Hello guys,
there are so many old threads how to render the big amount of particles, I'm guessing maybe there are some modern approaches in newer Houdini versions?

Here is my workflow:
I have saved 100 mln of particles (pos, vel, color)

It takes 3 min to render and 30GB of Ram if I load them by Load SOP.
It takes 2-3 times quicker if I use Delayed Load Shader. But it's hard to use it coz I have procedural particle scale, dynamically changing colors by vex, etc in SOP level. Re-make everything in shader seems too complicated in my case.

Also, I tried Packed Disk Primitive in Load SOP. But to make color, pscale changes I must unpack it and performance drops like loading file as usual geometry.

Any other suggestions how to deal with big point counts? Maybe there is any more modern way?

Thank you in advance.

I would personally save out the points with their final attribute values for rendering, and any unneeded attributes/groups deleted, then you can use delayed load/packed disk to make for more efficient rendering.

Another option would be to use the houdini engine procedural, which can use any sop network at rendertime, but I'm not sure of the implications on memory use with this method. It is also a bit unintuitive to set up.

Thank you for a reply!

Yeah, it's possible to save particles with all attributes, but you see, that's the point - when you have such big data, you save only main portion of them. Because when it took 24 hours and 1TB of disk space and if you need to change little color again, it's crazy unproductive to have another saving/rendering round.
Anyway, thank you for a proposition, will keep searching.

Save your sim (24 hours) then reload it and add the attribs you need to change, and recache. That should be fairly fast. Yes, uses more disk space but you can't make an omelette with breaking eggs

definitely explore hengine point procedural, especially as auto procedural on object node, it's much more convenient than having to constantly recache, especially if the operation performed is fast like changing attrib values

tamte
definitely explore hengine point procedural, especially as auto procedural on object node, it's much more convenient than having to constantly recache, especially if the operation performed is fast like changing attrib values

Thank you very much. It seems like a solution. Of course, it sounds crazy to make a new HDA just for such purpose but trying to make it happens. Man, I miss Krakatoa))
Thanks!