General

• All files from version 2.0 can be read by the programs in version 3.0.
  
• Temperature coupling can now be performed with either a Berendsen or a Nose-Hoover thermostat.
  
• Pressure coupling can now be performed with either a Berendsen or a Parinello-Rahman barostat.
  
• We have implemented Lennard-Jones-only inner loops, and a lot of special combinations like tabulated coulomb but normal Lennard-Jones interactions. This makes PME simulations faster, especially with the GROMOS96 forcefield (GROMOS96 needs a LJ cut-off of 1.4 nm, while PME works optimally with a cut-off of 0.9). It also speeds up simulations with a lot of uncharged atoms.
  
• Improved inner loops, which give a large performance increase on all platforms.
  
• Handwritten assembly inner loops for x86 hardware, using SSE (Intel P3/P4) and Extended 3DNow! (AMD Athlon/Duron) multimedia instructions. This will boost your performance almost a factor of two on these platforms. You need Linux kernel 2.4 or later to use the SSE loops (the AMD 3DNow loops work with Linux 2.2).
  No, it is not a typo - this release is twice as fast as gromacs 2.0 on pc hardware, which makes it 3-10 times faster than any other MD program!
  
• Solvent optimizations for solvent molecules consisting of one charge group with no internal non-bonded interactions.
  
• Special water-water optimizations, in addition to the already present water-other atom optimizations.
  
• All programs support triclinic boxes. mdrun uses grid search for triclinic boxes, which is just as fast as for rectangular boxes.
  
• Support for rhombic dodecahedrons and truncated octahedrons, since these are special cases of a triclinic box. The dodecahedron has a volume of 0.71 of that of a cubic box with the same box vector length, this saves CPU time.
  
• The efficiency of the grid search has almost doubled.
  
• Grid search is no longer limited to grids of at least 5x5x5, which means that it can also be used for small boxes.
  
• Molecules can now be longer than the box. You can even simulate polymers crossing the box hundreds of times.
  
• The block of a dummy and its constructing atoms can now cross a processor boundary when running in parallel. This is necessary e.g. to simulate polymers using anisotropic united atoms.
  
• All non-bonded interactions between pairs of energy groups can be excluded. This can speed up simulations with frozen groups or mdrun -rerun.
  
• Interpolation via soft-core interactions has been implemented for free energy calculations.
  
• Free energy contributions of constraints can also be calculated with SHAKE (in Gromacs 2.0 this was only supported with LINCS).
  
• We have added a section to the manual on writing topologies for free energy calculations.
  
• Made accurate calculation of viscosities possible by adding a cosine shaped acceleration profile, which is decoupled from the temperature coupling.
  
• More checks have been added for the consistency of the input, checks have been added for pressure scaling of more than 1%.
  
• The -e option for the last frame to read from trajectory now has a margin to avoid missing the last frame due to rounding errors in the time.
  
• Several tools now have a -tu option to change the time unit in the output files and the -b, -e and -dt options.
  
• Conjugate gradient minimization now also works in parallel.
  
• pdb2gmx can merge multiple chains into one molecule. This is useful e.g. for connecting chains with disulfide bridges.
  
• The FFTW library is not distributed as a part of GROMACS anymore, since it often required you to have two copies of FFTW. You might have to install FFTW separately now, but only the first time you compile GROMACS.
  
• The configuration has been completely automated by using GNU autoconf scripts.
  
• Shared libraries should be supported on almost all platforms.
  
• The GMXLIB variable can now be a unix-style path instead of only a single directory.
  
• As if this was not enough: Starting with version 3.0, the entire GROMACS code is free software, licensed under the GNU General Public License You don't have to sign a license - the source code and Linux RPM packages can be downloaded immediately from this site!
  

Important changes that might affect your simulation results

• The number of degrees of freedom for each group is now (3*natoms-nconstraints)*(ntot-3)/ntot, where ntot is the total number of degrees of freedom. This used to be 3*natoms-nconstraints-3. This will have a very small effect on the temperature when using more than one temperature coupling group.
  
• The maximum force for Steepest Descent and Conjugate Gradient minimization has been changed from per degree of freedom to per atom. This means a few more iterations may be needed for convergence compared to the previous Gromacs version.
  
• Forces on frozen particles are now not taken into account for the converge check for Steepest Descent and Conjugate Gradient minimization.
  
• The constant c_rf has been added to the reaction-field potential. This constant has been omitted in all previous versions of Gromacs. This change only affects the energies, not the forces. When all charge groups are neutral the energy is not affected.