NVIDIA Success Story

Life Sciences

Complex Multi-Phase Flow Molecular Dynamics Simulation


Molecular dynamics

Molecular dynamics (MD) simulation is an emerging scientific computing method widely used in the medical, materials, chemistry, energy and electro-mechanical industries. Scientists want to run MD simulations with larger number of particles and with greater level of detail. However, the computation requirements for these complex MD simulations far exceeds the capability of today’s supercomputers. Moreover, not every research group can afford to build a large supercomputer to run MD simulations.


Molecular dynamics

To achieve its goals of running complex MD simulations, the Laboratory of Multi-Phase Complex Systems at the Institute of Process Engineering (IPE) of the Chinese Academy of Sciences used a computing system equipped with a NVIDIA Tesla™ C870 GPU. The Tesla C870 GPU is based on the novel NVIDIA CUDA™ parallel computing architecture that delivers one to two orders of magnitude higher performance than CPUs for a range of applications. The capability of the GPU-based MD simulation method enabled researchers at IPE to create a microscopic simulation of mesoscopic behavior. This was demonstrated by actual phenomena such as cavity flow and particle-air bubble contact, and ran 20 to 60 times faster than using the single-core CPU method, achieving almost 150 Gflops of performance from a single Tesla C870 GPU In comparison, the CPU delivered just 2.4 Gflops of performance. The floating point performance of the Tesla C870 GPU was 20-30 times faster than a single-core CPU in the most time-consuming portion of the MD simulation, which calculates the force acting on molecular pairs interacting with each other, This enables IPE to do complex multi-phase MD simulations that were otherwise not possible using CPUs.


Molecular dynamics

By taking advantage of CUDA technology and MPI protocols, MD simulation can be applied to simulate extreme conditions that are difficult to do in physical experiments. This is a new development worthy of great attention in MD simulation. On balance, there is still huge potential for the application of GPU in MD simulation.