Liquid simulations on cloud-server

hi!

I will soon have to do a lot of fluid simulations in blender (2.91), probably more than I can do on my current computer in terms of how long I would have to wait for the baking of the fluid simulation. I’m expecting simulations worth 100s of hours if I did them on my machine. (Still not quite sure if I go with mantaflow or flip fluids addon, but for this question it shouldn’t make much difference).

I’m considering externalizing the simulation to a virtual machine (VM) in the cloud. And because the cache files are so big I’m expecting also to render on the same machine because the 100s of GB of cache files won’t transfer quick enough. I’m having some uncertainties as to the details of this approach. Here are some more or less general questions, if you have thoughts on any of them I would highly appreciate your answer!

  1. What is the bottleneck when doing physics simulations / liquid simulations? What are the most important specs? Is this more a RAM / CPU thing or is GPU an issue here too? Is disk access speed an issue?

  2. What are the requirements the cloud service should fullfill? Do I need an OS with GUI to start the simulations or is there a way to trigger simulation in a file (one that I prepare on my machine) from commandline, on a “headless” server?

  3. are there any services specifically for that purpose? Like a render-farm but then a sim-farm?

as I said before, ANY thoughts or experiences might be helpful. thank you!

kind regards, oyster

check. it. out.

this article covers it really well and in good detail. with step-by-step guides and even some convenient python code snippets for rendering files automatically on the cloud-vm.

for me that’s it!

only question that remains:
what are the good specs for the simulation part? do I need killer CPU with lots of cores or as much RAM as I can get? or both? what’s the bottleneck usually?

Hello, I have also been curious of how I could run a blender fluid simulation or smoke simulation over a network and as far as I know unfortunately I couldn’t find any renderfarms that support baking fluid simulations over multiple pc’s. However I believe something like this may be possible using a cloud service such as aws or google cloud but I still am not sure if the simulation could be spread across multiple pc’s since a fluid or smoke simulation requires temporal data from the previous frames.

But if you simulate locally many cloud providers allow you to load data onto a custom hard drive they send to you and then you could load your cache simulated locally onto the drive and send it back to the cloud provider but this will probably cost extra but may be much faster than uploading the files through the internet to the cloud depending on internet speed and cache size.

For simulation it’s all cpu and ram, you want to use a very high end cpu with lots of ram at least 32 gb preferably over 64 gb for large simulations you also need high speed storage using a solid state drive instead of hard drive to speed up caching.

Even with high end cpu mantaflow still probably wont take advantage of all cores but getting a high end cpu will improve simulation time. In the future if simulation could be done on gpu it would probably be much faster possibly even near real time for high end gpu’s but so far mantaflow only supports cpu.