CST2013: Acceleration

Simulation: Solver Start Simulation Acceleration
Simulation: Solver Start Simulation Acceleration
Simulation: Solver Start Simulation Acceleration
Simulation: Solver Start Simulation Acceleration
Simulation: Solver Start Simulation Acceleration
Simulation: Solver Start Simulation Acceleration

This dialog box provides full control over the different acceleration features used to increase the overall simulation performance.

CPU and GPU acceleration frame

Multithreading (CPU) up to [ or Multithreading (CPU) ]: Check this field to take advantage of the functionality of a multi-processor machine. The calculation will be split into corresponding parts and will run parallel in different threads to save simulation time. You may insert the desired number of threads in the provided field.

Please note: Some processors announce more cores to the system than they actually have (virtual cores). This feature is referred to as "Hyperthreading" for Intel processors. Using these virtual cores for CST solver processes deteriorates the performance. Therefore CST solvers will only use the physical processor cores but not the virtual ones.  

GPU acceleration up to: If this check button is enabled, GPU Computing is activated. Please note, that this setting has only an effect if your computer is equipped with GPU hardware supported by CST.

Please refer to the GPU Computing page for further information.

Notice: In case of MPI Computing or Distributed Computing the setting for the number of CPU cores and for the number of GPUs is meant per MPI node or per DC Solver Server, respectively.

Distributed Computing (DC) frame

Parameter sweep / Optimization up to: Activate this check button to use the Distributed Computing for parameter sweeps and optimizer runs to perform simulation runs with different parameter sets in parallel on the available DC Solver Servers.

Distribute excitation calculation up to: (Not available for all solvers) Activate this check button to enable the Distributed Computing (DC) facility. This offers the possibility to distribute solver runs for different ports of a multi excitation calculation to different DC Solver Servers and solve them in parallel.

Distribute frequency calculation up to: (Not available for all solvers) Activate this check box to distribute several frequency samples on different DC Solver Servers ("Distribute frequency samples") to speed up the S-parameter calculation. The distribution of frequency samples is only available for the general purpose solver. See the Frequency Domain Solver Overview for details about the distribution of frequency samples. All fixed frequency samples, such as monitor frequencies, equidistant and single frequency samples can be submitted to the DC job queue at once. However, if the automatic frequency sampling is active, the number next to "Distribute frequency samples" specifies how many jobs may be submitted at once to the job queue. Increase this number to use more DC Solver Servers in parallel for automatic frequency sampling. Please note that it does not affect equidistant frequency samples, monitor frequencies and other fixed frequencies, which are submitted to the job queue at once.

Remote calculation: (Not available for all solvers) Activate this check box to run a simulation remotely on a DC Solver Server.

DC matrix calculation: (Not available for all solvers) This check button controls whether the matrix calculation is done remotely or on the machine where the frontend is running. In any case the computation is only performed once before the transient solver is started.

DC Properties...: Use this button to edit the current settings for the Distributed Computing

Use only servers with more than x GB memory: Enable this check box to specify a lower limit for the required memory of the distributed jobs. The specified DC Main Controller will use this value to select appropriate DC Solver Servers where the requested amount of memory is currently available. Please note that the available memory is checked in intervals. This means that when the job is send to the DC Solver Server the available memory might differ from the one which is shown in the DC Main Controller's Connected Clients tab.

MPI Computing frame

MPI Computing on ... nodes:  (Not available for all solvers) Activating this check button allows the solver to run one large simulation in parallel on a homogeneous cluster of machines using MPI (Message Passing Interface) communication. You can select the cluster machines and all corresponding MPI settings in the MPI Cluster dialog box by pressing the MPI Properties button.

Please find more detailed information about MPI Computing on the MPI Simulation Overview page.

Token usage frame

Required tokens for this simulation shows the number of acceleration tokens needed to perform the simulation with the settings made in the other sections of this dialog box. This number is automatically updated when changing the above acceleration settings.

Tokens currently available displays the number of currently available acceleration tokens. The total number of acceleration tokens in the license is shown in brackets.

See also

Solver Overview,  GPU Computing, Distributed Computing, MPI Simulation Overview, Optimizer, Parameter Sweep