'WavZ' rendering in Sycles

[FXDude]

Hi! long time no post!

All this was from a while back, but I only recently made the video-edit



Translating and rendering is super quick, and I also tested with millions more,
the following has 263 million particles !
by duplicating the single patch without caching each unique patch only for this test,
and rendertime was barely phased at all.



Cheers, & nice to be back !

[FXDude]

... the following has 263 million particles !

Sorry that was 25 million and not 250 million,
I miscounted the number of digits in the commaless "number of points" scene stat lol.

But in the "infinite ocean" setup, I went up to 93 million unique points (without miscouting digits ),
with an average of about a million particles at any given frame in each "patch".



And with that many points, orbiting was about a second per frame, but was still navigatable, and that's using the much faster particle OGL shader,


And by curiosity, just recently did I try using instancing,
and instances of pointClouds in Sycles are really practically weightless!

So I tried "pushing the limits" to see how many instances could be handled,
and went ahead making an array of 120 x 120, totalling 14,000 instances.

14,400 instances
x 1 million particles
= 14.4 billion particles !




You can forget about displaying this many particles in the viewport, all instances were displayed as bounding boxes.

I then went ahead duplicating that array 4 times
making a total of ~56,000 instances and ~56 billion particles !



But then the bottleneck became viewport interaction speed with 56,000 instances in the scene,
as tens of thousands of even nulls will considerably slow down interaction, but was still workable at 2 frames per second navigating.

For the sake of it, I then tried duplicating the 56,000 instances array, and XSI crashed lol

So ya! ... 56 Billion particles without breaking a sweat !
(17 second region render, which is only slightly more than rendering a single patch ! )
while consuming less than 9 gigs of ram!

But you don't need TENS of billions of particles to cover a quite large surface area,
the following is a more easily manageable 2,000 instances (~2 Billion particles )
with randomization in the tiling to break-up the repeating tiling pattern, especially when looking-off towards the horizon.






__________________________________
Otherwise, correct me if I'm wrong, but rendering that many points in either Maya or Houdini seems to involve HUGE amounts of RAM.

In Houdini
How to render over 1 billion particles

I have a flip simulation cached out as points only, and there are round 13 million particles per frame, i want to bump up the particle count at render time with point replicate procedural, say about 100 times more. So the total particles to be rendered will exceed 1 billion.

Unfortunately, mantra can't handle such amount of particles, and houdini crashes. By the way, i have 64g ram. So is it possible to render such massive particles under 64g of ram?

What is the maximum particle count mantra can handle giving that 64g of ram.

it should be possible to render huge amount of particles but i think you will struggle with your ram.

have you seen this video?
https://vimeo.com/142517418

with 64gm ram i don't know how many particles you can render with it…. can't help there sorry….

Yes, i saw it. But it says 4 machine with 200GB memory each, so each will take one quarter, that is 250 million particles. Is it?

______________________________
In Maya, there aren't many attempts or examples of higher particle counts (for either simulation or rendering),
a while ago I've seen a video with 50-60 million particles using the first Bifrost (Naiad in Maya), which also involved 128 gigs + of ram
But basically no references of using the new Bifrost graph for large scale sims, maybe also because MPM is known to be rather slow?

Question about Maya simulation engines
https://old.reddit.com/r/Maya/comments/qqm7g9/question_about_maya_simulation_engines

Personally Autodesk as a whole doesn't make the simulation 'engines' that appeal to me. Bifrost is their newest craze and it's only geared toward high end production. You can't get quick, dirty results from it, either perfection or nothing. So it's slow, and my 8c/16t 32gb of RAM just chugs on even a basic sim.

_______

why Bifrost doesn’t use the GPU
https://www.reddit.com/r/Maya/comments/ns034x/one_of_the_big_mysteries_of_life_is_why_bifrost/

Bifrost is meant to be able to achieve high quality large scale simulations. The downside is that it uses A LOT of memory. It easily uses up 64 GB of memory and more when dealing with somewhat large simulations. Even smaller simulations take up over 10 GB of memory. You simply cannot fit this on most GPUs, unless you are willing to spend a fortune on one.

Nevertheless, in this case, it really seems like "radio killed the video star",
or that VHS overtook ... not just Beta or BetaMax, but Digital Betacam lol

.

[Maximus]

which version of scycles were you using there?
when i use latest scycles version on my renderings, mostly scattering geometry, grass, trees, proxy, it is insanely slow, like render starts after 5 minutes of loading all that stuff lol
im used to redshift which is blazing fast to cache stuff..

[FXDude]

which version of scycles were you using there?
when i use latest scycles version on my renderings, mostly scattering geometry, grass, trees, proxy, it is insanely slow, like render starts after 5 minutes of loading all that stuff lol
im used to redshift which is blazing fast to cache stuff..

Hey! sorry for the extra long delay, I have yet to try geometry instancing, but will do a few test and get back to you in a bit.

_____________________

Otherwise, just to show some of the latest "WavZ" renders (from a while back)
featuring post-simulation densification, and repeating wave patterns, not only in space as usual for spectral wave deformers (what makes it tilable), but also in time ( thus also making it loopable ! ) .

Never-ending beach waves



For the beach setup, (like the previously shown "infinite ocean") simulation is only the topmost whitewater particles instead of all surface particles normally required for beaches/oceans.

Allowing to VERY efficiently render very long sequences of beaches going off to infinity.
( mere 3 seconds per tile per frame to simulate, and 2 minutes per frame to render )

The following is very much a WIP or without look refinement ... or the rest of the ocean beyond beach waves,
( moved-on to other things in Blender and ComfyUI before getting there, but might get back to it at some point )
but should be enough as a "proof of concept" for very large scale seamless ocean/beach simulations.



In any event, hope you're all doing well, and cheers !

[FXDude]

which version of scycles were you using there?
when i use latest scycles version on my renderings, mostly scattering geometry, grass, trees, proxy, it is insanely slow, like render starts after 5 minutes of loading all that stuff lol
im used to redshift which is blazing fast to cache stuff..

Hi ,
I tested with ~50 instances, themselves each made of ~50 other instances, made of a bunch of textured fishes with 100k triangles each (making for 171 million triangles at render time)


and rendering was but a few seconds ...

________

I also tried with the somewhat recent "Create particle instances on edges" from Koyla to test with particle instances,
making 35,000 fish instances on the edges of a fish, and was also but a few seconds to initialize/render.



My Sycles ver. is 2.3, and so what's your version, and would you happen to have a scene file ?
(if you're still around lol)

Cheers!