Here you can see which new features / functionality are available in the released versions of CP2K

Features available or under development in the latest git version (https://github.com/cp2k/cp2k)

• Speedup grid kernels, especially non-orthorhombic on CPU and integrate on GPU
• Add support for ARM64
• Drop support for GCC 6 (#1203)
• Upgrade to COSMA 2.5 (#1303 )
• Upgrade to LibXC 5 and harmonize its input with built in functionals
• xTB/DFTB: Add stress tensor with efield
• Motion: Add space group symmetry
• Fix multi GPU by setting the active device consistently (#814)
• Fix MOLDEN output (#1335)
• XAS_TDP: Fix bug in open-shell SOC (#1304)
• PINT: Fix conserved quantity in PINT-RPMD restart (#1290)
• ELPA: As a precaution use only for large matrices by default (#1444)
• libvori : Augmented .voronoi file format provides improved support for TRAVIS (e.g. for spectra simulations)

• Fix bug affecting ADMM on GPUs (#893)
• Fix bug affecting Amber dihedrals (#984)
• Drop support for Python 2 and non-OpenMP builds
• Add interfaces to GRRM17 and SCINE codes
• Add support for Cosma (https://github.com/eth-cscs/COSMA)
• Add support for Voronoi integration of electron density (https://brehm-research.de/voronoi)
• Add support for output of electron density in compressed BQB format (https://brehm-research.de/bqb)
• OpenMP refactoring and speed-ups for one electron integrals
• Response code for polarizabilities: add finite difference debug, hybrid functionals, and ADMM
• Harris functional based on Kohn-Sham density
• TDDFPT code refactoring, add sTDA kernel, xTB/sTDA method
• NNP: Behler-Parrinello Neural Network Potentials
• XAS_TDP: Add OT solver and improve performance
• mGGA: Add stress tensor and fix bug (#1116)
• QMMM: Add benchmarks and speedup GEEP with OpenMP
• LS: Add sign calculation based on submatrix method
• ALMO: Add trust region methods
• RI-HFX: Add resolution of identity for Hartree-Fock exchange
• RPA/GW/MP2: Several optimizations and refactoring to low-scaling implementation
• CUDA: GPU acceleration of collocate and integrate grid operations (experimental)

• SIRIUS: Plane Wave module with GPU support, see also this tutorial for Quantum ESPRESSO users.
• xTB: Tight-binding module based on 10.1021/acs.jctc.7b00118
• RPA / GW / MP2: migrated to DBCSR tensors.
• HELIUM: New canonical worm algorithm based on 10.1103/PhysRevE.74.036701.
• XAS_TDP: X-ray absorption spectra simulations using linear-response TDDFT.
• NEGF: Contact-specific temperature, correct shift and scale factors.
• S-ALMO: Major refactoring, added wide variety of options.
• CDFT: Cleanup and bug fixing.
• FPGA interface for pw FFT.
• Updated libraries: DBCSR, ELPA, libint, libxc, libxsmm.
• The cp2k_shell was integrated into the main binary, simply call cp2k with -s or --shell.
• Development moved from SVN to Git

• Projection-operator adiabatization (POD) method
• CP2K can now do Plane Wave calculations using CPU and GPU, based on an electronic structure library SIRIUS
• Include NVIDIA P100 kernels for DBCSR
• Update toolchain
• Prevent ELPA diagonalization crashes with small matrices and/or large core counts
• Faster routines for reading and writing cube files using MPI I/O
• Docker based tests

• Sparse tensor framework based on DBCSR
• Flags __ELPA2 and __ELPA3 removed, instead use -D__ELPA=YYYYMM to specify library version.
• Constrained DFT, see FORCE_EVAL/DFT/QS/CDFT and FORCE_EVAL/MIXED/MIXED_CDFT
• Cubic scaling GW
• GW + image charge to compute electronic levels of a molecule on a metal surface
• Constraint cell optimization MOTION/CELL_OPT#CONSTRAINT

