When comparing different methods for performing SVD in a matrix completion problem, the study found that the fast.svd method from the corpcor package was generally the most efficient, taking both computational time and accuracy into consideration. Across different matrix dimensions and missing data rates, fast.svd consistently had either the fastest runtime or was comparable to the fastest methods, while also having relatively low training error compared to other methods like irlba that had faster speeds but lower accuracy. When the matrix size was very small, the built-in R svd function performed best. But for larger matrices, fast.svd emerged as the best overall method.