Hiya,
Just some quick benchmarks for flip simulations.
AMD 3990X 64 core
AMD 3975WX 32 core Pro.
Benchmark: Flip Sim
Houdini Benchmark Flip scene
Solver. Save to Disk
3975WX 1min17 - 3.96-4GHz SMT ON. 32 cores.
3990X 1min17s 3200MHZ RAM SMT Off. Cpu usage was about 75-80% and speed around 3.6GHz. 64 cores
3990X 1min55s 2133MHZ RAM SMT On. This uses slower memory settings.
3990X 1min43s 3200MHZ RAM SMT On
Comment. 32 Core Pro and 64 core has identical speed for the solver.
This is probably because this step is not that well multithreaded, and the Pro has better memory usage and also a bit higher clock speed. You can also see that disabling SMT (hyperthreadding) increase the sim speed.
Geometry building
3990X 2min25 3200Mhz SMT OFF. This step is at 100% so we can see its a bit faster than the Pro
3990X 2min56s. This used 100% of the cores at 3.5GHz
3975WX 2min43 - pretty much fully core utilisation at 3.8-3.9GHz
Windows 10. Air cooled systems.
I could maybe upload the file for others to test. Its not a very heavy flip, so worth testing higher res of it.
What I ideally would like is that we get 64 core AMD with more modern Zen 4 or similar. Zen 2 which is what 3990x is using is a bit old. In other programs like Terragen the 64 had a bigger advantage than this, about 50% faster than the 32 core.
Found 117 posts.
Search results Show results as topic list.
Technical Discussion » Flip Benchmarks - AMD Threadripper 3990X vs 3975WX Pro
- AndreasOberg
- 117 posts
- Offline
Technical Discussion » Flip Cache: How to save MULTIPLE cache nodes at once
- AndreasOberg
- 117 posts
- Offline
Hiya
So I have a flip where I have several steps of caching
1. I cache the solver into a file cache
2. I cache the Particle Fluid Surface
Is it possible to trigger these caches with 1 click. Right now I have to cache the first to disk, then wait, and then I can cache the second step.
Even better would be if I could also trigger a render so basically cache 1 and 2 and then render the sequence.
Cheers.
So I have a flip where I have several steps of caching
1. I cache the solver into a file cache
2. I cache the Particle Fluid Surface
Is it possible to trigger these caches with 1 click. Right now I have to cache the first to disk, then wait, and then I can cache the second step.
Even better would be if I could also trigger a render so basically cache 1 and 2 and then render the sequence.
Cheers.
Technical Discussion » SOP Flip: Why does some particle suddenly die mid air?
- AndreasOberg
- 117 posts
- Offline
Thanks for the help Tomas!
Your suggestion fixed almost all of the problems. Pity reseeding seemed useful to thin out areas that were loosing particles, but they did not add that much in my case.
Thanks again, very helpful.
Put phew, I wish the simulations were a bit faster. I have a 64 core machine but still so much waiting for the sims.
Cheers,
Andreas
Your suggestion fixed almost all of the problems. Pity reseeding seemed useful to thin out areas that were loosing particles, but they did not add that much in my case.
Thanks again, very helpful.
Put phew, I wish the simulations were a bit faster. I have a 64 core machine but still so much waiting for the sims.
Cheers,
Andreas
Technical Discussion » SOP Flip: Why does some particle suddenly die mid air?
- AndreasOberg
- 117 posts
- Offline
Thanks Tomas. That fixed almost all. There are still a few particles that I see are being killed. Do you know why this happens? There must be some logic to deleting them.
Technical Discussion » Xeon 8375C*2 and 6132*2 Only get 70% in flip sim .
- AndreasOberg
- 117 posts
- Offline
Technical Discussion » SOP Flip: Why does some particle suddenly die mid air?
- AndreasOberg
- 117 posts
- Offline
One problem I have with my splash sims is that after a while some of the particles just die. Usually if they are flying quite separately.
