Implementing multicomponent diffusion models in reacting-flow simulations is computationally expensive due to the challenges involved in calculating diffusion coefficients. Instead, mixture-averaged diffusion treatments are typically used to avoid these costs. However, to our knowledge, the accuracy and appropriateness of the mixture-averaged diffusion models has not been verified for three-dimensional turbulent premixed flames. In this study we propose a fast, efficient, low-memory algorithm and use that to evaluate the role of multicomponent mass diffusion in reacting-flow simulations. Direct numerical simulation of these flames is performed by implementing the Stefan-Maxwell equations in NGA. A semi-implicit algorithm decreases the computational expense of inverting the full multicomponent ordinary diffusion array while maintaining accuracy and fidelity. We demonstrate the algorithm to be stable, and its performance scales approximately with the number of species squared. We first verify the method by performing one-dimensional simulations of premixed hydrogen flames and compare with matching cases in Cantera. As an initial study of multicomponent diffusion, we simulate premixed, three-dimensional turbulent hydrogen flames, neglecting secondary Soret and Dufour effects. Simulation conditions are carefully selected to match previously published results and ensure valid comparison. Our results show that using the mixture-averaged diffusion assumption lead to a 15 % under-prediction of the normalized turbulent flame speed for premixed hydrogen air flames. This large difference in the turbulent flame speed raises questions on the appropriateness of using the mixture-averaged diffusion assumption for DNS of moderate to high Karlovitz number flames.