ESPRESO solver is a massively parallel sparse linear solver designed to take full advantage of todays most powerful petascale supercomputers such as Tianhe 2 with Intel Xeon Phi coprocessors, Titan with Nvidia GPUs or Sequoia – IBM’s CPU only MPP (Massively Parallel Processor) system, see TOP500 for more details. Of course ESPRESO will efficiently run on cluster of any size as well. In addition the next generation of DOE (U.S. Department of Energy) supercomputers such as Summit (Oak Ridge National Lab) or Aurora (Argonne National Lab) will also contain either GPU or well Xeon Phi chips.
massively parallel sparse linear solver designed to take full advantage of todays most powerful petascale supercomputers
|US Department of Energy CORAL Supercomputers|
|CPU Architecture||Intel Xeon||Intel Xeon||IBM POWER9||IBM POWER9|
|Accelerator Architecture||Intel Xeon Phi||Intel Xeon Phi||NVIDIA Volta||NVIDIA Volta|
|Performance (RPEAK)||180 PFLOPS||8.5 PFLOPS||130-300 PFLOPS||100+ PFLOPS|
|Laboratory||Argonne||Argonne||Oak Ridge||Lawrence Livermore|
In order to provide a tool that runs optimally on all these major platforms, we are concurrently developing three major versions: H for MPP systems and clusters, GPU for GPU accelerated systems, and MIC for Xeon Phi systems. In terms of implemented algorithms the solver is using either Total FETI or Hybrid FETI algorithm depending on problem size.