Early Experiences Porting the NAMD and VMD Molecular Simulation and Analysis Software to GPU-Accelerated OpenPOWER...

Area of Science:

• Computational biology
• Biophysics
• High-performance computing

Background:

• All-atom molecular dynamics (MD) simulations are crucial for studying biomolecular structure and dynamics.
• These simulations require significant computational resources and pose challenges in preparation, execution, and analysis.
• Optimizing MD simulations on advanced hardware is essential for scientific discovery.

Purpose of the Study:

• To report early experiences porting NAMD and VMD to GPU-accelerated OpenPOWER platforms.
• To evaluate the performance of molecular modeling kernels on these new hardware architectures.
• To assess the benefits of specific POWER8 features for MD simulations.

Main Methods:

• Porting NAMD and VMD software to GPU-accelerated OpenPOWER hardware.
• Comparing compiler autovectorization with hand-coded vector intrinsics on POWER8 CPUs.
• Analyzing performance gains from POWER8's 8-way Simultaneous Multithreading (SMT).
• Evaluating GPU-accelerated molecular modeling kernels.

Main Results:

• Successful porting of NAMD and VMD to OpenPOWER platforms.
• Demonstrated performance benefits from POWER8 architectural features, including 8-way SMT.
• Comparison of autovectorization and hand-coded intrinsics for the POWER8 CPU.
• Performance evaluation of GPU-accelerated kernels on the new hardware.

Conclusions:

• GPU-accelerated OpenPOWER platforms offer a viable and performant environment for large-scale biomolecular simulations.
• Specific architectural features of POWER8 can be leveraged to enhance molecular modeling tasks.
• Further optimization and exploration of these platforms are warranted for advancing computational biophysics.