mirror of https://github.com/JuliaFEM/JuliaFEM.jl.git synced 2026-09-09 12:42:19 +00:00

Files

T

Jukka Aho 6d583eb30c docs(demos): Add GPU and MPI demonstration guide

New 77-line guide documenting:
- Prerequisites (MPI and CUDA globally installed)
- Running commands for MPI communication test
- Running commands for combined GPU+MPI test
- Single-process GPU test instructions
- What gets demonstrated (type stability requirement, MPI fast transfer, real hardware)
- Success indicators and result interpretation
- Key insight: same patterns enable CPU speedup, GPU execution, and MPI efficiency

2025-11-09 10:47:32 +02:00

1.8 KiB

Raw Blame History

Running GPU and MPI Demonstrations

Prerequisites

These demonstrations require MPI and CUDA to be installed globally:

# MPI is already installed on your system
# CUDA is already installed on your system

The packages are loaded dynamically, so they don't need to be in Project.toml.

Running the Demonstrations

MPI Communication Test

Test data transfer between 2 MPI processes:

cd /home/juajukka/dev/JuliaFEM.jl
mpirun -np 2 julia benchmarks/gpu_mpi_demo.jl

Expected output:

Rank 0 sends typed arrays to Rank 1
Shows bytes transferred
Validates data integrity

GPU + MPI Combined Test

If CUDA GPU is available, runs full workflow:

mpirun -np 2 julia benchmarks/gpu_mpi_demo.jl

Expected output:

Detects GPU (if available)
Compiles type-stable kernel for GPU
Executes on real hardware
Transfers results via MPI

Single-Process GPU Test

To test GPU without MPI:

julia benchmarks/gpu_only_demo.jl

What Gets Demonstrated

Type Stability Requirement
- Matrix{Float64} transfers to GPU ✅
- Dict{String,Any} would FAIL GPU compilation ❌
MPI Fast Transfer
- Typed arrays: Fast buffer transfer (memcpy)
- Mixed types: Slow serialization (~100× slower)
Real Hardware Execution
- Actual CUDA kernel compilation and execution
- Actual MPI inter-process communication
- No mocks, no simulation

Interpreting Results

Success indicators:

✅ "MPI communication successful" - Type-stable data transferred
✅ "GPU execution successful" - Kernel compiled and ran on GPU
✅ "Combined GPU+MPI workflow successful" - End-to-end validated

The key insight: The same type-stable patterns that give 9-92× CPU speedup also enable GPU execution and efficient MPI communication.

1.8 KiB Raw Blame History Unescape Escape