Overview / Usage

This project is a research effort and proof of concept design for scalable FPGA-based acceleration of computation algorithms for molecular dynamics simulation focused on large molecules such as polyketides, peptides, proteins, RNAs in application for drug design and drug discovery.

For simplicity, and ability to immediately focus on FPGA-based acceleration software interface and algorithms I use cloned open-source UIUC simulation package NAMD that has full-fledged architecture for mass-parallel acceleration with classical SMP (mult-CPU/core), NVidia GPU(multi) CUDA, and infiniband network (verbs).

The project key goal is to investigate reasonable cost-effective FPGA-based acceleration outperforming GPGPU-based acceleration of molecular dynamics algorithms utilizing low on-crystal HBM memory, increased data bandwidth, and ability to build mass scalable design using direct 100G interconnect between FPGA cards.

​

Methodology / Approach

In this project I am going to develop FPGA-accelerated blocks (OpenCL kernels & RTLs) alternative to GPGPU implementation for key computationally-extensive core algorithms solving bonding, non-bonding, PME long-range forces calculation, generalized Born implicit solvent (GIBS), and solvent accessible surface area (SASA).

Technologies Used

Intel Quartus Pro

Intel FPGA OpenCL SDK

Terasic DE10-Pro

Repository

https://github.com/slvbsp2009/NAMD_XLNX_ACCEL