Running a simulation job using GPGPU acceleration can save time when you have a relatively
high-end graphics card. However, GPU acceleration has the following requirements and
limitations for fluid simulations:
- NVIDIA graphics cards with a compute capability of 4 or higher are required.
- You can run the simulation using multiple GPU cores and specify a different number of
cores for each flow analysis case in the simulation. The number of CPU cores you specify
must be a multiple of the number of GPU cores.
- Enable the graphics processor for computing in the execution preferences. For more
information, see Common Submission.
- Specify the coupled solver in the fluid physics. If you run a simulation with the
segregated solver selected, the app runs the job using regular execution.
- In the equation solver settings for steady-state flow steps and transient flow steps, two
linear solver options are supported for the coupled solver: AMG-BICGStab and ILU-FMGRES.
AMG-BICGStab provides better performance when you run simulations with GPU
acceleration.
- When you run the simulation job, use embedded licensing and run the job locally. For more
information, see Running a Simulation from an App.
- When you select the GPGPU solver, the app disables the error estimator approach for
stopping criteria and uses only the traditional residual-based criteria.
- The problem size that the GPU can solve depends on the memory capacity of the GPU card.