![]() NVIDIA Topology-Aware GPU Selection (NVTAGS) intelligently and automatically assigns GPUs to MPI processes, thereby reducing overall GPU-to-GPU communication time. Users can expect better performance from this container if they ensure that such GPU communication occurs along the strongest available GPU links. Parallelizing over multiple GPUs is done by utilizing the QMP library which provides a thin abstraction over MPI for most QCD oriented communications.Ĭhroma spends a considerable portion of its runtime in MPI-based, GPU-to-GPU communications. QDP-JIT is especially useful for gauge-field generation, in which expression templates are used to generate code dynamically using the JIT library of the LLVM Compiler infrastructure. Chroma can be built over the regular QDP++ package or over a version of QDP++ called QDP-JIT. This provides highly optimized solvers for GPU LQCD. Provides lattice wide types and expressions, using expression templates, to allow straightforward encoding of LQCD equations.Ĭhroma code can be built using the QUDA Library for Lattice QCD on GPU supported systems which includes an implementation of Adaptive Aggregation Multigrid. The Chroma package provides a toolbox of executables that carry out lattice Quantum Chromodynamics calculations (LQCD).Ĭhroma is built on top of the QDP++ (QCD Data Parallel Layer) which provides an abstract data parallel view of the lattice and ![]()
0 Comments
Leave a Reply. |