Greetings!
Still working to generate a ship on the ocean sim. Getting strange results with foam life outside the grid.
Nightly phoenixFD_adv_22501_max2017_vray_30_x64_26847.
This is frame 588 of 2000, attempting a resim to add splashes and foam. Detail reduction 2 in this screen shot.
cache.aur file size 412MB. Grid 73M cells, Liquid 25M, Foam 14.8M, Splashes 65M.
Seems as if only the foam particles with higher velocity are escaping the grid. Most of the foam is killed at grid even with outside life set to 10.
Does this happen only when you resim? Otherwise, the Outside life topic was discussed here in the forums not long ago - we will modify its behavior at some point, but for now you could check the other thread for more detailed explanation and try increasing the outside life.
Thanks for the reply!
I changed the foam outside life from 10 to 20 and am restarting the resimulation. I will let you know how it works, in a few days.
Ron
Update:
This is starting to take shape thanks to the tip from Svetlin.
Increasing the foam Outside Life from 10 to 20 helped some. I think it needs to increase even more. Grid 72.8M, liquid 25M, foam 17.2M, splashes 67k, mist 380k, no wetmaps. Noticed foam increased 3M.
The warship is at full military power run of 32 knots. I am impressed the sim produces good big rooster tail in the wake by itself. I need to also increase discharge rate of emitters on the ship screws to increase foam activity at the transom.
I understand now the foam particles from the rooster tail are able to exit the grid easily but foam particles from the bow wake are too slow at the back of the grid to cross outside within their lifespan. This adjustment gave the foam trail the expected taper and can be improved with more outside life.
My big wish list item is to use idle computers in my render farm to assist and speed the processing time of the sim and resim. With my i7-2600k@3.4 (water cooled, no overclock) my resim time is 6m45s and will increase with extra outside life, but that is the cost to produce this flow:).
good job man. I play with similar toys (boats), but my results are can’t even begin to compare with yours. you say the resim took 6.75m on your machine? how long did the simulation take?
I did a similar sim recently - grid 75 mil, liquid 45 mil, foam 116k, splash 3k, mist 9k, no wet. took 15h on a 2 x E5520 (a veteran begging for retirement).
tipped by our post I tried to resim my sim … just to find out I should have output those particles in order to be able to resim them. simulating it now all over again… and learning.
My original sim took 3 to 4 minutes per frame to start, up to 5 minutes at frame 1000+. Now I have made the grid wider so the sides of the foam spread is more natural, now 91M cells (500x2800x65), water 32M, spf 2, and am running the entire thing new. On this run I am also output wet particles since I think it did not work in resim to add wetmap. Now the base sim is taking 2.5m at the beginning but will likely be more like 5m++ each when the waves begin, and I expect the 2000 frames to take over 6 days.
you can add wetmap by resim, but with one limitation, the liquid consumption and the sticky effect will not work. the reason is obvious - they are affecting the liquid that is already simulated.
That is all you need to do. Just switch on “Enable Particle Resimulation” and press Start (You need the cache for the resimulation to be exported with Velocity channel).
It should change only the particle behavior and keep the liquid the same.
I thought I was doing something wrong as the Estimated time for the Resimulations was just about the same as the a Brand new simulations. Was this to be expected?
I thought Ron resimulates few days worth of simulation for just couple of minutes. Did I misunderstand or was it 6.75 min a frame?
Hi Baro!
I am running the base water simulation now with wetting on but with foam and splashes off. Each frame is taking 5.75 to 6.5 minutes each. When this is complete I will turn on resimulation and turn on foam and splashes. When I hit the simulation start button, I am expecting to take over 6 minutes additional each frame to add the foam and splashes to the base water. Some of the extra time is the outside life of the splashes and foam so the time required actually grows some more.
Hope it looks good, else I can do it over again! Won’t know for maybe 300 frames with render tests, etc. Time consuming!
By the way, I am trying to get around 45 seconds of usable material for creation of many long shots, so I am running 2000 frames so I can omit the first several seconds of grid startup to where the simulation appears stable. If average time to calculate the base plus foam and splash is 12 minutes combined then 400 hours of processor time or about 16 days (not counting the multiple cycles to test, adjust and try again). The ability to resimulate on top of the base provides a huge benefit compared to the time required to simulate complete with splash and foam, then realize something is not good and start over. Also my times include using a big solid state drive for my cache folder to speed the save process.
