Accelerating Molecular Dynamics Computation

• We selected to accelerate GROMACS (GROningen MAchine for Chemical Simulations), the molecular dynamics package used for simulations of proteins, lipids and nucleic acids. This software package was developed in the Biophysical Chemistry department of the University of Groningen.
• The accelerator we propose is connected to a x86 host computer using PCIe. The computation is performed by a Scalar unit in conjunctioin with a MapReduce unit [1] where:
  • Scalar unit is a RISC engine which fetches in each clock cycle from its program memory a pair of instructions, one for itself and another to be issued toward the Map array of cells
  • Map section is an array of cells, each with its predicated execution unit and local memory; each cell executes, according to its internal state, the instruction received from the Scalar unit
  • Reduction is a log-depth circuit which performs functions defined on vectors which return a scalar (for example: add, min, ...)

• the degree of parallelism in our MapReduce accelerator is 75.6% and scales with the number of cells
• the energy cost (in Wh per micoseconds of simulation):
  • 4.5 Wh/μs for FPGA implementation
  • 0.2 Wh/μs for ASIC 28nm implementation

• extend the approach form the system of forces considered in [2] to other kinds of forces
• combining the results of the Kinari analysis, developed in Ileana Streinu's research group LinkageLab, with our results in accelerating molecular dynamics simulations.

References

[1] Gheorghe M. Stefan, Mihaela Malita: "Can One-Chip Parallel Computing Be Liberated From Ad Hoc Solutions? A Computation Model Based Approach and Its Implementation", 18th International Conference on Ciruits, Systems, Communications and Computers (CSCC 2014), Santorini Island, Greece, July 17-21, 2014, 582-597.
[2] David Mihaita, : "N-Body Problem Application, on a Map-Reduce Accelerator, to Molecular Dynamics", Master Thesis, UPB, 2016
[3] Nicolae Goga, Mihaela Malita, David Mihaita, Gheorghe M. Stefan: "FPGA Based Accelerator for Molecular Dynamics", ACSE 2016, Rome, 27-29 July, ISBN: 978-84-944311-8-0