benchmark
The benchmark
command can be used to benchmark the platform to provide a performance baseline.
The routine preallocates a square matrix of size 5120 and repeatedly solves it for 50 times. Unlike other approaches that normally rely on dgemm
subroutine, here dgesv
subroutine is used as for FEM, the main operations are solving matrices rather than multiplying matrices.
Syntax
Notes
The basic idea of the benchmark resembles that of linpack, but it is not intended to be as versatile as linpack. Thus, it does not provide any configurable options.
The theoretical maximum GFLOPS is not often achievable in practical applications. Modern hardware is often bounded by other factors such as memory bandwidth.
One can perf
this benchmark to reveal the GFLOPS, for example,
The corresponding result is
This roughly gives 114 GFLOPS, which is around of the theoretical maximum GFLOPS of a i7-1185G7
processor. Of course, the actual performance depends on many other factors. Here the numbers are indicative. A rate of means the platform is two times faster than a rate of .
The practical FEM involves other processes such as updating elemental stiffness (small matrix manipulations) and assembling global stiffness (random memory accesses). As the result, the performance of analysis would further be a fraction of GFLOPS of this benchmark test.
Last updated