OrcaFlex can be used to perform demanding calculations, and so a powerful computer is best suited to this task.
Our experience is that typical engineering analysis with OrcaFlex involves the processing of a large number of individual simulations. Running those cases in parallel on a computer with multiple CPU cores will usually be the fastest way to obtain your results. However, we will begin with the advice that we would offer for a single-core PC. To get the best results we would recommend:
- A 64 bit edition of Windows.
- A powerful processor with fast floating point and memory performance. This is the most important factor since OrcaFlex is a computation-intensive program and simulation run times can be long for complex models. There are many general processor benchmark comparison tables available online.
- At least 4GB of memory. This is less important than processor performance but some aspects of OrcaFlex do perform better when more memory is available.
- As much disk space as you require to store simulation files. Simulation files vary in size, but can be hundreds of megabytes each for complex models.
- A screen resolution of 1920×1080 or greater.
- A DirectX 9 compatible graphics card with at least 256MB memory for the most effective use of the shaded graphics facility. Note that OrcaFlex does not offload any calculation to the graphics card of the PC, so the graphics hardware will not affect simulation run time.
- Microsoft Excel (Excel 2010, or later) in order to use the OrcaFlex automation facilities. Both 32 bit and 64 bit versions of Excel are supported although we strongly recommend the 64 bit version to avoid the severe constraints on memory usage of the 32 bit version.
For parallel processing of multiple simulations, a machine with many CPU cores can be chosen. When cases are processed in parallel, the completion rate of cases, or completion of an amount of simulation time per unit of real time is how speed can be measured. We refer to this measure as throughput. For the parallel processing situation, there is some additional hardware advice:
- OrcaFlex derives little benefit from running more than one execution thread on a single CPU core. This feature is usually called hyperthreading, or similar. It is hyperthreading that leads to a difference between the number of physical cores and logical cores for a given machine. We expect improvement of throughput until the number of parallel cases reaches the machine physical core count. Additional parallel cases after this limit will not improve OrcaFlex throughput further, and the additional contention they generate for other system resources can in fact reduce throughput.
- In our experience, it appears that an allocation of 2GB of memory per core is ample for multi-core machines. When running cases in parallel, certainly when the OrcaFlex batch processing facility is used, the simulation log files will be stored on the machine local disk, and are not held in memory.
- As noted above, disk access is an important part of parallel processing, so the speed and bandwidth of local storage should be considered in the machine specification.
For comparison of different machines, especially an indication of their ability to increase throughput via parallel processing, we provide a test program. More information can be found on the OrcaFlex benchmark page.
OrcaWave analysis typically involves processing multiple calculation tasks for a single modelled case, and these tasks can be processed in parallel on multi-core systems. The memory requirement for each calculation task is estimated on the OrcaWave validation page, and can be reviewed before an analysis is started. For analysis with a large diffraction mesh, OrcaWave might have a much larger memory requirement than OrcaFlex.
Our hardware recommendations above are therefore extended for OrcaWave to note that more memory might be required. The OrcaWave user can reduce the thread count before analysis is started in order that the expected memory usage fits into the free memory that the system has available.