Supported GPU boards

In order to run FRED^MPC, an NVIDIA GPU board is required.  Although it is subject to change in future versions of FRED^MPC, the GPU board must have NVIDIA compute capability greater than 6.0.  The graphics below show the relative performance of various GPU boards that have been successfully benchmarked with FRED^MPC in both 32-bit (single precision) and 64-bit (double precision) modes, where a GeForce GTX 1060 has been used as the reference GPU.  For example, the single precision benchmark results indicate that the GeForce RTX 3070 Laptop GPU raytraced ~3.5x faster than the GeForce GTX 1650.

Note that these plots only contain GPU boards that have been tested and verified to operate with FRED^MPC.  Boards that have compute capability 6.0 or larger will be compatible with FRED^MPC .  These graphs will be updated as additional test data is acquired.

Checking for supported hardware

To use FRED^MPC, your PC must have at least one local NVIDIA GPU board available with sufficient "compute capability". The compute capability is an NVIDIA numbering scheme that relates to the specific hardware architecture of each GPU board.  The following process can be used to establish whether your PC supports the use of FRED^MPC.

1.   Upon starting FRED with an MPC license, inspect the output window.  If the text in the output window looks like the image below, FRED^MPC cannot be executed on your PC because a sufficient GPU board is not available.

2.  If no message is received in the output window on FRED^MPC startup, no further action should be required in order to use FRED^MPC.   You may still wish to query the hardware available on the PC by issuing "$cudainfo" at FRED's command line as shown below.  This will not only report to you the number of CUDA capable GPUs installed on the computer, but will list specific driver, CUDA runtime and hardware capabilities for each GPU.

3.   The image below shows an example output from the $cudainfo command for a PC with a single RTX 3070 GPU board.  Note that the RTX 3070 has NVIDIA compute capability 8.6, which satisfies the requirement for FRED^MPC.  For each CUDA device found, the section below in blue will be shown for each individual device.

Information Request

GPU Device Drivers

If you believe that your GPU board(s) have compute capability 6.0 or newer but FRED^MPC still does not recognize your device, the issue is possibly with the device driver that is currently installed on the PC.  Photon Engineering recommends that the device driver for any GPU being used with FRED^MPC be updated to the newest version available from NVIDIA.  Device drivers are available from the NVIDIA website at the following link:  http://www.nvidia.com/download/index.aspx

Running the FRED^MPC benchmark file

A FRED^MPC benchmark file is included in your FRED installation in the following location: <FRED installation directory>/Resources/MPC/Benchmark.frd

After opening the benchmark file, go to File > Save As, and save the file to a directory location that has write-access (such as your Desktop).  Expand the Embedded Scripts folder, then right mouse click on the benchmarkAnalysis script and select "Run an Embedded Script".

The benchmark analysis may take 20-30 minutes to complete, depending on the PC configuration.  At the end of the script execution, two text files will be written to the same directory where the FRED file was saved (one file each for single and double precision).  Please submit these text files to Photon Engineering at fredsupport@photonengr.com.  Your benchmark contribution will help us build a database of information that will be made available to the FRED community in order to assist with selection of GPU hardware for use with FRED^MPC.