Skip to content

Reply To: Memory allocation error


Memory usage is quite model dependent. However, for a first approximation you can multiply the number of cells by the number of discrete velocities by the number of bytes per base type unit (e.g. 8 for double precision FPT).

Note that you probably don’t want to fully fill up your memory but rather increase the degree of parallelization for larger problems. While keeping the compute resource fixed while maximizing memory utilization will increase the raw throughput (MLUPs) the resulting discrete time steps per second will necessarily reduce.