CPU GPU Architecture

Titan uses NVIDIA graphics cards and CUDA to accelerate most of the computations needed to run mass-spring simulations through parallelization. Due to the specific requirements set by CUDA, a specific architecture was designed to enable a deeper level of in the parallelization and allow the user to run non-Titan code in parallel to the simulations.

The following diagram shows a comprehensive view of the model:

In particular Titan code runs and interfaces 3 concurrent computation processes, 2 within the CPU and 1 within the GPU. These are the CPU user thread, CPU kernel thread and GPU processes. For more information about these individual processes and their design, check out the CPU and GPU sections.