Excellent topical perspective article by John Stone, David Hardy, Ivan Ufimtsev and Klaus Schulten just published at Journal of Molecular Graphics and Modelling! It provides detailed analysis of algorithms that leverage GPU computing and gives future outlook:
The broad range of existing work demonstrates performance benefits from the use of GPUs as massively parallel co-processors for arithmetic-intensive molecular modeling applications. State-of-the-art GPU hardware designs, like the AMD Cypress and NVIDIA Fermi, offer greatly increased computational capabilities and solve some of the past limitations of GPUs. Double precision is now supported at just twice the cost of single precision operations, and error correcting memory is supported. CUDA driver and compiler technology is mature enough for production-level software development, and OpenCL is anticipated to follow this lead as vendor support improves.
The most exciting applications of GPUs to molecular modeling are perhaps those that have resulted in a “computational phase transition,” transforming what had previously been batch-mode computations on clusters into interactive computations that can now be performed on laptop or desktop computers. Recent and continuing improvements to the state-of-the-art GPUs, such as new special-purposes caches and hardware reduction operations, will enable development of GPU algorithms that were previously difficult to map to GPU hardware with high efficiency, while reducing the number of development hurdles encountered by computational scientists that are just beginning to learn GPU programming techniques.
Graphics processing units (GPUs) have traditionally been used in molecular modeling solely for visualization of molecular structures and animation of trajectories resulting from molecular dynamics simulations. Modern GPUs have evolved into fully programmable, massively parallel co-processors that can now be exploited to accelerate many scientific computations, typically providing about one order of magnitude speedup over CPU code and in special cases providing speedups of two orders of magnitude. This paper surveys the development of molecular modeling algorithms that leverage GPU computing, the advances already made and remaining issues to be resolved, and the continuing evolution of GPU technology that promises to become even more useful to molecular modeling. Hardware acceleration with commodity GPUs is expected to benefit the overall computational biology community by bringing teraflops performance to desktop workstations and in some cases potentially changing what were formerly batch-mode computational jobs into interactive tasks.
John E. Stone, David J. Hardy, Ivan S. Ufimtsev, Klaus Schulten, GPU-accelerated molecular modeling coming of age, Journal of Molecular Graphics and Modelling, In Press, Corrected Proof, Available online 8 July 2010, ISSN 1093-3263, [DOI: 10.1016/j.jmgm.2010.06.010] [PDF]