1: \begin{abstract}

2:   

3: A number of stochastic methods developed for the calculation of

4: fermion loops are investigated and compared, in particular with

5: respect to their efficiency when implemented on Graphics Processing

6: Units (GPUs). We assess the performance of the various methods by

7: studying the convergence and statistical accuracy obtained for

8: observables that require a large number of stochastic noise vectors,

9: such as the isoscalar nucleon axial charge. The various methods are

10: also examined for the evaluation of sigma-terms where noise reduction

11: techniques specific to the twisted mass formulation can be utilized

12: thus reducing the required number of stochastic noise vectors.

13: \end{abstract}