Fig. 6: GPU performance on TDDFT calculations of GFP.

a Speed-up and parallel efficiency per number of GPUs. The ideal speed-up is indicated by the dotted diagonal line. The number of GPU FP64 cores are also provided on top of the plot. A separate linear scale plot is provided in Supplementary Fig. 4. b Wall-time to convergence per number of GPUs. Alternate units of time are also provided to aid in interpretation of the results. Decomposition into different computations are also provided.