K-best MIMO detection technique is the prominent method of simplifying the detection complexity in MIMO systems while maintaining BER performance comparable with the optimum maximum-likelihood (ML) detection technique. However, sorting the candidate nodes in the tree search of the conventional K-best detection can take a significant number of cycles which would reduce the achievable data rate of the detector. In order to reduce this delay, and keep high performance at the same time, we propose using a novel sort-free based MIMO detector which avoids the demanding sorting step. Moreover, this detector utilizes a novel modified real-valued decomposition (M-RVD) ordering that, when compared to the conventional real valued decomposition scheme, can improve the BER performance at no extra computational cost. We show that our proposed detector can outperform the conventional K-best detector with a smaller combination of computation and latency requirements.