For now a lot of work with my studies as I’m finishing the bachelor’s degree, so less time for programming BUT more time for doing the planning stage so hopefully when I have more time the actual writing of the code will go significantly faster.
It’s not that I don’t have any new updates but they are just very rough as I implemented the first version of narrow-band like simulation which in the final form will be faster & less memory hungry but for now it’s just faster and much MORE memory hungry. But it’s like 3-4x faster at average and mostly even more but it’s still a bit buggy as it’s not based on any paper but on my own idea as I wanted to do it a bit better. I will give some samples soon but first I need to make it more visually stable
Okey, so as it’s now more stable than me than I’ve decided to do a little video with an explanation what I did and how much it’s faster (spoiler alert!!!) (from ~400ms/frame to ~120ms/frame) (!!!spoiler alert) for the level of detail of 70,000,000 particles.
So not great not terrible. I’ll be working on that more so hopefuly it’ll be faster in the future. And for the video link it’s below:
Your progression for the software is really great I appreciate that but till around what time you are planning to integrate with blender mantaflow or apply the smoke simulator or the liquid simulator as a addon for blender as your software is in realtime which is very efficient.Also thanks for making such wonderful softwares they are very useful and handy
Hi michal1000w, i thought you might want to know, that Sebastian Barschkis, the maintainer of mantaflow, doesn`t work for blender institute anymore. They might be looking for someone to takle that ??? I remember you wrote that you might want to add it to blender. Maybe you get a fund or something ???
Hi thanks for telling me. Yes, I would love to add it to blender and it would be an honor if it would ever be accepted (not mentioning that working at blender would be a dream-job). I’m afraid however that I may be a little too stupid for that as my work is fast but still far from being production ready . If I get a fund however I could work full time on both simulators instead of trying to make money as an AI engineer during the day, then afterwork doing the masters degree, in any free time trying to have a life aaaand spending nights on the simulator. As for now it’s just too much for me and that makes the development much much slower and from some time quite unpleasent.
But thanks for beliving in the project as it keeps me going. Without that I would give up a long ago as more and more I feel that nobody really cares
Have a nice day everysoul reading this and I hope You feel good in this time of darkness
Hi, i can imagine this is tough. I can`t program myself just a little scripting. Pablo said on yesterdays Blender Today they are looking for a new maintainer for mantaflow. Maybe you can apply. That would solve some problems. I am just desperate, because there is only one realtime simulator and that is very expensive. I wonder why you are the only one to do a free one, since there is the opportunity to do it on gpu now. I understand its very complicated stuff, but there must be a demand for that. I was baffled to hear that Sebastian quit after introducing a first draft for a realtime mantaflow.
That looks promising, especially the multi-material mix and it’s way more polished than my simulators. The only thing that I see mine doing better are the baking times. As test below shows a quite fast baking
6.8M points
300 frames (at 24fps)
40 minutes bake time (CUDA)
however my SPH solver does 80M points in about 26fps in realtime. But overall I’ll have a closer look at Taichi
Ahh. Cool. Any chance you both could put your brains together and collab?
Your work is out of my skill level ATM, otherwise I’d totally help out. ATM I’m working on basic stuff like prototyping and implementing grouping. lol
I mean if they would like to then I’m open. My goal is nothing more than to do something useful for anyone who like simulations as I’m myself in the situation where I like doing vfx etc. but I have just no money for Emb** *** and Hou**** and Mantaflow is great but not the fastest on the huge scale.
So one day I just thought why not to do it myself (still knowing that I’m not exactly bright enough for that). But the prototypes are working so with some help it should be initially ready in short time (or with enough money to drop current job and sit only on the sims because now I have too little time for huge steps) (and not production ready but like… hobby ready).
And sorry for that long monolog Hope it makes sense
For sure I hear ye. I hope that they’ll be able to help you. I’ll work on banging out this group feature as fast as I can and a couple of other high priority stuff on my plate and see if I can help you in whatever way I can. Perhaps with a good introduction from you on your code and how it works I can be of service
Thanks a lot
I’ll try to do an introduction to my code when I’ll make it more readable, as for now it’s a total mess. Like I have to pray to the Sphagetti Monster when I’m sitting in front of it (for the fluid sim at least as the smoke sim is much more clear). And I know it’s a really bad practise but with so little time most often I just write ideas as fast as I can and work on them until they are functioning properly and after that sprint I make a little refactor to make it not only work good but also look good. But when I have a lot of time for a project I usually try to make it as clear as possible from the beginning.
Haha. For sure! I can help you with stuff like formatting and organization / commenting no problem if I know what’s up.
Digging into the blender code base the last month and to trying and figure out how it works has been a nightmare. Stuff in there is all over the place with almost no comments. Hahaha. So you’re not alone by any means.
Okey. Sooooo long time no updates. The reason why is that I’m working on a new Smoke solver (and new iteration of the renderer), as the old one was fast (about 30fps for a huge domain), however it has some problems like:
uses a lot of memory, so on a 8GB card the max domain was about 500^3
the domain size was huge but the actual amount of detail was quite low
above 500^3 it gets really slow compared to the smaller scale
a lot of internal limitations
That’s why I decided to rethink the entire concept of how to do this (again). And today I’ve finished the first draft. To be honest I’m not so happy about how it turned out, as the simulation times feels much slower than the old ones when it comes to the relation of domain size → sim time, however on the contrary it looks A LOT better in terms of details and overall smoke behaviour. What’s more it uses over 3x less memory and the scaling of sim time vs resolution is a bit better when it comes to a huge scale. So it kinda is better but still there is a huge field for improvements in my opinion.
In summary it is fast, but I wouldn’t call it realtime. 968x624x408 domain is simulating at about 0.85sec /frame. So not great, not terrible but until it is in frames per second it won’t be called anywhere near “ready”
I’m really hopefull that I could do something useful
thanks for telling me. Yes, I would love to add it to blender and it would be an honor if it would ever be accepted (not mentioning that working at blender would be a dream-job). I’m afraid however that I may be a little too stupid for that as my work is fast but still far from being production ready 
. If I get a fund however I could work full time on both simulators instead of trying to make money as an AI engineer during the day, then afterwork doing the masters degree, in any free time trying to have a life aaaand spending nights on the simulator. As for now it’s just too much for me and that makes the development much much slower and from some time quite unpleasent.
