Triticale, a hybrid species between wheat and rye, is one of the newest additions to the plant kingdom with a very short history of improvement. It has very limited genomic resources because of its large and complex genome. Objectives of this study were to generate dense marker data, understand genetic diversity, population structure, linkage disequilibrium (LD), and estimate accuracies of commonly used genomic selection (GS) models on forage yield of triticale. Genotyping-by-sequencing (GBS), using PstI and MspI restriction enzymes for reducing genome complexity, was performed on a triticale diversity panel (n = 289). After filtering for biallelic loci with more than 70% genome coverage, and minor allele frequency (MAF) > 0.05, de novo variant calling identified 16,378 single nucleotide polymorphism (SNP) markers. Sequences of these variants were mapped to wheat and rye reference genomes to infer their homologous groups and chromosome positions. About 45% (7,430), and 58% (9,500) of the de novo identified SNPs were mapped to the wheat and rye reference genomes, respectively. Interestingly, 28.9% (2,151) of the 7,430 SNPs were mapped to the D genome of hexaploid wheat, indicating substantial substitution of the R genome with D genome in cultivated triticale. About 27% of marker pairs were in significant LD with an average r2 > 0.18 (P < 0.05). Genome-wide LD declined rapidly to r2 < 0.1 beyond 10 kb physical distance. The three sub-genomes (A, B, and R) showed comparable LD decay patterns. Genetic diversity and population structure analyses identified five distinct clusters. Genotype grouping did not follow prior winter vs. spring type classification. However, one of the clusters was largely dominated by winter triticale. Genomic selection accuracies were estimated for forage yield using three commonly used models with different training population sizes and marker densities. Genomic selection accuracy increased with increasing training population size while gain in accuracy tended to plateau with marker densities of 2,000 SNPs or more. Average GS accuracy was about 0.52, indicating the potential of using GS in triticale forage yield improvement.