The Kokkos Lectures: Module 6 Q&A

General Kokkos

Will Kokkos arrays in Cuda space work with Pyhton, i.e. cupy?

• Yes, that should work.

What is the minimal supported Kokkos version to support Python Interop?

• 3.0

Why was the HostPinnedSpace generally faster than the standard CudaSpace for the MPI comm? Is this specific to the Spectrum MPI implementation on Summit or sth to generally expect?

• The best choice for the best MPI communication strategies varies wildly. It is advised to write code that supports all cases and then measure which variant is fatest for a specific application/hardware combination.

Is GPU-aware MPI still required for HostPinned buffers?

• HostPinned buffer still lives on the CPU so GPU-aware MPI is not necessary.

What kind of precautions need to be taken with Kokkos against oversubscribing with MPI+Kokkos?

• Set OMP_NUM_THREADS appropriately if using OpenMP.

Is the remote space on the default device (host/cuda/etc.)?

• Per default, NVSHMEM runs on the Cuda Execution Space, SHMEM and MPI One-Sided runs on the host execution space.