The Versatile Video Coding (VVC) standard introduced several novel encoding tools for intra frame prediction, increasing the encoder complexity when compared to previous standards. Among these novelties is the MIP tool, whose acceleration has not been tackled in the literature. Therefore, an efficient parallelization of MIP prediction targeting GPU platforms is now proposed. The presented technique makes use of alternative reference samples and computes the distortion in an approximate manner to expose and potentiate massive parallelism. Moreover, the adopted prediction scheduling and memory communication were tailored by considering the GPUs' architecture and memory hierarchy. When compared with a CPU execution, this work is capable to accelerate the MIP prediction up to 105 times at the cost of a negligible coding efficiency loss of 0.284 % BD-BR.