I am very impressed with RC after finally swapping out my radeon vii cards to test it out. I am currently running it with 3x gtx1080ti. I noticed this about cuda utilization:
1. During model construction (which is a combo of point map and meshing), all 3 gtx1080ti go to near 100% cuda for around half the time, then they all perform a copy task, and it seems the second half is done 100% by the cpu. Also while all 3 gtx1080ti were going full cuda, the process was so fast, I imagine having just 1 would only be a little slower.
2. A small burst of maybe 5% cuda utilization during coloring and texturing, only for a few seconds.
That's the only time I found cuda used. Now obviously RC is very special because that first part with point cloud/depth map generation being so heavily gpu accelerated does magic compared to anything else I've found. It can be the difference of 5 minutes vs 60 minutes for a small project, or the difference between 1 day vs 30 days.
I'm going to keep a second computer around with nvidia cards to use RC, but I really wish something could be done to interface with AMD cards. Once I have my meshes exported I immediately switch to the radeons. In davinci resolve a radeon vii compute performance is toe to toe with a rtx2080ti, and on top of that it has more vram.
Please sign in to leave a comment.