Skip to content
Live
NVIDIA Technical Blog· Erica Tsai·· 6 hours ago35

Tackling GPU Bottlenecks in Plane-Wave DFT with K-Point Batching and Mixed Precision

Tackling GPU Bottlenecks in Plane-Wave DFT with K-Point Batching and Mixed Precision

Brief overview

NVIDIA employs K-point batch processing and mixed precision technology to optimize plane wave DFT calculations, significantly enhancing the GPU performance required for training next-generation machine learning atomic interatomic potentials (MLIP) using density functional theory (DFT) data. This technology boosts computational speed by 3.5 times while maintaining high precision, suitable for generating tens of millions of energy, force, and stress data.

Article

Training a next-generation machine learning interatomic potential (MLIP) requires ground truth: tens to hundreds of millions of density functional theory (DFT)...

Training a next-generation machine learning interatomic potential (MLIP) requires ground truth: tens to hundreds of millions of density functional theory (DFT) calculations of energies, forces, and stresses across diverse structures and chemistries. Plane-wave DFT, with VASP as its dominant implementation, is the gold standard for generating that data. It is accurate…

Source

Source: NVIDIA Technical Blog · developer.nvidia.com