Modern video encoders, such as the recently proposed AV1 and VVC, offer significant encoding gains at the cost of a corresponding increase of the computational effort. This is the case of the adopted intra prediction techniques, comprehending an increased number of prediction modes and range. To mitigate this computational cost, the presented work proposes a new mode-adaptive algorithm that significantly reduces the number of SAD/SSE operations during intra prediction, by generating an optimized subsampling pattern adaptive to each prediction mode. The method can be applied to any video codec and, when applied to AV1, it led to an encoding time reduction and BD-BR impact of 15.36% and 0.6%, respectively, or 7.97% and –0.02%, depending on the selected subsampling parameters. When implemented in hardware, the proposed technique provides an effective reduction as high as 75% of both the area and power on the modified distortion calculation module.