Why is it so hard to get help on larger scale FLIP sims?

[coccosoids]

I was just wondering why it is so hard to get help om the topic of larger scale FLIP simulations, like oceans and similar...

In the span of a few weeks I opened a few threads here and in the official houdini community forum and I have gotten zero replies.

It would be helpful to know if this is some kind of a taboo topic in general.

[lukeiamyourfather]

Looking at your previous posts, I would say it's your questions rather than the topics that are leading to no replies. In my opinion your time would be best spent on tutorials and checking out the documentation to get a better handle on the fundamentals of Houdini. That would enable you to ask better questions that would be more helpful to you and more efficient use of time for everyone else. In general I wouldn't start the process of learning Houdini via complex topics like FLIP simulations and dynamics.

https://www.sidefx.com/learn/getting_started/

Master the basics and then everything else will be much easier. The beginner stuff might seem boring but it's essential to understanding the more complex topics. 

[coccosoids]

19 minutes ago, lukeiamyourfather said:

Looking at your previous posts, I would say it's your questions rather than the topics that are leading to no replies. In my opinion your time would be best spent on tutorials and checking out the documentation to get a better handle on the fundamentals of Houdini. That would enable you to ask better questions that would be more helpful to you and more efficient use of time for everyone else. In general I wouldn't start the process of learning Houdini via complex topics like FLIP simulations and dynamics.

https://www.sidefx.com/learn/getting_started/

Master the basics and then everything else will be much easier. The beginner stuff might seem boring but it's essential to understanding the more complex topics. 

First off a big thank you! For at least shattering my illusion that I was invisible on this forum...

I understand your point of view, even though I believe you've made a sort of generalization. But indeed I can see why presenting my questions in a wider and more detailed context would lead to a more attractive engagement proposition to the more experienced members of this forum. Because I am (somewhat exclusively) focused on the ocean tools now, I think most would agree that trying to level up on every aspect of Houdini before attempting to finish a simulation is not necessarily the best thing to do -but if indeed that were the case, then a more focused recommendation on what I should catch up on would be undoubtedly followed up on.

On the FLIP side of things even if I am only familiar with the concepts at the most basic level I can edit collision sources and wrangle v fields, I can decouple my collision from my particle separation and use different guides to debug what feeds into the solver, I can work my way around a VOP node to build a custom velocity field, I could probably build from scratch a basic setup for a guided ocean layer simulation and am getting into post processing flip sims with wrangles before meshing...

I know I still have a lot to learn, but can you at least tell me when and where on this journey would I be able to find the answer to a question like: distributed flip sims on a single machine? And why is it so wrong to ask that now and learn if what I need is more ram, a second machine or just... more experience?

[lukeiamyourfather]

1 hour ago, coccosoids said:

I know I still have a lot to learn, but can you at least tell me when and where on this journey would I be able to find the answer to a question like: distributed flip sims on a single machine? And why is it so wrong to ask that now and learn if what I need is more ram, a second machine or just... more experience?

It would be easier/faster/cheaper to get more computers to distribute the workload or get one computer that's better. I realize money isn't infinite but neither is time to kludge stuff. You can get a used workstation with 24 cores and 256 GB of memory for less than $2,000.

https://www.theserverstore.com/dell-precision-t7810.html

A machine like that would be plenty of power to learn with and create some impressive demo material. There are lots of other options but my point is to address the hardware problem if there's a hardware problem rather than looking for other less efficient ways to solve the problem.

[coccosoids]

14 minutes ago, lukeiamyourfather said:

It would be easier/faster/cheaper to get more computers to distribute the workload or get one computer that's better. I realize money isn't infinite but neither is time to kludge stuff. You can get a used workstation with 24 cores and 256 GB of memory for less than $2,000.

https://www.theserverstore.com/dell-precision-t7810.html

A machine like that would be plenty of power to learn with and create some impressive demo material. There are lots of other options but my point is to address the hardware problem if there's a hardware problem rather than looking for other less efficient ways to solve the problem.

 

Thank you.

Right now I have 14 cores and 64GB ram. The 'hardware problem' presents itself due to the intended final sizes of the volume of liquid. Due to the dimensions on the ship and desired wake length I will need to have a 200m+ long by 60m+ wide volume which in my currents tests fills up my ram capacity quite fast. This is why I was wondering if on a single machine it was possible to slice this in 2 and 'distribute'. I do understand that the simulations need to exchange data between them but I was willing to compromise quality and fidelity of the simulation for a larger simulation size. Would you say something like this would be possible? Otherwise I can augment my ram and bite the bullet on simulation times, as long as I can fit everything in memory.

I am open to any optimization tricks if there are any.

[tamagochy]

Distribution not help you, because with current FLIP solver it have no profit. So you should increase particle separation for your fluid or change size of the container. You should start thinking from camera point of view. If it wide shot you no need to detailed sim, if it close up you can cut container by camera and save details.

In general FLIP simulations very complex process and for beginners better to start from smallest dimensions to understand how its work. A lot of people cant do good flip sims alone. You can watch my FLIP RnD, I spent 3 month learning it and result far away from best, but I know how it works now, I hope))

[Andrea]

1. If you don't need a super high resolution sim/details, you can simulate a very low resolution ocean to have some sort of start velocities and then do some post sim processes on the ocean surface or work a lot on the white water details. Possible but not so straight forward

2. Making a lot of assumptions there, but in regard of the topic I would add that if you look around on the forum there aren't many discussions about big ocean simulations, and I would consider this forum a sample of the Houdini users out there.

The past summer I've been working on a show where I needed to do some complex/out of the shelf ocean simulations and I have to admit that I struggled quite a lot. The main issue was that I wasn't able to find materials / references or answers to my questions on the web or in papers. This, in addition to some high quality hardware requirements probably discourage most of the people that want to try on their free time.

I guess also that companies that often work on such shots have their own proprietary tools or digital assets (see for example ILM that recently developed a tool for creating oceans in the latest Star Wars). Whoever worked on them probably can't share informations or the setups were so different from what we have available by default in Houdini that would anyway be """useless their experience""" without explaining things in very detail other than say what everyone already knows ( see 1. for example)

Would be awesome if in the future SideFX will release some webinairs with the "suggested workflow" since they are improving so much the flip solver and related SOPs

[bunker]

You could also include some visuals with your question, that would help understand what you are trying to achieve. And no, you can't distribute anything on a single machine, "distributing" refers to using multiple machines. From what I understand, you're trying to use FLIP to simulate an entire ocean?
Usually ocean sims are done with a small FLIP patch where you need splashes (eg: a boat) and blend with an ocean surface (displaced grid).
If you absolutely need to, you can cluster simulations. The patches won't interact with each other but for an ocean that shouldn't be visible. That would require some to time to setup obviously.
 

 

Edited July 22, 2020 by bunker

[CeeGee]

Hi,

I did few rnd for distribute simulation and you can do distribute simulation on single workstation, for example if you have 64core you can split 2-4slices sometihng like 32core or 16core per slice. But you need to have a lot of memory
here are some stats and tests - http:// https://vimeo.com/380432448

In your case 14 core with 64GB of ram, is not good option for single workstation distribute simulation. Also simulating on linux can be a lot better then on windows, also simuilating something with flip is more of optimization to solve problem then actual brtue force.
 

You can see what Pixomondo did on Midway https://www.youtube.com/watch?v=kJ8Hz2Bjejc&t=10962s

 

Edited July 23, 2020 by CeeGee

[bunker]

well, I did learn something today... nice one Igor!