In the flip solver I do not see any settings for deleting particles as they get spread too thin, but this is what seems to happen.
Any ideas if this can be fixed? Ideally I do not want any particles to die as they fly out.
In the flip solver I do not see any settings for deleting particles as they get spread too thin, but this is what seems to happen.
Any ideas if this can be fixed? Ideally I do not want any particles to die as they fly out.
Technical Discussion » How to render pyro and flip in the same SOP?
- AndreasOberg
- 117 posts
- Offline
Thanks. I tried both of the ways. I think it felt like a better way to have 2 geom nodes. That way it is easier to filter what lights will light what. In this case I think I want a dedicated light on the mist and one on the liquid.
Technical Discussion » How to render pyro and flip in the same SOP?
- AndreasOberg
- 117 posts
- Offline
Basic Question. I have now have a SOP flip and a SOP pyro in the same geometry node.
I'm just adding mist to the geometry liquid.
How do I render them with different materials?
Right now I have them in a merge node as the last step.
I'm just adding mist to the geometry liquid.
How do I render them with different materials?
Right now I have them in a merge node as the last step.
Technical Discussion » How to create a pyro volume with velocity from geometry?
- AndreasOberg
- 117 posts
- Offline
CYTEThanks that seems to work! How is the transfter done. I was just thinking I start with just having points on the surface of an object, but then the transfer has to figure out what to do inside the geom.
You can Attribute transfer the velocity from the mesh onto the emitter points.
Cheers
CYTE
Technical Discussion » How to create a pyro volume with velocity from geometry?
- AndreasOberg
- 117 posts
- Offline
Trying the new Pyro SOP here.
I have a sphere that I'm animating with Mountain.
I use Trail to create velocity for every point. Now I want to throw it into a pyro so the smoke inherites the velocity.
I tried with Pyro Source but it refuses to take the velocity attribute it just overwrites with the default one, at least when I fill points inside the mesh.
if I use Volume rasterizeattributes then I get the velocity but only on the surface.
Basically for an impact explosion I want to be able to animate the mesh, then fill the mesh with smoke that has velocity moving along with it.
Any tips would be appreciated.
Andreas
I have a sphere that I'm animating with Mountain.
I use Trail to create velocity for every point. Now I want to throw it into a pyro so the smoke inherites the velocity.
I tried with Pyro Source but it refuses to take the velocity attribute it just overwrites with the default one, at least when I fill points inside the mesh.
if I use Volume rasterizeattributes then I get the velocity but only on the surface.
Basically for an impact explosion I want to be able to animate the mesh, then fill the mesh with smoke that has velocity moving along with it.
Any tips would be appreciated.
Andreas
Technical Discussion » How to construct foam shader?
- AndreasOberg
- 117 posts
- Offline
Hi,
Anyone has tips for how to make a good foam shader for water splashes. I'm using Redshift to render currently but could use Karma as well.
I looked at the mantra whitewater shader, but I do not see that much going on. What I'm considering is using say vorticity to control where there is foam and make it white and opaque and then having areas where we have more transparent liquid.
But I'm not sure vorticity will actually give the right places. Another idea would be to use the age with some randomness to animate foam more. I'm not overly keen on using Mantra to render with.
Anyone has tips for how to make a good foam shader for water splashes. I'm using Redshift to render currently but could use Karma as well.
I looked at the mantra whitewater shader, but I do not see that much going on. What I'm considering is using say vorticity to control where there is foam and make it white and opaque and then having areas where we have more transparent liquid.
But I'm not sure vorticity will actually give the right places. Another idea would be to use the age with some randomness to animate foam more. I'm not overly keen on using Mantra to render with.
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
BenyeeOk, have not really tried that will check. Thanks!AndreasObergtry obj/geo1/fliptank1:Grid Scale with 1.5?hope it helps
looks a bit like a drop on a window slowly falling down. Any idea what is going on?
