There are three main hardware options for running Turbostream and we offer support for installing and using the solver on all of them.
Desktops with two or three GPUs are the most common starting point for running Turbostream simulations. Even on such relatively inexpensive hardware, it is possible to run simulations with 15-20 million grid nodes.
GPU clusters are the natural next step up from desktop machines. Turbostream is routinely run on the University of Cambridge's 128-GPU cluster for tackling simulations with hundreds of millions of grid nodes, and we have experience in creating specifications for similar machines. The solver also runs on CPU-only clusters, thereby ensuring that existing hardware can be fully utilised.
Some organisations only occasionally need access to significant hardware resources. In such cases, renting time on a GPU cluster such as the University of Cambridge's Darwin facility or the Amazon Elastic Compute Cloud (EC2) is a perfect option. We can provide help with provisioning either CPU or GPU clusters on EC2.