GPU Simulation

Fluid Simulation needs quite a bit of processing power. Mostly because there is a huge amount of data to be pushed around. This makes memory bandwidth the most important factor for simulation speed. Today's fastest memory interfaces are found in GPUs - about 10 times faster than those of CPUs. Coupled with the appropriate amount of parallel compute power, GPUs are the ideal type of processor for fluid simulation.

TurbulenceFD makes use of GPUs for it's simulation pipeline. Unlike with some GPU accelerated tools, this is not just a stripped down version of the CPU pipeline. All features are supported at the same quality. In fact, you can switch between CPU and GPU simulation on-the-fly (see Simulation Window). This is also what TurbulenceFD will do automatically, should it run out of GPU memory. It will then continue the simulation on the CPU.

Supported GPUs

Nvidia GPUs with Compute Capability 2.0 or newer, listed at http://developer.nvidia.com/cuda-gpus

While TurbulenceFD technically works with less than 1GB of GPU memory, a GPU with 4GB or more memory is highly recommended.

Please make sure to use the latest driver for your graphics card.

Hardware setup tips
  • When choosing a GPU, prefer the one with the most per-GPU memory (note that per-GPU memory for Dual-GPU boards is usually half of the advertised amount)

  • Even prefer a slower GPU if it has more memory than an alternative faster GPU (see this post for the rationale behind this advice)

  • Ideally use two GPUs: one (possibly smaller one) as primary display GPU and one as a secondary GPU for simulation only.

  • When your system has only one GPU and you run large simulations, disable the viewport preview to speed up the simulation.

  • If you have a supported GPU but don't have anything to select but "Use CPUs", update your driver (see Supported GPUs above).

  • The larger the resolution the better the speedup (GPU vs. CPU) will be. At very low resolutions, the GPU sim may not be much faster.

  • Also keep in mind the general performance guidelines.