I find that with min 1 and max 3 I do not get the stuck drops. To me its very odd though, how on earth can they detect a collision in mid air when there is nothing there even at min 1 and max 1 steps.
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
If I use the checkpoints the whole simulation just disappears. No matter what I do I just cannot see anything. Anyone else has better luck.
Another bug?
Another bug?
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
animatrix_
Have you tried playing with FLIP Min/Max Substeps?
Also I wouldn't compare individual velocity components like that, unless you want to check for movement on specific axes, but even then a tolerance would be better:float tol = 0.0000001; if ( length2 ( @v ) < tol ) removepoint ( 0, @ptnum );
A problem I found with this version is that this will delete the particles as they turn around if they fly straight up. That is why I changed so I looked mostly at x and z.
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
HM_2020
I still say this is some bug, never really had this issue in DOP version of FLIP before. (except as I said, sometimes if you key activation in some sources, it would make strange things like this happen.)
Can you rebuild it in 18.5 or just not a SOP in 19 and see?
It could be that the keyed activation created this problem. I have a new version where no longer key it, I just scale the collision mesh to 0 instead. Maybe that fixed it.
I will try the DOP Flips as well, but they are quite difference so will probably work there!
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
Here are the benchmarks
Global Substeps 1 Min:1 Max: 1 3min 15sec
Global Substeps 1 Min:1 Max: 3 8min 33sec surprised that this is shorter
Global Substeps 1 Min:1 Max: 2 18min 31sec
Global Substeps 2 Min:1 Max: 2 34min 40sec
To me it seems like global really increases time. After that it seems mostly that the size of the shape depends the time. The max 3 is fast, but its because it creates a quite small splash.
Global Substeps 1 Min:1 Max: 1 3min 15sec
Global Substeps 1 Min:1 Max: 3 8min 33sec surprised that this is shorter
Global Substeps 1 Min:1 Max: 2 18min 31sec
Global Substeps 2 Min:1 Max: 2 34min 40sec
To me it seems like global really increases time. After that it seems mostly that the size of the shape depends the time. The max 3 is fast, but its because it creates a quite small splash.
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
animatrix_Thanks! Just doing some benchmarks.
Yes FLIP substeps is cheaper than global DOP substeps.
Would you use the previous DOP for flip now or the new SOP?
To me the largest benefit with the older DOP is that it can do whitewater, but you maybe do yours manually?
Do you have any tip for how to do a white water shader? I got a pretty good material for transparent water, but the foam part I'm not sure what the best way is to tackle. I looked at the default whitewater but it does not seem to do anything special in the material what I can see.
Andreas
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
animatrix_
Have you tried playing with FLIP Min/Max Substeps?
Also I wouldn't compare individual velocity components like that, unless you want to check for movement on specific axes, but even then a tolerance would be better:float tol = 0.0000001; if ( length2 ( @v ) < tol ) removepoint ( 0, @ptnum );
I tested now to set substeps min 1 and max3. That got rid of the stuck particles. Not sure if its faster than having 2, maybe. I do find it a bit odd the the particles would just stop in mid air and then slowly fall down.
The reason I compared the velocities is I got very odd characteristics. They would be completely 0.0000 in x and z axis but they would have a bit of -y velocity.
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
Of course having 2 substeps really increases simulation time so its not exactly the best solution. The manual they say its very rare that you need to increase this from one. So would be nice to find another way if possible.
Technical Discussion » Flip Splash leaves little drops almost static in the air
- AndreasOberg
- 117 posts
- Offline
I managed to fix it by increasing the Simulation Global Substeps to 2. Takes longer to calculate of course but at least I do not get odd fake collisions in mid air.
This is the wrangle I used before btw
if(v@v.x==0.0f || v@v.y==0.0f) removepoint(0,@ptnum);
This is the wrangle I used before btw
if(v@v.x==0.0f || v@v.y==0.0f) removepoint(0,@ptnum);
-
- Quick Links