This release contains support for CUDA 11 and support for Pascal, Volta, and
Ampere cards. The code offloads to one GPU per MPI rank. This requires running
as many MPI ranks per node as there are GPUs available on a node.
It was compiled against OpenMPI 4.0 and GCC 4.8.5.
This release contains support for CUDA 10 and the new Tesla cards with the Turing chip (sm 7.5). The binary also works for Kepler (sm 3.5), Maxwell (sm 5.0), Pascal (sm 6.0), and Volta chip (sm 7.5).
It was compiled against OpenMPI 3.1 and GCC 4.8.5.
Switched the output format for reporting the results from YAML to a basic
line-oriented key-value format with nested naming scheme for the keys.
Added faster search for optimal 3D grid partitioning of a given integer that
does not require combinatorial search through the all 3-set partitioning of
the prime factors.
Closed the outstanding bugs reported as issues on HPCG's Github project page
and incorporated the fixes in the source code.
This release contains additional optimizations that improve
performance for SC17 runs.
This new binary (dated Oct 8, 2017) supports CUDA 9 and the new Tesla
cards with the Volta chip (sm 7.0). The binary also works for Kepler
(sm 3.5), Maxwell (sm 5.0), and Pascal (sm 6.0).
It was compiled against Open MPI 1.10.2 and GCC 4.8.5.
