GPU加速在第一性原理输运研究中的应用

分子器件中的第一性原理输运计算比普通的密度泛函计算要慢很多，其最根本原因在于密度矩阵计算方法的不同。本文将 GPU 加速应用于第一性原理的输运计算，重点实现相关矩阵运算的加速。测试结果表明，在单次迭代中，对于较大的体系，相对于调用 MKL 库，密度矩阵的运算速度在单个 Tesla M2090 可以提高一个数量级以上，在 Tesla K20m 上则可以提高 20 倍以上，从而取得了很好的加速效果，而且体系越大，加速效果越好。

GPU Acceleration for First Principles Transport Calculations

First principles transport calculations in large molecular devices are generally very slow, especially when compared with the density functional calculations for electronic structures. For transport calculations, the most time consuming part is the calculation of the density matrix. In this work, GPU acceleration is applied to the first principles transport calculations and we focus mainly on the speedup for the matrix operations. Tests show that, in a single iteration, there will be a speedup of more than one order of magnitude for large systems on a single Tesla M2090 and more than 20 times on a single Tesla K20m, compared with the MKL library. Thus a very good acceleration is achieved, and the bigger the system is, the better speedup we will achieve.