Two simulated data sets and one commercial data set were used to evaluate computing options for models in which the effects attributable to QTL were fit as covariables. The simulated data sets included records on 24,000 animals for 10 traits. Data sets 1 and 2 were simulated with low and high correlations among traits, respectively. The model included an overall mean, 160 covariables as effects attributable to QTL, the random animal genetic effect, and the random residual effect. A commercial data set included records on approximately 110,000 animals for 11 growth, reproduction, and other traits. The model included the effects usually fitted for these traits as well as 116 covariables as effects attributable to QTL; models including the number of covariables varied by trait. Initial computing was by the BLUP90IOD program, which applies iterations on data by using a preconditioned conjugate gradient algorithm with a diagonal preconditioner. Modifications included adding block preconditioners for effects attributable to QTL (BQ) and for traits (BT). With the simulated data sets and the original program, one-trait analyses without the covariables took 7 s, whereas the 10-trait analyses with the covariables took 15 min for a data set with low correlations and 1 h 40 m for a data set with high correlations. The BQ improved the convergence rate but increased the computing time. The BT decreased the computing time from 1.5 times (low correlations) to 7 times (high correlation) at a cost of greater memory requirements. For the commercial data and the complete model, computing took 10.3 h with the unmodified program and was reduced to 6 h with BT. Relative changes in computing time and convergence rate with the commercial data set were close to those of the simulated data set, with low correlations among the traits. The BQ decreased the number of rounds by less than expected. Genetic evaluation with a large number of effects attributable to QTL fit as covariables is feasible.