Appendix B: Hardware and Software Requirements for GT-CONVERGE

For the GT-CONVERGE solver, the amount of recommended RAM depends on the size of the domain; the level of grid refinement; the number of species, passives, and reactions in the mechanism; and other simulation-specific criteria. A small, simple simulation (e.g., an engine sector case) can be efficiently executed in serial on a single machine with less than 4 GB of RAM. Larger simulations (e.g., a port-fuel injected, full-cylinder simulation using the SAGE detailed chemistry solver) may require several multi-core machines to achieve a solution in a reasonable time.

GT-CONVERGE 6 supports any hardware architecture that uses the x86-64 instruction set. GT-CONVERGE is designed for consumer-class hardware and does not use enterprise-class additions to that instruction set. Contact the Convergent Science Applications team to discuss processor selection.

We recommend running GT-CONVERGE on a parallel file system such as Lustre or BeeGFS. If you are running GT-CONVERGE on an asynchronous file system such as CIFS, we recommend using a hint file for your MPI library with the format shown in Figure 6 below. Set the mpi_control > io_hints_filename parameter in inputs.in to the name of this file. Refer to the GT-CONVERGE Manual for more information about inputs.in.

romio_no_indep_rw true

cb_nodes 1

Figure 6: Example MPI hint file for CIFS.

We recommend disabling hyper-threading for all general GT-CONVERGE use.

For optimal geometry rendering, run GT-CONVERGE Studio on a machine with OpenGL-compatible video drivers (3.3+). Refer to the GT-CONVERGE Studio Manual for more information about command line options for GT-CONVERGE Studio.

To manipulate a geometry in GT-CONVERGE Studio, we recommend the minimum RAM listed below in Table 17.

Table 17: RAM recommendations for manipulating geometries in GT-CONVERGE Studio.

Triangles in geometry	Recommended minimum RAM
10,000	200 MB
120,000	475 MB
200,000	650 MB
400,000	1 GB
800,000	1.8 GB
1,000,000	2 GB

Table 18 below lists the MPI (Message Passaging Interface) support packages that can be used to execute the solver in parallel. Other MPI packages are not supported.

Table 18: Compatible MPI support packages.

MPI package	Supported version for GT-CONVERGE 6
HPC-X (Linux only)	2.25
Intel MPI	21
MPICH (Linux only)	4.23
MS-MPI (Windows only)	10.1
Open MPI (Linux only)	5.0

Earlier versions of HPC-X, Intel MPI, MPICH, MS-MPI, and Open MPI are not supported.

For multi-node simulations, GT-CONVERGE works best with a high-performance interconnect such as InfiniBand or Omni-Path. The HPC-X, Intel MPI, and Open MPI packages that are included with the GT-CONVERGE solver work with InfiniBand automatically. We do not recommend using HPC-X unless you are running on InfiniBand with NVIDIA InfiniBand cards, for which HPC-X was specifically developed.