mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-05-14 14:06:53 +08:00
aeab82ff30
- Added Python samples for CUDA Python 1.0 release - Renamed top-level `Samples` directory to `cpp` to accommodate Python samples.
22 lines
777 B
Plaintext
22 lines
777 B
Plaintext
# 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>=13.0.0
|
|
|
|
# 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>=0.6.0
|
|
|
|
# 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/
|