Octgrav is a very fast tree-code which runs on massively parallel Graphical Processing Units (GPU) with NVIDIA CUDA architecture. The algorithms are based on parallel-scan and sort methods. The tree-construction and calculation of multipole moments is carried out on the host CPU, while the force calculation which consists of tree walks and evaluation of interaction list is carried out on the GPU. In this way, a sustained performance of about 100GFLOP/s and data transfer rates of about 50GB/s is achieved. It takes about a second to compute forces on a million particles with an opening angle of $ heta approx 0.5$.

To test the performance and feasibility, we implemented the algorithms in CUDA in the form of a gravitational tree-code which completely runs on the GPU. The tree construction and traverse algorithms are portable to many-core devices which have support for CUDA or OpenCL programming languages. The gravitational tree-code outperforms tuned CPU code during the tree-construction and shows a performance improvement of more than a factor 20 overall, resulting in a processing rate of more than 2.8 million particles per second.

The code has a convenient user interface and is freely available for use.

Traditionally, a simulation of a dense stellar system required choosing an initial model, running an integrator, and analyzing the output. Almost all of the effort went into writing a clever integrator that could handle binaries, triples and encounters between various multiple systems efficiently. Recently, the scope and complexity of these simulations has increased dramatically, for three reasons: 1) the sheer size of the data sets, measured in Terabytes, make traditional 'awking and grepping' of a single output file impractical; 2) the addition of stellar evolution data brings qualitatively new challenges to the data reduction; 3) increased realism of the simulations invites realistic forms of 'SOS': Simulations of Observations of Simulations, to be compared directly with observations. We are now witnessing a shift toward the construction of archives as well as tailored forms of visualization including the use of virtual reality simulators and planetarium domes, and a coupling of both with budding efforts in constructing virtual observatories. This review describes these new trends, presenting Starlab as the first example of a full software environment for realistic large-scale simulations of dense stellar systems.

AMUSE is an open source software framework for large-scale simulations in astrophysics, in which existing codes for gravitational dynamics, stellar evolution, hydrodynamics and radiative transport can be easily coupled and placed in the appropriate observational context.

McScatter illustrates a method of combining stellar dynamics with stellar evolution. The method is intended for elaborate applications, especially the dynamical evolution of rich star clusters. The dynamics is based on binary scattering in a multi-mass field of stars with uniform density and velocity dispersion, using the scattering cross section of Giersz (MNRAS, 2001, 324, 218-30).

Roche is a visualization and analysis tool for drawing the Roche-lobe geometry of evolving binaries. Roche can be used as a standalone program reading data from the command line or from a file generated by SeBa (ascl:1201.003). Eventually Roche will be able to read data from any other binary evolution program. Roche requires Starlab (ascl:1010.076) version 4.1.1 or later and the pgplot (ascl:1103.002) libraries. Roche creates a series of images, based on the SeBa output file SeBa.data, displaying the evolutionary state of a binary.

SeBa is the stellar and binary evolution module, fully integrated into the kira N-body integrator, for Starlab (ascl:1010.076), although it can also be used as a stand-alone module for non-dynamical applications. Due to the interaction between stellar evolution and stellar dynamics, it is difficult to solve for the evolution of both systems in a completely self-consistent way. The trajectories of stars are computed using a block timestep scheme, as described earlier. Stellar and binary evolution is updated at fixed intervals (every 1/64 of a crossing time, typically a few thousand years). Any feedback between the two systems may thus experience a delay of at most one timestep. Internal evolution time steps may differ for each star and binary, and depend on binary period, perturbations due to neighbors, and the evolutionary state of the star. Time steps in this treatment vary from several milliseconds up to (at most) a million years.

Sapporo mimics the behavior of GRAPE hardware and uses the GPU to perform high-precision gravitational N-body simulations. It makes use of CUDA and therefore only works on NVIDIA GPUs. N-body codes currently running on GRAPE-6 can switch to Sapporo by a simple relinking of the library. Sapporo's precision is comparable to that of GRAPE-6, even though internally the GPU hardware is limited to single precision arithmetics. This limitation is effectively overcome by emulating double precision for calculating the distance between particles.

