Looking for MG solver for sparse matrix on GPU


I’m looking for an algebraic multigrid solver that run on GPU. The AlgebraicMultigrid.jl package is not compatible with CUDA and I can’t find other package. I could use the AMGX library in Cuda and C++ but then I’ll have to rewrite a large part of my code.

Does somebody know a julia package for this ?

Thank you

I am not aware of a Julia package for this. You can call the C++ solver from Julia. I have implemented a baby GMG in Julia but I guess that your AMG choice is strongly constrained.

On juliahub I found AMGX.jl


OMG ! Thank you so much ! I don’t know why I did not try to type AMGX.jl !

1 Like

Hi Ludovic,

Did you try to use the AMGX.jl? Your impressions, if you could share?


Hi Niceno,

I did not use this package, I changed the description of the problem and now use a simple iterative solver.



1 Like

But still something on GPUs?

yes, I just made a very simple iterative Jacobi solver, nothing optimized but it is ok for me now.

1 Like