Modeling thrombus growth in pathological flows allows evaluation of risk under patient-specific pharmacological, hematological, and hemodynamical conditions. To this end, we have developed a 3D multiscale framework for the prediction of thrombus growth under flow on a spatially resolved surface presenting collagen and tissue factor (TF). The multiscale framework is composed of four coupled modules: a Neural Network (NN) that accounts for platelet calcium signaling, a Lattice Kinetic Monte Carlo (LKMC) simulation for tracking platelet positions, a Finite Volume Method (FVM) simulator for solving convection-diffusion-reaction equations describing soluble agonist release and transport, and a Lattice Boltzmann (LB) flow solver for computing the blood flow field over the growing thrombus. A reduced model of the coagulation cascade was embedded into the framework to account for TF-driven thrombin production. The 3D model was able to effectively capture the evolution and morphology of the growing thrombus when compared against in vitro microfluidics experiments of whole blood perfusion with various antiplatelet agents. The generalizability of the 3D multiscale solver enabled simulations of important clinical situations, such as cylindrical blood vessels and acute flow narrowing (stenosis). Enhanced platelet-platelet bonding at pathologically high shear rates (e.g., von Willebrand factor unfolding) was required for accurately describing thrombus growth in stenotic flows. To enable large computations in a reasonable amount of time, each module within the multiscale framework was individually parallelized. Parallelization was achieved by developing in-house parallel routines for NN and LKMC, while the open-source libraries OpenFOAM and Palabos were used for FVM and LB, respectively. The parallelized model was validated against a reference serial version for accuracy, demonstrating comparable results for both microfluidic and stenotic arterial clotting conditions. Moreover, the parallelized framework was shown to scale essentially linearly on up to 64 cores for a benchmark simulation of thrombus growth in a stenotic vessel of size ~1 mm. Overall, the parallelized multiscale framework allows consideration of patient-specific platelet signaling and vascular geometry for the prediction of thrombotic episodes.