GPU Accelerated
MD Simulation

GPUMD is a highly efficient open-source molecular dynamics simulation software. Developed natively with CUDA/HIP, it achieves extreme simulation efficiency on GPUs and DCUs. GPUMD supports various empirical potentials and NEP machine learning potentials, a single GPU delivers a computing speed of tens of millions of atom-steps per second.

• Extreme Performance: Built for GPU/DCU, efficiently handles systems with tens of millions of atoms.
• Easy to Use: Comprehensive docs, rich tools, active community.

NEP is a machine learning potential function natively supported by GPUMD, based on neuro-evolution algorithms. It predicts energy, force, and virial with ab initio accuracy at a cost close to empirical potentials. NEP is suitable for large-scale MD simulations of complex materials.

• Simulation Efficiency: A single GPU delivers a computing speed of tens of millions of atom-steps per second.
• Data Efficiency: High quality potentials from small datasets.