Cardiac muscle contraction is primarily driven by excitation waves travelling through myocytes within the tissue, and in silico studies of their dynamics have contributed to clinical advancements. While state-of-the-art computer models, like the bidomain model, average over hundreds of myocytes, experimental studies suggest that e.g. arrhythmias can originate from microscopic tissue heterogeneities. To address this, we employ the extracellular-membrane-intracellular (EMI) model, which explicitly represents myocytes in the discretised space. We implemented the EMI model in FEniCSx and solved it on a highly resolved mesh. We then progressively coarsened the mesh using mmg3d, varying element lengths, as well as Hausdorff distances and gradation, where the latter two control the myocyte shape preservation. Since, for example, extracellular domains, can be coarser than the membranes, where relevant processes for the excitation dynamics occur, we also considered different resolution requirements for these regions. For each remeshing parameter set, we determined the number of degrees of freedom (DOFs), which influence the compute time. Finally, we determined the smallest number of DOFs that preserves quantities characterising excitation dynamics, such as conduction velocity and electrograms, from the original mesh dynamics with a maximal deviation (e.g. 1%). This trade-off between mesh resolution and runtime provides an estimate for the number of representable myocytes given a computational time constraint, which in turn can then be used to design large, clinically motivated EMI model simulations with high energy efficiency.