NAMD runs very fast on the Itanium processor under Linux, including the SGI Altix. Native binaries for 64-bit x86 processors such as AMD Opteron and Intel EMT64 are 25% faster than 32-bit binaries.
NAMD has been ported to the new Intel-based Macintosh platform. Binaries are only 32-bit, but should run on newer 64-bit machines.
These two large, scalable, but immature platforms are now supported. Neither platform supports dynamic linking and our experience shows that binaries would be quickly out of date, so only source code is released. Also, BlueGene/L requires the latest development version of Charm++, rather than charm-5.9 as shipped with the NAMD source code.
Tuning of the NAMD inner loop has provided a 30% performance boost with the IBM XL compiler. New Mac OS X binaries are up to 70% faster, but users must download the IBM XL runtime library from http://ftp.software.ibm.com/aix/products/ccpp/vacpp-rte-macos/
A new method, implemented in Tcl, efficiently calculates the potential of mean force along a reaction coordinate by applying adaptive biasing forces to provide uniform sampling.
The replica exchange method is implemented as a set of Tcl scripts that use socket connections to drive a set of NAMD jobs, exchanging temperatures (or any other scriptable parameter) based on energy.
Tcl-scripted forces may be efficiently applied individually to large numbers of atoms to implement boundaries and similar forces.
Memory usage for simulations of large, highly bonded structures such as covalent crystals and for sparse simulations such as coarse-grained models has been greatly reduced without affecting the performance of typical biopolymer simulations.
Both NAMD and psfgen (standalone or VMD plug) read and interpret the new CHARMM 31 stream files (combining topology and parameters) and the new CMAP crossterm (dihedral-dihedral) potential function.
Geometric combining of Lennard-Jones 
(radius) parameters
(
), as required by the OPLS force
field, is available with the option vdwGeometricSigma.