A good tip I read somewhere, also learned the hard way: turn off “Show Mesh” as well as Grid Content Liquid and Particles in the Preview Rollup while running the simulation and resimulation. This keeps time required per frame to a minimum. Check it often by turning on the “show mesh” to make sure the sim is not doing something unusual, but turn it off again when possible to conserve processor time.
Best wishes!
Ron
The direct splash/mist size control was removed, because it leads to significant problems in the render setup. When we initially introduced it, the idea was to follow the general physical correct model, however you can’t simulate a ship or waterfall with the real particle size and count, the particles are just too big in count. The real pipeline is to use lower in count and bigger in size particles, that works pretty well. Using manual size control, you have to guess what size to use with the current particle count, and this was hard for the most of the users. That’s why now there is automated count-size relationship, controlled by the parameters splash and mist multiplier. The meaning of the multipliers is how many result particles correspond to single parent particle, for example splash multiplier 5 means that single liquid particle is turned into 5 splash particles, when the physical correct model is used (affect liquid=1). The multiplier determines the splash size giving in account the volume preservation, this makes the setup much easier.
can someone please tell me what a the benefits of re-simulation vs simulation. in my few runs at this I observed no time savings (resim took almost the same time as a sim) and after outputing all particles, the frame solder turned form about 10GB to 124GB for 600 frames. was this to be expected?
Ivo, please clarify which exactly is the multilayered you are mentioning here “splash multiplier 5 means that single liquid particle is turned into 5 splash particles”. Is it the simulations settings:
or is it int the particle system settings:
One other thing - I certainly like the idea of lower in count particles which should reflect in faster simulations, but I noticed in the FLIP optimized ship tutorial you posted in another thread, the grid count I think was 91 mil, whereas in the Classic ship tutorial there were ~50 mil cells. What I am trying to figure is what would be the minimum workable sell size/particle count that will produce dissent results.
The benefit of the resimulation exists only if the resimulated particle groups have negligible simulation time compared to the liquid, this is highly dependent on the settings. For example a beer with thick foam head has no benefits to be resimulated, because 99% of the simulation time is spent for the foam. As opposite example if you have already finished simulation and decide that you need wetmap for the rendering, the resimulation is much faster than full simulation. There is no simple rule, you just have to try. If there is no advantage, then better use full simulation, because some features are not accessible in resimulation.
Ivo, please clarify which exactly is the multilayered you are mentioning here “splash multiplier 5 means that single liquid particle is turned into 5 splash particles”.
It’s the splash amount.
One other thing - I certainly like the idea of lower in count particles which should reflect in faster simulations, but I noticed in the FLIP optimized ship tutorial you posted in another thread, the grid count I think was 91 mil, whereas in the Classic ship tutorial there were ~50 mil cells. What I am trying to figure is what would be the minimum workable sell size/particle count that will produce dissent results.
The FLIP solver has different performance dependencies, the grid cells were increased until the simulation time reaches the old one (or rather is a bit faster).
The main factor in the flip solver is the particle count, of course the grid size is also affecting the speed, but the relation is not simple. The main difference that you will observe after some experience is that the old solver has constant performance, regardless the liquid content, but the flip solver is highly related to the liquid amount.
have been playing with the splash amount number.
cell size .3m, grid 75 mil
splash amount =20 (stolen from the ship example)
liquid 45 mil, foam 107k, splashes 1700
both foam and splash looked grainy, but splashes looked rather large particles
then I decrease foam size from 0.05 to 0.025 and increased splash amount to 50
results: foam 225k, splash 28k
splash particles still too large:
one more try: up-ed splash amount to 100
result: foam 396k, splash 54.9k
splash particles size still noticeably large (tinted red to be distinguishable)
somehow I don’t get the feeling I am on the right path to solve this issue. What would the right path be? I thought of decreasing the cell size to 0.128m (as in the ship example), but that send the grid count to half a billion, which doesn’t seem a workable solution, despite the unarguably superior solution speed of the new core.
any pointers where would be the balance between sell size/count and splash amount…?
can you run the ship scene to confirm that it produces the expected result?
your settings seem good for me, we have to be sure that you are using good build first, run the ship scene and after one night we will know.
For now it’s single machine only…
