In this contribution, adaptive sub-block MV clip is proposed for affine blocks to reduce worst case memory bandwidth while keeping coding efficiency based on constraining max difference among 4x4 level MVs within an 8x8 sub-blocks to no larger than 1 pixel. It is further proposed to use different thresholds for x and y components for better trade-off. It is reported 0.16%/0.06%/0.03%, 0.09%/0.03%/0.01%, 0.11%/0.03%/0.03%, and 0.01%/-0.02%/0.00% luma BD-rate increase over VTM-3.0 for RA, LDB, and LDP when {Tx=0, Ty=0}, {Tx=1, Ty=0}, {Tx=0, Ty=1}, and {Tx=1, Ty=1}, respectively. With the proposed methods, the worst case memory bandwidth for affine blocks is reportedly reduced to 7.03, 7.5, 7.5 and 8 luma samples/pixel when {Tx=0, Ty=0}, {Tx=1, Ty=0}, {Tx=0, Ty=1}, and {Tx=1, Ty=1}, respectively. Note that 7.03 luma samples/pixel is the memory bandwidth for regular 8x8 bi-predicted blocks. SIMD implementation of the proposed method is also provided.