Simulating fluids, fire, and smoke in real-time

Original Article ↗ Hacker News Discussion ↗

Real-time simulations of fluids, fire, and smoke are now feasible on consumer hardware, but most visual effects rely on heavily simplified physics that prioritize plausibility and artistic control over physical accuracy. Commenters compare game and movie techniques (like vorticity confinement, curl noise, and tools such as EmberGen and Houdini) with industrial and scientific CFD, noting that the latter still requires massive compute, careful numerics, and long runtimes. The thread also touches on GPU vs CPU roles, usability and performance issues in browsers, and career paths for software engineers interested in high-performance fluid simulation.

Real-time fluid effects vs “true” CFD

  • Commenters note that real-time fluids in games have existed for years on modest hardware.
  • These real-time systems heavily simplify physics and aim to “look right” rather than solve Navier–Stokes accurately.
  • Industrial/high‑Reynolds‑number CFD instead often adds artificial dissipation for stability and is far slower.

Visual plausibility, cheating, and game design

  • In graphics, internal consistency and plausible reactions (water vs. shoreline, obstacles) matter more than strict realism.
  • Many effects are procedural; when they don’t interact with the world, they feel fake despite looking pretty.
  • “Cheating” is seen as necessary and even desirable; the target is plausibility and fun, not physical truth.

Noise-based turbulence and vorticity techniques

  • Several comments discuss vorticity confinement and curl/curl-noise (“curly noise”) as ways to add detailed swirling motion to coarse sims.
  • Curl noise fields are divergence‑free and can create fluid‑like motion without time-stepped simulation, though individual evaluations can be expensive.
  • There’s some disappointment that original vorticity-confinement work is not always credited.

Hardware, browsers, and performance

  • The article’s interactive sims run smoothly for some and are nearly unusable for others.
  • Reports vary by OS, browser, GPU, and whether hardware acceleration is enabled; Safari on Apple Silicon tends to fare well, Chrome on some Intel laptops less so.
  • Some suggest simulations should be user‑toggled to avoid pegging CPUs/GPUs.

GPU workloads and iGPU use

  • Discussion covers how modern GPUs can run multiple compute tasks per frame (fluids, physics, AI, rendering) within a frame-time budget.
  • Others point out that many games are CPU‑bound at high FPS and that using an iGPU alongside a dGPU is rare, partly due to software complexity and thermal limits.
  • There’s debate on how far GPGPU will erode the role of traditional CPUs.

Learning and working in CFD

  • Multiple requests for entry paths lead to suggestions: start with basic conservation laws, shallow‑water equations, numerical stability concepts.
  • Recommended skills include MPI, some Fortran, and general HPC knowledge.
  • CFD jobs are often in aerospace, energy, weather, and specialized software vendors; competition can be high, timing matters.

Tools and learning resources

  • Mentioned tools: EmberGen (real-time fire/smoke), Houdini (node-based VFX), various GPU/CPU fluid tutorials, and course notes/texts on graphics-oriented fluid simulation.