Graph computing models show great popularity in tracking the fraud detection problem defending merchant losses with illegal behaviors. However, current solutions mainly focus on capturing features of user historical behavior sequences with supervised learning, which makes it difficult to cope with the vagaries of fraud patterns. To gain a lot of traction, this paper proposes a novel PARallel Graph Computing model for Unsupervised Fraud Detection (PAR-GCUFD). The unsupervised graph construction and graph contrastive learning are developed to effectively identify unseen fraud patterns. Furthermore, we innovatively design a parallel strategy to accelerate the speed of training and inference. Extensive experiments on real-world industrial datasets also prove the superiority of the proposed PAR-GCUFD framework.