Dr Saul Cohen (NVIDIA)
This year, 2018, the QUDA library for Lattice QCD on NVIDIA GPUs celebrates its tenth birthday. QUDA has evolved from an acceleration library for solvers into an open-source framework for developing QCD simulations. It supports many different fermion discretizations, and features algorithms like adaptive multigrid, deflation and block Krylov-space methods. QUDA uses various techniques such as mixed precision and communication hiding to maximize performance. We will present the current state of QUDA and discuss performance on current hardware. We strongly encourage discussions about user requests and contributions.