Killer.Penguin Posted March 3, 2014 Share Posted March 3, 2014 Well I'm having more problem with your cloth system. It seems that cloth in our scene is really gpu expensive. On our good GPU card, having 6 characters with fully-cloth-rigged got our FPS down from 60 to 4. (I'm not having a lot of polygons on them either, roughly less than 500 tris per character.) Try bringing iteration down to 1 didn't help much neither... I tried the cloth7 demo and those cloths seem to have no effect on GPU at all. So I'm wondering what am I doing wrong? Is it b/c they use particle attachments? Link to comment
Killer.Penguin Posted March 3, 2014 Author Share Posted March 3, 2014 I think I kinda spotted the problem, it seems that sometimes the cloth system got stuck in a weird loop, and it cause the calculation to go berserk. Any help on this? Link to comment
sebastianbah Posted March 3, 2014 Share Posted March 3, 2014 Hi, I'm afraid we can give you any specific advices without a test scene.It would be very helpful if you can send us your scene with heavy GPU load. Thanks! Link to comment
Killer.Penguin Posted March 4, 2014 Author Share Posted March 4, 2014 Well, we spent a whole day trying to debug it ourselve and fail, so we just send you a chuck of our game to your support mail. Link to comment
unclebob Posted March 5, 2014 Share Posted March 5, 2014 Hi there! Don't be afraid of that, the response is coming... right now! I've run some tests and found interesting results. First of all, I noticed significant performance drop only on some characters' hair so if you leave only cloth there'll be no hiccups (poor characters). Also, hair simulation works nicely on Fay and Menalisa characters. I believe it's related to hair topology and collision shapes. You can notice that most of the pivot points of the characters' hair are inside collision shapes — this might lead to significant performance drops. So from the concent side try to leave enough space for hair and cloth or replace hair by skinned mesh. From the tech side try to reduce physics ifps (engine.physics.setIFps) and scale (engine.physics.setScale). Your art is beautiful! Link to comment
Killer.Penguin Posted March 6, 2014 Author Share Posted March 6, 2014 Okay... so "cloth inside collision shapes" caused a significant drop in FPS. With enough play time, that's something that's bound to happen no matter what. Can't you help fix this issue? May be add some exit loop inside your cloth system if it goes too far? Using skinned mesh instead is not what we had in mind.. ps. I also suggested you adding a rubber co-efficient to cope with this issue (https://developer.unigine.com/forum/topic/2271-solved-rubber-cloth/), but it seems to get no serious response yet. This is something that will help making this cloth system more workable, but that's another issue though. Link to comment
Killer.Penguin Posted March 25, 2014 Author Share Posted March 25, 2014 Seems like everyone moved to hybrid cloth system: "Regular simulated cloth behaves too unpredictable." - nVidia (http://physxinfo.com/wiki/APEX_Clothing) Any case, we found out the hard way because we ran into many cloth problem in Unigine, making this feature unusable in real practice. So will I hear any response on this issue? Link to comment
Recommended Posts