Abstract In this study we investigated full-field fine-scaled thermal reservoir models with multilateral wells that are independent of grid refinements and thus allow efficient optimization of well placements. This facilitated the production optimization and well interference study in thermal modeling framework. We observed that multilateral wells with multiple sidetracks enhanced the steam injection efficiency by enlarging steam chambers. We also examined the benefit of utilizing high performance computing (HPC) in multiple-node cluster environments. Results showed that excellent scalability was achieved with up to thousands of computing CPUs and total simulation time was reduced from weeks to hours. More importantly, the study confirmed that in thermal simulation coarser grid models overestimate steam chambers leading to overestimating total field production and reserves.