![fxphd fluid dynamics in realflow 2013 fxphd fluid dynamics in realflow 2013](https://1.bp.blogspot.com/-8WQtcOBVop4/WeY7g3ZS0zI/AAAAAAAAQzM/q5EkvkKQqug5KQQ2LQoKNdVLLL1rr3-jgCLcBGAs/s1600/Mastering%2BDestruction%2Bin%2BHoudini%2Bwith%2BKeith%2BKamholz%2B02.jpg)
Plugins also fallback automatically with fluids. GPU and CPU plugins can be mixed, but isn't recommended due to an initial performance overhead (we have to read back more buffers). All plugins have a GPU implementation included and use the same public API that you can use to create your own GPU fluid plugins.You can supply a fallback particle count and fallback emission rate to automatically adjust particle counts accordingly as well. Fallback is automatic - if GPU acceleration is unavailable (non-DX11 platforms at the moment), fluid simulation will run on the CPU.GPU acceleration is just a checkbox, and can be switched on and off at runtime - even while a fluid is simulating.Fluid simulation is a really good fit for GPU computing, and I've wanted to get Fluvio running on the GPU for a while now. I'm super excited about this, because the performance difference is like night and day, especially compared to 2.x. I also plan on implementing a geometry shader on DX11 that will allow rendering of fluid without any reading back of data - this should easily allow 100k+ particles on a desktop. GPU accelerated fluids are currently only supported on DX11, but I have plans to get them working on OpenCL as well.
![fxphd fluid dynamics in realflow 2013 fxphd fluid dynamics in realflow 2013](http://3.bp.blogspot.com/-eBSagEt5QMo/UrB8_1Nd5KI/AAAAAAAAAqQ/uVclbAP64fM/s1600/Isotropix+Clarisse+iFX+1.6.jpg)
![fxphd fluid dynamics in realflow 2013 fxphd fluid dynamics in realflow 2013](https://1.bp.blogspot.com/-jN7UGUBrcFM/Wal7-ObaftI/AAAAAAAAFNA/lXgK32evpKU_jhC9Yq69zNblrToMEFQ4ACLcBGAs/s1600/Maya%2B2018%2BUpdate%2BReview%2B%2BUV%2BEditing%2BImprovements%2B00.jpg)
With GPU fluids, my desktop is able to simulate up to around 12,000 particles before any FPS drops, and that's still while using Shuriken to render the particles (so the bottleneck is on reading data back to the CPU). It will be available with the first beta as well. I have something very awesome to share with you guys - I've had something in the works for a while and it's only really been feasible once the groundwork with the new solver was set. So I think the beta is finally at a state where it can get its first public release - there are a few things that have to be done at this point (compatibility checking with 4.5/4.6, mainly).