Pose estimation of known objects is fundamental to tasks such as robotic\ngrasping and manipulation. The need for reliable grasping imposes stringent\naccuracy requirements on pose estimation in cluttered, occluded scenes in\ndynamic environments. Modern methods employ large sets of training data to\nlearn features in order to find correspondence between 3D models and observed\ndata. However these methods require extensive annotation of ground truth poses.\nAn alternative is to use algorithms that search for the best explanation of the\nobserved scene in a space of possible rendered scenes. A recently developed\nalgorithm, PERCH (PErception Via SeaRCH) does so by using depth data to\nconverge to a globally optimum solution using a search over a specially\nconstructed tree. While PERCH offers strong guarantees on accuracy, the current\nformulation suffers from low scalability owing to its high runtime. In\naddition, the sole reliance on depth data for pose estimation restricts the\nalgorithm to scenes where no two objects have the same shape. In this work, we\npropose PERCH 2.0, a novel perception via search strategy that takes advantage\nof GPU acceleration and RGB data. We show that our approach can achieve a\nspeedup of 100x over PERCH, as well as better accuracy than the\nstate-of-the-art data-driven approaches on 6-DoF pose estimation without the\nneed for annotating ground truth poses in the training data. Our code and video\nare available at https://sbpl-cruz.github.io/perception/.\n