XF's calculation engine, XFsolver, utilizes a NVIDIA graphics processing unit (GPU) when the GPU is properly installed, the drivers are up to date, and the number of GPUs was specified when the simulation was created. The Output tab in the Simulations window confirms the utilization of GPUs.

xfsolver output text

If the GPU does not initialize, confirm that the NVIDIA CUDA-capable card is properly installed.

  1. Insert the NVIDIA CUDA-capable card firmly into the motherboard.
  2. Check for a proper connection between the card and the machine's power supply unit.

This process will vary depending on the type of card. The exact methods for connecting a CUDA-capable card to the motherboard are beyond the scope of these instructions, so users should refer to the documentation provided with the card.

After ensuring a proper installation, proceed to the Specify the GPU(s) and Check Driver and Update sections below.

Specify the GPU(s)

Use XStream Acceleration option

XFsolver does not automatically search for and run on GPUs. It requires instruction on how many GPUs to utilize. When running a simulation through the user interface, select the Use XStream Acceleration option in the Simulations window. When running through the command line, provide the --use-xstream and --xstream-use-number options, similar to those shown below.

[remcom@cluster 000001]$ xfsolver --verbose --use-xstream=true --use-cpu=false --xstream-use-number=1 --num-threads=1

Additional command line options are available for utilizing GPUs. Refer to xfsolver --help

Check Driver and Update

The GPU driver must be up to date in order for XFsolver to utilize it in a simulation, so users should check the driver version before updating the driver. To accomplish this within the user interface, click the Show Detailed Information in the Simulations window.

Alternatively, the following command line option can be used:

$ xfsolver --xstream-list-devices

Review the message's content and locate the two CUDA version numbers, similar to the example below:

CUDA runtime version: 4010
CUDA driver version: 4020

If the driver version is less than the runtime version, update the driver using the following steps:

  1. Download the card's newest driver from the NVIDIA website and save it to the disk.
  2. Install the driver by following the steps in the installation wizard.
Once the driver update has been installed, confirm that the driver version is greater than or equal to the runtime version using the process above. If the driver version is less than the runtime version, ensure that the latest available GPU driver from the manufacturer was installed. If an error message still occurs when initializing a simulation, look at XFsolver's error message under the Output tab for troubleshooting clues or contact Remcom support.