And I will just add that performance is sensitive to parameters. The defaults try to be conservative and hypre's seem to be geared for 2D low order discretizations. GAMG is probably a bit more geared for 3D. If you are interested in looking this carefully you can run GAMG with '-info', and grep on GAMG, and send us the results and we can verify that it running OK. Also run with -log_summary and send that separately. The two solvers should be about the same speed on your problem.
On Fri, Sep 18, 2015 at 7:36 AM, Jed Brown <[email protected]> wrote: > Justin Chang <[email protected]> writes: > > I see that there's GAMG and there's Hypre's BoomerAMG (and perhaps others > > too?) > > ML > > > What exactly is the difference between these two? Do they have very > > different implementations under the hood? Does one have better > > scalability over another? Or more importantly, when would I choose one > > over the other? > > BoomerAMG is classical AMG, which is a different coarsening strategy > From GAMG and ML which use smoothed aggregation. The math is different > and the implementation is different. ML and GAMG have similar math and > there exist configurations that are equivalent. >
