juliacon-2025-semidiscretization-benchmark

This repository provides the source code and instructions to reproduce the CPU and GPU semidiscretization performance benchmarks presented in the JuliaCon 2025 talk TrixiCUDA.jl: CUDA Support for Solving Hyperbolic PDEs on GPU.

Reproduce the Benchmarks

First, clone the repository to your local machine:

git clone https://github.com/trixi-gpu/juliacon-2025-semidiscretization-benchmark.git

Once the clone completes, move into the source directory:

cd juliacon-2025-semidiscretization-benchmark/code

Next, start Julia with the project environment and install all required packages:

julia --project=.
] instantiate

After dependencies have been fetched, precompile the core module:

include("src/TrixiCUDA.jl")

With the module precompiled, run the full benchmark suite:

include("benchmark/benchmark.jl")

You'll see progress and information printed to the terminal as each example runs, for example:

julia> include("benchmark/benchmark.jl")
=============== Mode: Fixed DOFs Approach ===============
[1] Linear Advection Equation (Basic), running examples_fixed_dofs\advection_basic_1d.jl
[ Info: Benchmarking rhs! on CPU
[ Info: Benchmarking rhs! on GPU
[2] Linear Advection Equation (Basic), running examples_fixed_dofs\advection_basic_2d.jl
[ Info: Benchmarking rhs! on CPU
[ Info: Benchmarking rhs! on GPU
[3] Linear Advection Equation (Basic), running examples_fixed_dofs\advection_basic_3d.jl
[ Info: Benchmarking rhs! on CPU
[ Info: Benchmarking rhs! on GPU
Saved: ..\plots\fixed_dofs_approach_linear_advection_equation_basic.png
...
=============== Mode: Variable DOFs Approach ===============
[1] Linear Advection Equation (Basic), running examples_variable_dofs\advection_basic_1d.jl
[ Info: Benchmarking rhs! on CPU
[ Info: Benchmarking rhs! on GPU
[2] Linear Advection Equation (Basic), running examples_variable_dofs\advection_basic_2d.jl
[ Info: Benchmarking rhs! on CPU
[ Info: Benchmarking rhs! on GPU
[3] Linear Advection Equation (Basic), running examples_variable_dofs\advection_basic_3d.jl
[ Info: Benchmarking rhs! on CPU
[ Info: Benchmarking rhs! on GPU
Saved: ..\plots\variable_dofs_approach_linear_advection_equation_basic.png
...

Note that benchmarks are conducted using two approaches: (1) fixed DOFs and (2) variable DOFs.

Once it finishes, all generated figures will be saved in the plots directory.

Sample Benchmark Results

Sample benchmark figures are available in the figures directory, and a complete report can be found here.

Sample benchmarks were conducted on an NVIDIA RTX 4060 (Ada architecture). The software environment was as follows:

# Julia version information
julia> versioninfo()
Julia Version 1.10.9
Commit 5595d20a28 (2025-03-10)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 20× 13th Gen Intel® Core™ i9-13900H
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, goldmont)

# CUDA version information
julia> using CUDA; CUDA.versioninfo()
CUDA runtime 12.6 (artifact installation)
CUDA driver 12.8
NVIDIA driver 572.42.0

CUDA libraries:
  CUBLAS    12.6.4
  CURAND    10.3.7
  CUFFT     11.3.0
  CUSOLVER  11.7.1
  CUSPARSE  12.5.4
  CUPTI     2024.3.2 (API 24.0.0)
  NVML      12.0.0+572.42

Julia packages:
  CUDA             5.6.1
  CUDA_Driver_jll  0.10.4+0
  CUDA_Runtime_jll 0.15.5+0

Toolchain:
  Julia  1.10.9
  LLVM   15.0.7

Device:
  0: NVIDIA GeForce RTX 4060 Laptop GPU (sm_89, 6.706 GiB / 7.996 GiB available)

Note that results may vary on different GPU models, Julia releases, and CUDA versions. Please take these benchmark results for reference only.

Acknowledgment

Many thanks to the following professors for their guidance on this project:

Prof. Hendrik Ranocha (Johannes Gutenberg University Mainz, Germany)
Prof. Jesse Chan (University of Texas at Austin, U.S.)
Prof. Michael Schlottke-Lakemper (University of Augsburg, Germany)

If you have any questions or find any issues, please contact the author, Huiyu Xie.