search

SciML / DiffEqGPU.jl

GPU-acceleration routines for DifferentialEquations.jl and the broader SciML scientific machine learning ecosystem
https://docs.sciml.ai/DiffEqGPU/stable/
MIT License
274 stars 28 forks source link

Just wonder how faster DiffEqGPU.jl is compared to DifferentialEquations.jl #204

Closed licheng0794 closed 1 year ago

licheng0794 commented 1 year ago

Any quantitative comparsion between two packages DiffEqGPU.jl and DifferentialEquations.jl? Thank you.

ChrisRackauckas commented 1 year ago

DiffEqGPU.jl is a part of DifferentialEquations.jl. It's the same interface and the same solvers. DiffEqGPU.jl is just some compilation tricks and helpers for using GPUs more effectively with DifferentialEquations.jl

You cannot put one number on it. Some cases without enough trajectories are slower: GPUs have a kernel cost overhead. Some cases with EnsembleGPUKernel are over 100x faster. It's application-dependent. We will be starting to write the paper on this library in the next month, aiming for it to be done in the next 6, and adding benchmarks to https://github.com/SciML/SciMLBenchmarks.jl, so definitely stay tuned and read those when they are ready. Until then, come up with your own benchmarks and share the results!