Pseudo-continuous arterial spin labeling combined with 3D segmented readouts is recommended for acquiring ASL perfusion data. However, the total number of k-space encodings limits the trade-off between motion sensitivity and image blurring. To tackle this problem we implemented an accelerated 3D-GRASE sequence with a time-dependent 2D-CAIPIRINHA sampling pattern to increase the temporal incoherence between averages or PLDs. High quality images can be gained from the under-sampled time series by a variational image reconstruction approach with total-generalized-variation (TGV) regularization in space and time. This allows acquisition of single-shot 3mm isotropic ASL data with whole brain coverage within 1min22sec.