Uploaded August 2022 | Updated September 2026, 2 weeks ago
This video may not change your life, but my #FluidX3D software will, if you do research in #CFD. On the same #GPU(s) it's 100-2000x faster than expensive commercial solvers. The entire source code is on GitHub: github.com/ProjectPhysX/FluidX3D
This 10s video shows 10s in real time with 1m/s wind speed. 476×952×476 #LBM grid (215 million voxels), 28k time steps, 23 minutes for compute+rendering on my PC with Titan Xp GPU.
How is it possible to squeeze 215 million grid points in only 12GB?
I'm using two techniques here, which together form the holy grail of lattice Boltzmann, cutting memory demand down to only 55 Bytes/node for D3Q19 LBM, or 1/6 of conventional LBM codes:
1. In-place streaming with Esoteric-Pull. This almost cuts memory demand in half and slightly increases performance due to implicit bounce-back boundaries.
Paper: doi.org/10.3390/computation10060092
2. Decoupled arithmetic precision (FP32) and memory precision (FP16): all arithmetic is done in FP32, but LBM density distribution functions in memory are compressed to FP16. This almost cuts memory demand in half and almost doubles performance, without impacting overall accuracy for most setups.
Paper: researchgate.net/publication/362275548_Accuracy_and_performance_of_the_lattice_Boltzmann_method_with_64-bit_32-bit_and_customized_16-bit_number_formats
Graphics are done directly in FluidX3D with #OpenCL, with the raw simulation data already residing in ultra-fast video memory. No volumetric data (1 frame of the velocity field is 2.5GB!) ever has to be copied to the CPU or hard drive, but only rendered 1080p frames (8MB) instead. Once on the CPU side, a copy of the frame is made in memory and a thread is detached to handle the slow .png compression, all while the simulation is already continuing.
Paper: researchgate.net/publication/360501260_Combined_scientific_CFD_simulation_and_interactive_raytracing_with_OpenCL
This video may not change your life, but my #FluidX3D software will, if you do research in #CFD. On the same #GPU(s) it's 100-2000x faster than expensive commercial solvers. The entire source code is on GitHub: github.com/ProjectPhysX/FluidX3D
This 10s video shows 10s in real time with 1m/s wind speed. 476×952×476 #LBM grid (215 million voxels), 28k time steps, 23 minutes for compute+rendering on my PC with Titan Xp GPU.
How is it possible to squeeze 215 million grid points in only 12GB?
I'm using two techniques here, which together form the holy grail of lattice Boltzmann, cutting memory demand down to only 55 Bytes/node for D3Q19 LBM, or 1/6 of conventional LBM codes:
1. In-place streaming with Esoteric-Pull. This almost cuts memory demand in half and slightly increases performance due to implicit bounce-back boundaries.
Paper: doi.org/10.3390/computation10060092
2. Decoupled arithmetic precision (FP32) and memory precision (FP16): all arithmetic is done in FP32, but LBM density distribution functions in memory are compressed to FP16. This almost cuts memory demand in half and almost doubles performance, without impacting overall accuracy for most setups.
Paper: researchgate.net/publication/362275548_Accuracy_and_performance_of_the_lattice_Boltzmann_method_with_64-bit_32-bit_and_customized_16-bit_number_formats
Graphics are done directly in FluidX3D with #OpenCL, with the raw simulation data already residing in ultra-fast video memory. No volumetric data (1 frame of the velocity field is 2.5GB!) ever has to be copied to the CPU or hard drive, but only rendered 1080p frames (8MB) instead. Once on the CPU side, a copy of the frame is made in memory and a thread is detached to handle the slow .png compression, all while the simulation is already continuing.
Paper: researchgate.net/publication/360501260_Combined_scientific_CFD_simulation_and_interactive_raytracing_with_OpenCL
![PhysX3D [GRAVITY SIMULATION] Galaxy fly-through + Blooming Effekt demo](https://i.ytimg.com/vi/WbbjvGIfqdc/mqdefault.jpg "PhysX3D [GRAVITY SIMULATION] Galaxy fly-through + Blooming Effekt demo
Recently I have improved the blooming effect in terms of performance by about two magnutudes.
Blooming is a rendering effect which makes the otherwise black space between close pixels brighter. It is a post-processing effect, meaning the completely rendered frame is being edited as an image right before being drawn on the screen.
The improvement that I implemented was instead of going through the frame pixel by pixel and checking the surrounding of each pixel, at first to scan the frame for all non-black pixels and to save their positions, so the algorithm only has to iterate through the amount of drawn pixels instead of the whole frame.
More information at: http://www.projectphysx.de
Music: Puddle of Infinity - Open Sea Morning")
![Line3D [GRAPHICS ENGINE] Dynamische Beleuchtung](https://i.ytimg.com/vi/_9i8RSxTkjA/mqdefault.jpg "Line3D [GRAPHICS ENGINE] Dynamische Beleuchtung
Ich hab noch ein bisschen an der Java-Grafikausgabe aus meinem Projekt PhysX3D weiter geschrieben. Aus einfachen Linien sind Polygone geworden, die dynamisch von einer (oder auch mehreren) frei beweglichen Lichtquellen beleuchtet werden. In diesem Beispiel berechnet die CPU auf einem Kern 25.600 Dreiecke pro Frame, und das etwa 30 Mal in der Sekunde.
Ich verwende dazu nur die elementarsten Java-Bibliotheken wie das JFrame, ein Fenster, in dem ich Linien, Text und Vielecke zeichnen kann.
Wie funktioniert die Berechnung?
Für jedes Dreieck wird der Normalenvektor berechnet. Das ist sozusagen eine Linie, die immer senkrecht auf der Fläche des Dreiecks steht. Dann wird der Normalenvektor skalar mit dem Vektor vom Mittelpunkt des Dreiecks zur Lichtquelle multipliziert. Einfach ausgedrückt vergleicht man dabei, ob der Normalenvektor zur Lichtquelle zeigt. Wenn er das tut, ist die Fläche heller, sonst dunkler. Mit dem Ergebnis der skalaren Multiplikation wird die Helligkeit des Dreiecks skaliert.
Die Lichtquelle muss nicht immer am selben Punkt sein, sie kann sich auch bewegen, wie hier im Video demonstriert.
Was ist der praktische Nutzen davon?
Das Prinzip der dynamischen Beleuchtung ist heutzutage in jedem Computerspiel und in jeder Rendering-Software zu finden, bloß werden dort die Dreiecke noch viel kleiner gemacht, damit zum Beispiel ein animiertes Gesicht nicht eckig aussieht.")
