see it from distance. Concept of our world is visibility of all forests cca 3km.... My problem is, that because of lots of kind of trees in this scene, I am not able to set density low enough. For standard forest with 3-4 trees it is no problem, but by using 10+ I need to add on more decimal place to density. And having individual clutter mask for each tree is, well wierd.....Anyway it is what i am doing right now. Link to comment
demostenes Posted July 30, 2013 Author Share Posted July 30, 2013 Well, I faked it little bit (billboards are not 1:1) with all trees, only with some, this is result: http://imageshack.com/a/img62/8651/zb9q.png Btw, do you know what has by far biggest performance impact? Billboards. Typically I have only cca 13k grass billboard triangles in view and it eats 15 FPS (1024*256 BB texture, with mip maps). In comparison hundreds of thousands of mesh tree polygons have impact like 8-9 FPS (few hundreds of mesh trees). There is definitely HUGE space for speeding up grass/BB rendering. Link to comment
sebastianbah Posted August 7, 2013 Share Posted August 7, 2013 Hi,What type of billboards do you use for grass? ObjectBillboard or ObjectBillboards? Thanks! Link to comment
demostenes Posted August 8, 2013 Author Share Posted August 8, 2013 I am using ObjectGrass with grass_impostor_base material Link to comment
frustum Posted August 9, 2013 Share Posted August 9, 2013 Is "ambient light" material toggle enabled? Link to comment
demostenes Posted August 13, 2013 Author Share Posted August 13, 2013 Is "ambient light" material toggle enabled? Sure. Link to comment
frustum Posted August 24, 2013 Share Posted August 24, 2013 We can check Engine bottlenecks if you can provide a small scene sample. Link to comment
demostenes Posted October 29, 2013 Author Share Posted October 29, 2013 I am afraid that in this case simple test case would be whole project with 15+GB of data and thousands of meshes.... Anyway I have interesting discovery. For clutter 4096*4096 12000*12000 appropriate density for vegetation is 0,005 (our case, actually it makes pretty dense forest). When I subdivide such clutter for performance reasons(*) to for example 2048*2048 6000*6000, I need to change density to 0.026, to have identical vegetation density. Which is quite problematic, because lowest number possible is 0,0026, I am also loosing lots of "detail" of vegetation distribution. I really dont how to cope with this, how to prepare mask to not have this problem. Big areas are no problem at all, but smaller are. *I tried to workaround this problem by having full size mask (4096*4096 12000*12000) even for smaller areas, for example only lower right corner with real coverage 1000*1000m, so almost whole mask was blank (black). And surprisingly there is quite significant performance drop, when I am stayng on "empty" place of mask, several thousands meters from first objects, it still eats 3-5% of rendering performance. This is surprisingly lot, I wasnt expecting that. Link to comment
frustum Posted November 4, 2013 Share Posted November 4, 2013 The limit of per-cell vertices count is 65536. The grass stops seeding when the number of vertices exceed this number. The only option is to reduce cell side. Link to comment
Recommended Posts