Bonsai is a gravitational N-body tree-code that runs completely on the GPU. This reduces the amount of time spent on communication with the CPU. The code runs on NVIDIA GPUs and on a GTX480 it is able to integrate ~2.8M particles per second. The tree construction and traverse algorithms are portable to many-core devices which have support for CUDA or OpenCL programming languages.

MPWide is a light-weight communication library for distributed computing. It is specifically developed to allow message passing over long-distance networks using path-specific optimizations. An early version of MPWide was used in the Gravitational Billion Body Project to allow simulations across multiple supercomputers.

The use of graphics processing units offers an attractive alternative to specialized hardware, like GRAPE. The Kirin library mimics the behavior of the GRAPE hardware and uses the GPU to execute the force calculations. It is compatible with the GRAPE6 library; existing code that uses the GRAPE6 library can be recompiled and relinked to use the GPU equivalents of the GRAPE6 functions. All functions in the GRAPE6 library have an equivalent GPU implementation. Kirin can be used for direct N-body simulations as well as for treecodes; it can be run with shared-time steps or with block time-steps and allows non-softened potentials. As Kirin makes use of CUDA, it works only on NVIDIA GPUs.

Smoothed Particle Hydrodynamics (SPH) is a Lagrangian particle method that approximates a continuous fluid as discrete nodes, each carrying various parameters such as mass, position, velocity, pressure, and temperature. In an SPH simulation the resolution scales with the particle density; StarSmasher is able to handle both equal-mass and equal number-density particle models. StarSmasher solves for hydro forces by calculating the pressure for each particle as a function of the particle's properties - density, internal energy, and internal properties (e.g. temperature and mean molecular weight). The code implements variational equations of motion and libraries to calculate the gravitational forces between particles using direct summation on NVIDIA graphics cards. Using a direct summation instead of a tree-based algorithm for gravity increases the accuracy of the gravity calculations at the cost of speed. The code uses a cubic spline for the smoothing kernel and an artificial viscosity prescription coupled with a Balsara Switch to prevent unphysical interparticle penetration. The code also implements an artificial relaxation force to the equations of motion to add a drag term to the calculated accelerations during relaxation integrations. Initially called StarCrash, StarSmasher was developed originally by Rasio.

Torch simulates coupled gas and N-body dynamics in astrophysical systems such as newly forming star clusters. It combines the FLASH (ascl:1010.082) code for gas dynamics and the ph4 code for direct N-body evolution via the AMUSE framework.

HUAYNO implements integrators derived from second order Hamiltonian splitting for N-body dynamics. This integration scheme conserves energy and momentum with little or no systematic drift. The code uses an explicit but approximate formula for the time symmetrization that is compatible with the use of individual time steps, making an iterative scheme unnecessary. HUAYNO is available as part of the AMUSE package (ascl:1107.007).

GWToolbox simulates gravitational wave observations for various detectors. The package is composed of three modules, namely the ground-based detectors (and their targets), the space-borne detectors (and their targets) and pulsar timing arrays (PTA). These three modules work independently and have different dependencies on other packages and libraries; failed dependencies met in one module will not influence the usage of another module. GWToolbox can accessed with a web interface (gw-universe.org) or as a python package (https://bitbucket.org/radboudradiolab/gwtoolbox).

TRES simulates hierarchical triple systems with stellar and planetary components, including stellar evolution, stellar winds, tides, general relativistic effects, mass transfer, and three-body dynamics. It combines stellar evolution and interactions with three-body dynamics in a self-consistent way. The code includes the effects of common-envelope evolution, circularized stable mass transfer, tides, gravitational wave emission and up-to-date stellar evolution through SeBa (ascl:1201.003). Other stellar evolution codes, such as SSE (ascl:1303.015), can also be used. TRES is written in the AMUSE (ascl:1107.007) software framework.