Since the introduction of General-Purpose GPU computing, significant increase has been made in performance for regular path tracing. However, more advanced versions of path tracing such as BiDirectional Path Tracing (BDPT) and Energy Redistribution Path Tracing (ERPT) have not been implemented successfully so far due to their stochastic sampling characteristics. The goal of this thesis is to find efficient GPU implementations for these unbiased physically based rendering methods. In this thesis improved streaming versions of these algorithms have been developed that better exploit ray coherence, reduce memory-footprint, and improve convergence, in order to make the algorithms more feasible for the GPU. The performance of the GPU versions is compared with the CPU versions and it is shown that the convergence characteristics of the original methods are preserved in our GPU implementations, while the processing has been speeded up with an order of magnitude.