Tweet
I noticed on some recent exterior renders that Vray was taking much longer to render than I would have expected (on the order of 4-5 times longer than I'd expect). To make a long story short it was due to the presence of wparallax planes in the scene. Backing this up, it seems like while rendering the windows or anything that might reflect a window (e.g., leaves are SLOOOW) knocks the cpu usage down to 20% in task manager. If I removed or hid the planes everything rendered nice and fast at 100%. The wparallax planes weren't a huge portion of the scene (no giant windows or windows most of the frame). I'm in contact with Art at wParallax to see if there's some issue but I realize the OSL code isn't unique/original to wparallax. He suggested posting it here to see if any one has any insight or similar issues. My email to him is below that explains it a bit more in depth...
I've been using wparallax for a while now (I think shortly after it was released) and don't recall similar behavior previously) which is why I tried reverting back to Next. As I said though, no change. Maybe I just didn't notice the slowdowns before though since I don't really watch the task manager that much.
I realize it's not a good idea to directly compare VRay CPU to VRay GPU directly but it seems like something is getting messed up on the CPU side since the results are so drastic. I've attached the scene if anyone wants to try it out. You don't need the right EXR, any bitmap will do. I've also tried recreating the material using the VRayOSLTex map but haven't been able to get that to function as expected.
Is anyone having similar problems using wparallax? Any ideas what might be causing this?
Thank you for the quick reply. I'm not sure what's going on either. To answer your initial questions, no I'm not using any type of multlimap or multi-sub material and no I don't have double faces. The old OSL code actually seems to be a bit faster in my testing, but usually just by several seconds over 2-3 minutes of rendering. It seems to get slowed down in every scene I try it in so I've boiled it down to just a plain scene with 25 identical planes with one textured applied which I've attached. In the scene, the first two material map slots have the old (1st) and new (2nd) OSL shaders in them if you want to compare (I think they are set up the same just with the newer code).
Some additional info:
-to clarify, I'm running 3ds max 2020.3.7 on Windows 10. I've got a few other plugins running but nothing that I think should affect this: railclone, forestpack, rpmanager, bercon maps, colocorrect, to name a few.
-the slowdown is quite dramatic in the test scene for me. I've got a Threadripper 3970x with 32 cores/64threads and 64gb ram and I get times of around 2:30 (150 seconds) at 1920x1440 with VRay 6 Hotfix 3 (just regular CPU mode). The same scene but with a standard map in the illumination slot (for example I used the same "wP_Retail_v1...exr" from the sample scene) finishes in about 3 seconds (10 seconds if I use progressive rather than buckets which is somewhat inefficient in this case).
-if I switch it over to VRay GPU 6 Hotfix 3, it finishes in 3.8 seconds using both the old and new OSL shader using a GTX 3070 and the CUDA engine (3.9 sec using RTX). Even if I switch the CUDA engine over to using the CPU (3970x) only I get a time of 8 seconds.
...
I've recently updated to VRay 6 from Next so I thought that might be the problem but I tried reverting back to next and the problem persisted.
Some additional info:
-to clarify, I'm running 3ds max 2020.3.7 on Windows 10. I've got a few other plugins running but nothing that I think should affect this: railclone, forestpack, rpmanager, bercon maps, colocorrect, to name a few.
-the slowdown is quite dramatic in the test scene for me. I've got a Threadripper 3970x with 32 cores/64threads and 64gb ram and I get times of around 2:30 (150 seconds) at 1920x1440 with VRay 6 Hotfix 3 (just regular CPU mode). The same scene but with a standard map in the illumination slot (for example I used the same "wP_Retail_v1...exr" from the sample scene) finishes in about 3 seconds (10 seconds if I use progressive rather than buckets which is somewhat inefficient in this case).
-if I switch it over to VRay GPU 6 Hotfix 3, it finishes in 3.8 seconds using both the old and new OSL shader using a GTX 3070 and the CUDA engine (3.9 sec using RTX). Even if I switch the CUDA engine over to using the CPU (3970x) only I get a time of 8 seconds.
...
I've recently updated to VRay 6 from Next so I thought that might be the problem but I tried reverting back to next and the problem persisted.
I realize it's not a good idea to directly compare VRay CPU to VRay GPU directly but it seems like something is getting messed up on the CPU side since the results are so drastic. I've attached the scene if anyone wants to try it out. You don't need the right EXR, any bitmap will do. I've also tried recreating the material using the VRayOSLTex map but haven't been able to get that to function as expected.
Is anyone having similar problems using wparallax? Any ideas what might be causing this?
Comment