The non-commutativity error is generally compensated by the multi-subsample rotation vector method, but the performance of the traditional multi-subsample rotation vector method is degraded under the condition of large half cone angle or other more complicated angular motion. The new multi-subsample quaternion algorithm and the direction cosine algorithm do not directly consider the non-commutativity error, and the two algorithms directly solve the attitude differential equation. In order to comprehensively compare the performance of these algorithms, a simulation analysis is carried out under conical motion and large angle maneuvering conditions. Two data conditions, ideal and non-ideal samplingare used to transform different sampling frequencies, and comparison is made to the computational burden of the algorithms. The results show that under ideal conditions, the periodic error of the new algorithm is smaller than that of the rotation vector method, and the accuracy of the new algorithm is improved significantly with the increase of the number of subsamples. By comparing ideal sampling with non-ideal sampling, it can be seen that the accuracy of the algorithm differs by 4~5 orders of magnitude. Each of the three algorithms has its own advantages and disadvantages. Only by fully compensating the inertial device data can the accuracy potential of the algorithm be fully realized.