Approaching QUDA 1.0
The QUDA library for Lattice QCD on NVIDIA GPUs will be around for 10 years in 2018. In this time QUDA has developed from an acceleration library for solvers to a framework for developing QCD simulations. It support many different fermion discretizations, and features many algorithms like Adaptive Multigrid, deflation and Block Krylov space methods, using various techniques such as mixed-precision and communication hiding to maximise performance. We will present a current state of QUDA, discuss performance on current hardware and present a roadmap to QUDA 1.0. We also encourage discussions about user requests.
Preferred track (if multiple tracks have been selected)