The dataset contains per-matrix convergence traces for 9 widely used preconditioners evaluated on 79 SPD matrices from the SuiteSparse Matrix Collection. Each plot shows the floating-point work required for the preconditioned Conjugate Gradient method to reach a relative residual of 10‑10. The horizontal axis denotes the cumulative work expended (in FLOPs) as a function of iterations, while the vertical axis shows the corresponding residual norm. Both axes are on a log-log scale. The control (no preconditioner) is shown in blue. All configurations of the highlighted preconditioner appear in red; other tested preconditioners are shown in gray. The results span classical iterative techniques, incomplete Cholesky variants, sparse approximate inverses, and algebraic multigrid schemes, allowing direct side-by-side comparisons of preconditioner effectiveness and robustness on real-world matrices.