Beach Waves questions

[elyptic]

Hi, I was checking out the included Beach Waves scene and decided to run the simulation (as is) and found the simulation time to be a day and a half. Is that normal for this scene? I killed it after a couple of hours on frame 42 and rendered it, still didn't see any foam. When I study the particles in the preview, I saw there were a few foam particles out at the back but nothing like I saw in the preview video. I figured you must have started the render at some point 50+ frames in, so I left it overnight to simulate and checked it in the morning. Still only 130+ frames had been simulated, but I rendered it and finally saw the foam in the video.

In an attempt to speed this sim up, I changed the grid resolution to 10cm from 4.7 in the demo, disabled the output of several of the particles(ID, age, RGB), and the sim took still took 3 and a half hours, but the results look like the water is now a blob of gel. I assume it's got to do with those extra particle exports so I'll re-enable them and try again.

Is simulation really supposed to take this long to do? Watching the promo for FD 3.0 I saw the Flip Simulator and it looked like it was near real time. That promo video got me excited enough to put the money together and buy it... but I can't imagine making changes to these scenes and waiting a day or two to see the results.

I think the mention of GPU acceleration has already been shot down so I fear the only hope I have is for those new RYZEN CPUs coming out soon, I'm running FX-8350 with 32GB of RAM on a Samsung Pro 850 SSD.

[Svetlin.Nikolov]

Hey,

The video is sped up - I guess our 3D guys have to be more careful to put an indication where this happens. The scene is quite big, but the beach geometry is not okay, being opened, so I hope we can straighten this out and reduce the sim times a lot. With lower resolutions it would not look very good, so it's up to us to optimize it. GPUs could not benefit the thing too much, because half the time of the simulation goes into interacting with the scene, which is single threaded, and saving and loading the cache files. You can speed up the sim by turning off everything in the preview - the mesh, all the grid channels, the particles and eventually the GPU preview. We did some improvements to the cache loading speed in the nightlies, but there is a lot more that can be improved. The simulation bottleneck, on the other hand, is usually the bus speed, so you should be very careful when selecting a MB - in the office here we have a couple of monster CPUs with crappy motherboards and they simulate more slowly than my older i7.

I hope we'll have the scene reworked soon and it would be up on the docs site.

Cheers!

[Zakkorn]

The simulation bottleneck, on the other hand, is usually the bus speed, so you should be very careful when selecting a MB - in the office here we have a couple of monster CPUs with crappy motherboards and they simulate more slowly than my older i7.
Cheers!

Just a question on this particular phrase. I apologize as it is not very appropriate for the this thread.
How can we choose the m/bs?
What someone must check / look to be sure that simulations like phoenixfd will use all the power of the cpus and nothing will be wasted?

Because all the m/bs of the same line (x99 chipset for example) give the same bus speed.

I am asking this because with my dual Xeon I did not have many options about m/bs
And because I am thinking to buy a new machine dedicated to phoenixfd.

Regards,
Zach

[Svetlin.Nikolov]

Yup, well that's about it - you should get a motherboard with as high bus (FSB) speed as possible and if it's going to be a simulation machine, for now dual-processor or NUMAs are not recommended since they waste time communicating with one another and sometimes using just a single NUMA node for simulation is faster than using all. Since you already have an SSD, this point is already covered

[flipbook]

Is there a way to disable a NUMA node for the sim? How is the thread limit option done, does it limit threads from one CPU first or randomly?

Thanks

[Svetlin.Nikolov]

Sure, the NUMA nodes are in the Simulation rollout - ctrl+click to enable or disable one of them, if there are any. The thread limit takes the threads in the order Windows returns them - should be the same order you see them in the Task Manager, but I gotta check for sure

[waaazoo]

For simulation, would it be better for a CPU to have less cores but higher speed (4 x 4.5Ghz), or more cores at lower speed (10 x 2.3Ghz), or does it not affect the simulation render speed?

[Fragster]

Hey guys,

so I watched your nicely done "Beach Waves" tutorial and re-created the scene from scratch.
At first glance the result is actually looking good but there are still a few issues.

1. Is there a way to get a smooth or clean transition between the grid content and the adjacent ocean? (simulator and displacement now share the same map)
2. The collision of the water is not continuously accurate like there is an offset or sth. (00:16) - due to SPF (it was 10)? or grid resolution (cell size was 2,6cm)?
3. There seem to be some gridlike artifacts on the ocean surface which are visible especially in bright reflection (00:14) - should I use Mesh smooth options?
4. Mist creation - Do I need to add a plain force to activate that?
5. Wetmap - great feature! It would be great to achieve individual drying times for different mesh objects (wet rocks should stay glossy while sand is already dry)?



Thanks!

[Svetlin.Nikolov]

Hey, sorry for the late reply!

Make sure you try out the same setup with a very latest nightly - since the official 3.0 there are many fixes and improvements related to collisions, mist birth and ocean meshing.

The mist can be created only from the 'Splash to mist' param - do you have it above zero?

Thanks very much for the suggestion about per-object wetmap drying times! Also, if you still have the artifacts on the border between the sim and ocean, ping me

Also, just in case, check if your ground geometry is a closed mesh and is not just a plane, or it might cause an entire different simulation running underground, which might mess with the behavior of the visible sim.

Cheers!