Hey all, I am attempting to adopt a VRay RT workflow to render on my video card with CUDA. For this reason, I just installed a GTX 970which has 4GB VRAM. I incorrectly assumed that the 4GB would be sufficientto cache my Maya scenes on the video card before rendering, but it seemsI was wrong. I continue to receive logs like the following. Here is an excerpt from the time a render fails:
“VRayRTEngine Fail Log
Lightmap successfuly loaded[2015/Aug/19|02:03:39] GPU light cache: 18848 samples stored[2015/Aug/19|02:03:39] GPU light cache: 4210 nodes stored[2015/Aug/19|02:03:39] GPU light cache: 4211 leaves stored[2015/Aug/19|02:03:39] Building optimized CUDA kernel for comp cap 5.2 for device 0...[2015/Aug/19|02:03:39] Optimized CUDA kernel assembled successfully[2015/Aug/19|02:03:39] Total memory for device 0 : 4095 MB, currently used 4070 MBBufman 2875.67MB, Mtls 6.89MB, Geom 0.20MB, Light Cache 1.38MB, Total 2444.89 MB (1588)[2015/Aug/19|02:03:49] error: ./src/ocl_tracedevice.cpp(3477) : CUDA error 2 : unable to allocate enough memory to perform the requested operation (out-of-mem)[2015/Aug/19|02:03:49] Closing RTEngine (cancelled)”
My question is this: can anyone point me to some resources that addressoptimizing scene sizes at render time? I do not know what is causing the VRAM cache requirements to be so high. I have no idea how to estimate howmuch average VRAM a scene will require, 8GB? 16? Any insight will be dearlyappreciated.