# Multi-GPU Gradient Average Sample Requirements

# MPI Python bindings for distributed communication
mpi4py>=3.1.4

# GPU array library (NumPy-compatible arrays on CUDA)
# Use cupy-cuda11x, cupy-cuda12x, or cupy-cuda13x depending on your CUDA version
cupy-cuda13x>=14.0.0

# CUDA Python bindings (low-level CUDA driver API)
cuda-python>=13.0.0

# cuda.core - Modern Python interface for CUDA
# Provides Program, LaunchConfig, Device, and launch APIs
cuda-core>=1.0.0

# Note: This sample uses host-staging for MPI communication
# Standard MPI installation is sufficient (no CUDA-aware MPI required)
# Install MPI using system package manager:
#   Ubuntu/Debian: sudo apt-get install openmpi-bin libopenmpi-dev
#   Or build from source: https://www.open-mpi.org/software/ompi/
