跳到正文
NVIDIA Technical Blog· Erica Tsai·· 2 小时前AI 评分20

NVIDIA 如何用 K 点批处理与混合精度解决平面波 DFT 的 GPU 瓶颈

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

AI 导读

NVIDIA 针对 VASP 平面波 DFT 训练 MLIP 时的 GPU 瓶颈,提出 K 点批处理与混合精度两种优化手段。训练下一代 MLIP 需数千万至数亿次 DFT 计算提供能量、力与应力真值,而 VASP 是生成这类数据的主流实现。

正文 · 原文

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

来源:NVIDIA Technical Blog · developer.nvidia.com