Vectorization of the generalized Born model for molecular dynamics on shared-memory computers

Vectorization and performance of the generalized Born solvation model (GB) as it is implemented in the AMBER program is presented in this study. Nonbonded interactions computed within the generalized Born model use the pairwise approximation of Hawkins et al. [J. Phys. Chem. 100 (1996) 19,824], which is in turn a variant of the model proposed by Schaeffer and Froemmel [J. Mol. Biol. 216 (1990) 1045]. The performance of this implementation on CRAY SV1 vector machines is discussed with reference to several proteins. Loop vectorization has shown improvements by as much as a factor of 10. Comparison of the timings of the GB method against simulations using fully hydrated proteins favors the GB method.