• Maximum Overlap Method (MOM)
• Modified Atomic Orbitals (MAO) Analysis
• Easier installation with an improved toolchain
• Improved Development Tools: prettifier, API documentation
• Improved Coding Standards
• More collective variables
• Transport with Omen: improvements
• libcp2k.h interface (C/C++ header)
• Remote Memory Access (RMA) for future architectures
• Various performance improvements and bug fixes
• GTH-PBE pseudopotentials for the Lanthanide elements
• Polarized Atomic Orbitals from Machine Learning (PAO-ML)
• Cubic-scaling RPA
• Fast method for periodic ERI reducing overhead of image charge correction in QM/MM and used in cubic-scaling RPA
• Drop support for PLUMED 1.3 (PLUMED 2.x is now required)
• Improved linear scaling (LS) DFT MD with curvy steps
• Support for Hybrid density functionals in TDDFT

• Improvement of the Path integral code and use of PIGLET thermostat
• Workaround for ifort 'feature' leading to incorrectly ignored 1-4 interactions in classical MD simulations.
• Gradients for MP2 in the unrestricted case
• Current output for EMD
• constant E/D simulations
• RMA based DBCSR
• Improved portability (xlf90)
• G0W0 and eigenvalue self-consistent GW
• Improved testing: the make target 'test' will now regtest the code
• Basic k-point functionality for GGA DFT
• Faster TRS4 for semi-empirical runs.
• Interface to the PEXSI library
• PLUMED 2.0 interface
• Interface to ELPA2015 ELPA
• Filtered Basis method
• More optimized CUDA kernels
• Coupling with the quantum transport code OMEN
• Implicit Poisson solver, with dielectric, and different boundary conditions
• Rho mixing for LS SCF enabled
• Speedup for LRIGPW method
• Support for the ASE Python toolkit.
• Updated toolchain
• Saveguard against known issue with CUDA cufft 7.0
• Polarized atomic orbitals
• REPEAT variant of the RESP atomic charge fitting method
• Streamlined error handling
• Support for Intel's libxsmm
• Various bug fixes

• Check the list of back-ported bug fixes here

• Check the list of back-ported bug fixes here

• Allowing GPU acceleration for full matrix multiplies by using the DBCSR multiply routines
• RTP and EMD with Hartree-Fock exchange and ADMM NONE
• Improved FULL_SINGLE_INVERSE preconditioner and linear scaling PRECOND_SOLVER INVERSE_UPDATE for large systems
• Self-consistent continuum solvation (SCCS) model
• K-points (partial implementation for some methods)
• Improved linear scaling routines.
• Improved RPA frequency integration methods.
• QUIP Manybody potential
• Optimization of LRI basis sets
• Full Fortran 2003 compliance
• Various bug-fixes, refactoring, and speed-ups
• Collection of production grade parameters (basis-sets, pseudo-potentials, etc.)
• File discovery mechanism (controllable via compile flag -D__DATA_DIR or environment variable $CP2K_DATA_DIR )
• New build-system, replaced makedepf90 with novel makedep.py
• Started splitting code into sub-directories (packages) with well defined dependencies among each other
• Auto-tunning framework for DBCSR cuda-kernels and many readily optimized kernel-parameters
• QM/MM for DFTB
• LRIGPW
• Hirshfeld population analysis
• DM and Charge Constraint Projection based ADMM

• MP2 gradients and stress
• emacs plugin for input syntax highlighting (→ CP2K Tools)
• vim plugin for input syntax highlighting (→ CP2K Tools)
• Post-SCF linear response, including Raman
• Energy use framework for Cray
• RI-MP2 auxiliary basis optimization
• Global optimization of geometries
• SCP tight binding
• Relativistic corrections for atomic blocks
• CUDA enabled DBCSR
• Improved OMP parallelism
• Tree Monte Carlo: additional parallelism in MC
• ALMO: linear scaling for molecular systems
• Removed internal ELPA, using it as an external library instead
• Integrated molecular basis set optimization
• Langevin dynamics regions
• DCD dump option for an aligned cell (allows a reconstruction of scaled coordinates)
• and many bug fixes …

• GPW-MP2 & RPA
• Adaptive QM/MM
• Non-local vDW functionals, PBEsol
• Integrated basis set optimisation
• Support for non-linear core corrected pseudos
• Additional linear scaling algorithms and properties
• Possibility of using image charges
• Periodic RESP charges
• PLUMED support
• ELPA eigensolver support
• libxc support
• Improved ifort/MKL support
• Process topology mapping for Cray Gemini

The commits until end 2011 can be viewed as a short animation .