Tweet
We are currently converting a project over to vray from mental ray that involves quite a lot of xgen trees. They are currently low res standins around 5k tris each, three different trees about 110,000 total. It's taking a long time, upwards of 5 minutes during the compiling geometry stage, the actual render takes less than 30 seconds at this point, it's just a grey shaded first pass trying to get all the geometry switched over.
We need more trees and a lot higher res in the end here, I've tried quite a bit of different things.
It started out as the particle instancer with instanced vray proxy trees, that took even longer to start rendering and hung up the machines for 15-20 minutes on the farm. This is how it was setup before in MR with MR proxies and rendered 400,000 trees instantly so I started with that. I thought xgen would be better as it's a bit more modern and flexible, I tried the vrscene cache in xgen, that shaved off some time in the earlier stages of the render where is is translating the scene (around 40 seconds) so that's nice, but it still takes the same amount of time to compile the geometry. I have taken a fresh scene and imported the .vrscene file, same thing happens.
Is this the expected behavior? This is going to cost us thousands of dollars on the online render farm if we can't get the initial phase of this render down to lower times. I was expecting the .vrscene to avoid the compiling geometry phase, but I suppose I was mistaken on it's intent.
As a last ditch effort, I was going to try to export as the mel file in xgen and run that to generate regular old instances, but it's going to take a long to to complete and then we can't modify our trees easily anymore.
Would modifying the generated mel file to create vray proxies instead of instanced .abc files work faster?
We need more trees and a lot higher res in the end here, I've tried quite a bit of different things.
It started out as the particle instancer with instanced vray proxy trees, that took even longer to start rendering and hung up the machines for 15-20 minutes on the farm. This is how it was setup before in MR with MR proxies and rendered 400,000 trees instantly so I started with that. I thought xgen would be better as it's a bit more modern and flexible, I tried the vrscene cache in xgen, that shaved off some time in the earlier stages of the render where is is translating the scene (around 40 seconds) so that's nice, but it still takes the same amount of time to compile the geometry. I have taken a fresh scene and imported the .vrscene file, same thing happens.
Is this the expected behavior? This is going to cost us thousands of dollars on the online render farm if we can't get the initial phase of this render down to lower times. I was expecting the .vrscene to avoid the compiling geometry phase, but I suppose I was mistaken on it's intent.
As a last ditch effort, I was going to try to export as the mel file in xgen and run that to generate regular old instances, but it's going to take a long to to complete and then we can't modify our trees easily anymore.
Would modifying the generated mel file to create vray proxies instead of instanced .abc files work faster?
Comment