Memory Calculator

To aid you in the planning and setup of your simulation PIConGPU provides python tools for educated guesses on simulation parameters. They can be found under lib/python/picongpu/extra/utils.

Note

The utils have been moved to the picongpu.extra submodule.

Calculate the memory requirement per device.

from picongpu.extra.utils import MemoryCalculator
class picongpu.extra.utils.memory_calculator.MemoryCalculator(n_x, n_y, n_z, precision_bits=32)

Memory requirement calculation tool for PIConGPU

Contains calculation for fields, particles, random number generator and the calorimeter plugin. In-situ methods other than the calorimeter so far use up negligible amounts of memory on the device.

__init__(n_x, n_y, n_z, precision_bits=32)

Class constructor

Parameters:

  • n_x (int) – number of cells in x direction (per device)

  • n_y (int) – number of cells in y direction (per device)

  • n_z (int) – number of cells in z direction (per device)

  • precision_bits (int) – floating point precision for PIConGPU run

mem_req_by_calorimeter(n_energy, n_yaw, n_pitch, value_size=None)

Memory required by the particle calorimeter plugin. Each of the (n_energy x n_yaw x n_pitch) bins requires a value (32/64 bits). The whole calorimeter is represented twice on each device, once for particles in the simulation and once for the particles that leave the box.

Parameters:

  • n_energy (int) – number of bins on the energy axis

  • n_yaw (int) – number of bins for the yaw angle

  • n_pitch (int) – number of bins for the pitch angle

  • value_size (int) – value size in particle calorimeter {unit: byte} (default: 4)

Returns:

req_mem – required memory {unit: bytes} per device

Return type:

int

mem_req_by_fields(n_x=None, n_y=None, n_z=None, field_tmp_slots=1, particle_shape_order=2, sim_dim=3, pml_n_x=0, pml_n_y=0, pml_n_z=0)

Memory reserved for fields on each device

Parameters:

  • n_x (int) – number of cells in x direction (per device)

  • n_y (int) – number of cells in y direction (per device)

  • n_z (int) – number of cells in z direction (per device)

  • field_tmp_slots (int) – number of slots for temporary fields (see PIConGPU memory.param : fieldTmpNumSlots)

  • particle_shape_order (int) – numerical order of the assignment function of the chosen particle shape CIC : order 1 TSC : order 2 PQS : order 3 PCS : order 4 (see PIConGPU species.param)

  • sim_dim (int) – simulation dimension (available for PIConGPU: 2 and 3)

  • pml_n_x (int) – number of PML cells in x direction, combined for both sides

  • pml_n_y (int) – number of PML cells in y direction, combined for both sides

  • pml_n_z (int) – number of PML cells in z direction, combined for both sides

Returns:

req_mem – required memory {unit: bytes} per device

Return type:

int

mem_req_by_particles(target_n_x=None, target_n_y=None, target_n_z=None, num_additional_attributes=0, particles_per_cell=2, sim_dim=3)

Memory reserved for all particles of a species on a device. We currently neglect the constant species memory.

Parameters:

  • target_n_x (int) – number of cells in x direction containing the target

  • target_n_y (int) – number of cells in y direction containing the target

  • target_n_z (int) – number of cells in z direction containing the target

  • num_additional_attributes (int) – number of additional attributes like e.g. boundElectrons

  • particles_per_cell (int) – number of particles of the species per cell

  • sim_dim (int) – simulation dimension (available for PIConGPU: 2 and 3)

Returns:

req_mem – required memory {unit: bytes} per device and species

Return type:

int

mem_req_by_rng(n_x=None, n_y=None, n_z=None, generator_method='XorMin')

Memory reserved for the random number generator state on each device.

Check random.param for a choice of random number generators. If you find that your required RNG state is large (> 300 MB) please see memory.param for a possible adjustment of the reservedGpuMemorySize.

Parameters:

  • n_x (int) – number of cells in x direction (per device)

  • n_y (int) – number of cells in y direction (per device)

  • n_z (int) – number of cells in z direction (per device)

  • generator_method (str) – random number generator method - influences the state size per cell possible options: “XorMin”, “MRG32k3aMin”, “AlpakaRand” - (GPU default: “XorMin”) - (CPU default: “AlpakaRand”)

Returns:

req_mem – required memory {unit: bytes} per device

Return type:

int