I have nasty problem with V-Ray. All my projects usually have about 2M polygons. Those are rather detailed and complex scenes. It always come to a point when V-Ray crashes on any attempt to render.
The same thing occured in my latest project - swimming pool. See attached screenshot.
When it had 1 970 351 polygons, it rendered just fine, starting a couple of seconds after pressing Render.
After I added Bisazza mosaic made of 3D tiles (see below), polycount jumped to 2 115 907 and the file refuses to render anymore!
Where is the problem? Is V-Ray somehow limited to 2 000 000 polygons? Is it memory allocation problem? Maybe the problem is that every tile is textured? The texture is small, 100x100 pixels. Each tile has front face and 4 beveled side faces (so 5 polygons in total). Of course tiles are not NURBS but polygons. With NURBS the file would be huge and it wouldn?t move in OpenGL.
When I start to render, Render Progress pops up, showing some information. When it comes to Prepass, it… well, not actually crashes, but doesn?t do anything. Seemingly it?s working but if you let it run for 1 day or 1 year, it would still look like this.
I found a method how to find out that it actually crashed. I can click into black area of frame buffer and zoom using mouse wheel in and out. It instantly crashes, showing Unhandled exception error message.
This bug danger for me before it appears a couple of days before the huge project must be finished. This is the time when polycount approaches 2mil. border. I must finish the project in 3 days but I am not able to render one single pixel. Any ideas? Has anybody tried V-Ray on such complex scene?
UPDATE: when I deleted some other objects in the scene so that the polycount dropped to 1.6 million, it rendered fine even with texturesd 3D mosaic so the problem will be burried really in polycount, not in textured mosaic. Can anybody (Vlado) confirm that there is some technological limitation with large scenes?
If I remember me right, I have posted this problem some weeks befor. A simple test was, I have create a sphere, mesh it with very fine mesh settings - 500.000 or more per sphere. Than I have copied the sphere several times. My conclusion: polygon count problem.
I doubt its an actual polycount limitation - its just memory in general most likely. If you’d like to send us the scene for internal testing we may be able to figure something out. You can upload it to ftp://ftp.asgvis.com/incoming
If its sensitive material you can password protect it and just email the name of the archive and the password to joeb@asgvis.com
Don’t worry about the size - all textures etc should be shipped along as well so we can completely replicate the problem. Please make sure the paths for the textures are something off the C drive so we don’t have to reassign everything.
the big scene that I render on Chaos: Industry-leading design and visualization software is at 2 284 563 polygons count I think,
it renders fine, but as you I passed through a period when it was crashing everytimes I pressed Render. Especially with Light Cache calculation.
Upgrading the computer in ram and performance solved my problem “for now”
Jp
It is all very scene/system specific. Right now the way we handle blocks is not very efficient either. So a scene with a lot of blocks can eat up the RAM very quickly.
Today I added new RAM to my machine - total 3GB. 3GB switch enable.
Test: scene with HDRI and directional light → one meshed sphere with 1 Mill polygons. OK. Duplicate the sphere → more than 2 Mill polygons → OK. … 3 Mill polygons → Rhino hang up at 2GB RAM usage … 4 Mill polygons → Rhino crash with 2.18GB RAM usage.
So, my extra RAM dosn’t help. The reason could be a Rhino.exe 2GB limit or a Vray for Rhino Bug. From the chaosgroup I know, that Rhino for Max good handle very high polycounts.
What I might try in an effort to get the poly count down is to mesh some objects before you render. Either mesh the simple objects and set the render mesh for the more complex ones or mesh the complex objects and set the render mesh for the simple objects. That way the objects that don’t need alot of polys can be simple. In Rhino V4 there is the option to set the render mesh individually for objects… Hope this helps
Thank you. It’s not realy a solution for arch viz with much deatails and trees, and people and cars … . I hope the reason for the crashes can be find and fixed. Maybe Rhino 4 will be a solution.
Maybe we can try to collect all tricks and hints for a better RAM usage:
- objects based meshing per meshmanager (old Rhino bonus tool)
- object based meshing per NURBS to mesh conversation
- blocks (here we need a better workflow to edit the material of blocks)
- usage of low res textures (an old wish is, that the user can setup a downscale factor in the texture dialog, so Vray automatic downscale map - could be work together with the filter map size)
A new special trick from me: after press “render” and during Vray is loading textures, the user can open an empty Rhino scene. The current problem is, that the Rhino.exe hold twice the scene in the RAM - one in the viewport and one in the Render engine. The OS Windows allow each task 1.7GB .. 2GB only. The big advantage of an from Rhino.exe independent render engine would be, that Rhino and the engine could use independent RAM. I suppose so that the 2 Mill polygon limit stay so long Vray use the Rhino.exe. In scenes with many texture maps are only 1 Mill polygons in the scene a problem too. Here it helps to clean the rhino scene during the render start process.
Any suggestions for other system settings?
Idea: could it be possible to use a second rhino.exe for VfR render process only?
The blocks one is a huge weakness in the current VfR setup. Each instance of a block is currently separately meshed ( as is the default Rhino render sdk setup ). I plan to pull us off the default sdk implementation so I can customize the block stuff so there is only one meshed instance of a block.
We did want to look into directly using nurbs geometry - but we need to do some testing to see if that would truly be a benefit to us, since I haven’t tested out V-Ray’s handling of NURBS
Direct NURBS support? Sounds great - this could be a great extra feature. The question is, how good handle Vray NURBS? Maybe, Rhino user could delete all rendermeshes and render NURBS without a big RAM usage.
Joe, what do you think about an extra rhino3.exe task? Is this possible? The same *.exe for modelling and rendering is a big disadvantage.
Other question: how can I clean the Rhino RAM usage? … the undo buffer?
Thats what I have to test. I don’t know if V-Ray handles NURBS better than McNeel does already.
Joe, what do you think about an extra rhino3.exe task? Is this possible? The same *.exe for modelling and rendering is a big disadvantage. Rendering is a very expensive operation - even with low thread priority it will probably eat up a decent amount of your system resources - especially if its a large scene with large textures, etc. I suppose it could be easier to work if you had the rendering in a separate process and then manually took system resources away from it. I have never personally tested this, but its worth a shot I suppose.
Other question: how can I clean the Rhino RAM usage? … the undo buffer?Thats a question for McNeel, I’m not sure to be honest. The memory imprint that VfR maintains is geometry / material / light information from the previous render so it doesn’t remesh unless absolutely necessary.
#1 does your specs of your motherboard told you that it’s ok too use more than 2Gb ?
2 Does the new Ram is the same company, capacity…etc than the other two ?
#3 If no, does it support the same overclocking that you did ? (cause instability if one is less perf than the others. you can test it by cancel the overclocking to test)
4 Sometimes, the ram works better in pair of two. you can put 3, but the overall gain is better on pair.
#1 does your specs of your motherboard told you that it’s ok too use more than 2Gb ?
No, it’s not recommended for 32bit.
2 Does the new Ram is the same company, capacity…etc than the other two ?
100% the same
#3 If no, does it support the same overclocking that you did ? (cause instability if one is less perf than the others. you can test it by cancel the overclocking to test)
I have lower it a little, but I will test it without overclocking.
4 Sometimes, the ram works better in pair of two. you can put 3, but the overall gain is better on pair.
Good to know.
Thank you. My hope is the next Windows that allow to use more than 2GB per task.
I think you can easily find some closed to release version of Vista online. I didn’t try it. but you could…just to see if there’s gonna have hope at the end.
