Since Schnorr-Euchner Sphere Decoding (SE-SD) does not guarantee a fixed throughput, the searching cycles of SE-SD should be limited for the practical implementation. Given SE-SD with runtime constraint causes degradation in performance due to the variance of searching cycles, an enhanced SE-SD architecture with a small variance of searching cycles is proposed in this paper for a multi-input multi-output(MIMO) system. Small variance in number of searching cycle is achieved by applying parallel partial Euclidean distance (PED) calculation units to the one-node-per-cycle architecture. Since the proposed architecture is able to evaluate more children nodes in a single cycle, average processing cycles and error performance are significantly improved with a per-block run-time constraint. Our proposed parallel architecture increases the complexity about two times, but it can obtain a 2 dB gain in a 4times4 16QAM system when the runtime constraint is 7 cycles.