I have glu3d gpu particles exported as PRTs and loaded back with PRT loader (total 1.3mil). Now Frost is created out of PRT loader and PRT volume object created out of from Frost to increase particle counts. I set 1.2 spacing for both viewport and render. This gives around 2.2mil particles in viewport (subdivision is set on in PRT volume and also “disable subdivision” is turned off in viewport display) So it shows all particles.
It is 2.2 millions particles around for each frame in PRT volume. But when I go to certain frame, Frost gets calculated fast (set to 2.0 viewport/render resolution, relative to radius) and then PRT volume, it takes a little time but takes upto 20GB of ram for just 2.2 millions particles.
Why does it take 20GB of RAM? Is this normal or I am doing something wrong? I am running versions posted on thinkbox site for both frost and krakatoa.
hmmm but it’s just 2 millions faces in frost. and when it calculates frost, it doesn’t go that high in RAM.
Do you mean PRT volume needs less faces in frost?
Try to render the Frost itself in VRay without the PRT Volume generation. If you run out of memory, you will know that Frost is the one using up the memory. Since you are using relative meshing, it is possible that the radii get too small and your Frost adapts to that, producing increasingly complex topology. But since you said it was fast, this might not be it.
Back to PRT Volume, try to increase the Render Time Spacing 2x while increasing the Subdivisions by 1. This will sample the Frost mesh less precisely into voxels, but then produce a similar number of particles. If the PRT Volume is the one that is running out of memory while trying to sample the Frost mesh, this should improve the memory usage.
If neither of these approaches helps you figure out what is going on, please post some screenshots of the “good” and “bad” frame’s PRT in the viewport so we can muse about what could be causing the difference…
But Here are the screenshots I am posting which explains.
In 1.jpg, it shows glu particles exported to PRTs and loaded in PRT loader.
2.jpeg, shows frost mesh made using that PRT loader.
3.jpeg, shows PRT volume created out of frost. As you can see, It generates 1.1 mil particles and in task manager RAM usage goes very high (Particles are lower than PRT loader)
Based on the screenshots, I see two things happening. Over time, the surface area of the Frost increases dramatically. This will cause a lot more faces to be generated as the particles turn from a block to a spray. Second, the bounding volume of the input to the PRT Volume is increasing at the same time. Maybe like 5 or 6 times larger? This will result in a much larger levelset volume to place the points into, and that will increase memory consumption.
You cannot apply a Magma OVER a Frost.
But you can either apply a Magma UNDER the Frost (to a PRT Loader, setting the Radius channel), or a Genome over the Frost (shifting vertices along normals). I would suggest the former…
I tried with Bobo’s suggestion of increasing “subdivision” and it worked quite nice so far in viewport. Also tried “absolute quality” in frost with 1.0 which gave low res 5 lacs poly mesh. Easily and fastly increased particles counts to 5 millions (frost helped to remove some random particles as well)
