Fig. 3: Multi-GPU mapping of numerical quadrature for exchange–correlation. | npj Computational Materials

Fig. 3: Multi-GPU mapping of numerical quadrature for exchange–correlation.

From: Kohn–Sham time-dependent density functional theory with Tamm–Dancoff approximation on massively parallel GPUs

Fig. 3

a Schematic illustration of the partitioning scheme of the quadrature points. Data were generated for water with (10, 14)-grid and partitioned into 44 boxes. See Method for the grid definition. Points from two select boxes are shown in magenta/green. Two s-type AOs are displayed in gray; the boundary implies their significance threshold. Significance table with a true or false value with the related elements entailing the overlap between boxes and AOs is given at the bottom. b Evenly-distributed mapping of boxes to multi-GPUs. Each grid, comprising a one-dimensional array of blocks corresponding to boxes, is scheduled to a single GPU. Each block contains threads corresponding to points with the predetermined list of AO contribution in accordance with the significance table.

Back to article page