DOE OASIS DAV SDK project supported by the DOE PESO project.
One of the primary challenges on Perlmutter is handling the compiler wrapper for the NVHPC compilers, in particular how they wrap MPI and CUDA.
Info | GCC Toolchain | NVHPC Toolchain |
---|---|---|
Version | v11.2 | v22.4 |
Cray MPICH | v8.2.3 | v8.2.3 |
CUDA | v11.3 | v11.3 |
Project | GCC | GCC + CUDA | NVHPC + CUDA |
---|---|---|---|
ADIOS2 | β | N/A | N/A |
HDF5 | β | N/A | N/A |
PNetCDF | β | N/A | N/A |
Ascent | β | (π« π«) | (π« π«) |
DIY | (π) | (π) | (π) |
ParaView | β | (π) | (π) |
VisIt | π | (π) | (π) |
VTK-m | β | β | (π) |
ZFP | β | β | (π) |
β - Verified π - In Progress π« - Broken N/A - Not Applicable
Ascent - Blocked by build errors in RAJA package.
Ascent - The way MPI is set up on Perlmutter conflicts with Ascentβs spack recipe and CMake. Fixes for this are being developed.
DIY - Not yet in DAV SDK.
ParaView - ParaView VTK-m using CUDA.
VisIt - VisIt utilizes a VTK version locked to a Python that is not compatible with the Python requirements of PyH5, the HDF5 python interface used by Cinema.
VisIt - VTK-m enabled GPU support for CUDA and ROCm is available VisIt, but is not officially tested as part of the DAV-SDK due to (HDF5 conflict) only recently being resolved.
VTK-m - VTK-M NVHPC using CUDA.
VTK-m - VTK-M NVHPC using CUDA.