Various methods have been devised to multiply N×N matrices using O(NE) arithmetic operations when N→∞, for various exponents E with 2<E≤3.However, in practice, the schemes with least known E are not the best, because they only start to win for infeasibly large N. Prior literature has unhealthily been fixated on E alone, i.e. on the asymptotic large-N performance alone. To address that, we propose investigating, for each method, not merely its E, but also its "breakeven N," meaning the least N causing that method to use fewer than the obvious algorithm's N3 bilinear multiplications [or fewer than Strassen'sO(N2.807355) scheme]. The set of (E,logN) datapoints then form a subset of the infinite rectangle (2,3]×[0,∞). Part of that rectangle is filled with datapoints, while another part contains none. What is of interest is the curve delineating the boundary between those two regions.