Core subset selection from collections hosted by seed banks, grow in importance as the number of accessions and genetic marker information rapidly increases. A data set of 20,526 single-nucleotide polymorphism (SNP) markers characterizing 7986 Mexican creole wheat landraces, was used to test 11 methods for core subset selection, through optimization criteria containing average genetic distance and genetic diversity. Allele richness was used as an additional criterion to qualify the generated core subsets. Three replications with random samples of 1500 SNP loci, each comprising a maximum of 3000 alleles, were used to perform the method evaluations through four different objective functions. The LR greedy search (LR) and LR with random first pair (LRSemi) were consistently best across all assays for maximizing the objective functions, and they performed well even for criteria not included in those functions. The Tukey's HSD (honest significant difference) multiple comparisons grouped those methods together with the sequential forward selection (SFS) and SFS with random first pair (SFSSemi) strategies as the top set of approaches. All of them are simple heuristic maximization algorithms, and outperformed two more sophisticated optimization approaches: parallel mixed replica exchange and replica exchange Monte Carlo. For their efficiency to optimize the objective functions and computing speed, the LRSemi and SFSSemi methods demonstrated to be good alternatives for core subset selection from large collections of highly homozygous accessions characterized by many biallelic markers.