Daniel Silva

Turns out regular RAM usage still goes up when using OpenCL as only small parts of the solver are accelerated. The estimation I was thinking of is way more complicated. It would be pretty neat to have a shelf tool that analyses system resources and one or multiple sims and lets you know what the max particle density you can hit on each is and also a way to decide how to split resources between multiple sims...

I would like to have a tool that allows me to easily max out and not overflow the vram when using OpenCL. First step would be knowing what exactly is loaded in Vram. Is it only the raw sim data for the previous and/or current frame? Do other things get loaded as well? There's also the size per frame changing and I'm thinking you could run a 1/3 res sim and then use that to estimate what the peak mem usage would be at full res...

I think I figured it out after running it a couple more times and getting out some diagnostic flipbooks. I think the feedback scale (1.2) was making the angular velocity on the small pieces go haywire. You can see it clearly in the video I've attached. I'm resimming with 2 substeps at the DOP level, slightly more particles, tweaked densities for the rocks and the boat and higher rotational stiffness on the rocks. I'll report back with the results tomorrow

https://drive.google.com/file/d/1rC9Ewc2dzjrOXp83bKRgxYSRXz-tCGUv/view?usp=sharing [drive.google.com]