Simulate magnetic textures — and differentiate through them.

02 · Run online

Describe a simulation, get a report

Write what you want to simulate, in English or German. An assistant turns it into a jaxmag run: one of the templates below, or a script written for your case (other geometries and materials, several phases, custom observables such as the skyrmion number, extra plots). It rents a GPU, runs the job and returns a PDF report; afterwards it can analyse the stored results on request. Every run keeps its own folder with the exact script, inputs, outputs, report and log (Files panel, one zip).

03 · Meshing service

Turn your geometry into a mesh

Upload a measured topography or an existing mesh and get back a tetrahedral mesh for jaxmag.fe, built with gmsh — with a preview, the element count and the curvature-induced DMI to expect from it.

Then use it locally

# the download contains topo_verts.npy, topo_tets.npy and mesh.msh
import numpy as np
from jaxmag.fe import FEMesh, fix_orientations

v = np.load('topo_verts.npy'); t = np.load('topo_tets.npy').astype('int32')
mesh = FEMesh(v, fix_orientations(v, t), mu0_Ms=4e-7*np.pi*860e3, scale_len=1e-9)

04 · Viewer

Watch the magnetisation move

Every simulation keeps 24 snapshots of the full magnetisation. Play them back, scrub through time, switch between the components and read the topological charge as it changes.

05 · Files

Everything you have made here

Reports from your simulations, meshes from your uploads, and the jaxmag source bundle — in one place, scoped to this browser.

06 · Install & use

Get jaxmag running

Install from GitLab, or download the self-contained bundle with vendored dependencies, every example and the documentation.

# from GitLab (pulls jax-ode and jaxfmm automatically)
pip install git+https://gitlab.com/dietersuess1/jaxmag.git

# or from the downloaded bundle, offline
pip install "jax[cuda12]"                       # or plain "jax" for CPU
pip install ./vendor/jax-ode ./vendor/jaxfmm .
python examples/sp4_fd.py                        # muMAG standard problem 4

Download jaxmag-bundle.zip browse the repository ↗

What the bundle documents

A skyrmion in ten lines

import jaxmag as jm

# the only backend-specific line (jm.fe.make_mesh / jm.fe.from_gmsh for tets + FMM):
mesh   = jm.fd.make_mesh(lx=100e-9, ly=100e-9, lz=2e-9, dx=2e-9)

fields = [jm.make_exchange(mesh, A=13e-12, Ms=860e3),
          jm.make_dmi(mesh, D=3e-3, Ms=860e3),
          jm.make_demag(mesh, Ms=860e3)]

m_sky  = jm.relax(jm.init_m(mesh, (0, 0, 1)), fields)   # Néel skyrmion

07 · Benchmarks

Measured, on real hardware

Pure integration time, measured on the MUSICA cluster (ASC Vienna) on 16–17 October 2026 with the scripts in jaxmag_paper/; setup (demag kernel or FMM tree, JIT compilation) is measured separately and not included, for jaxmag and mumax3 alike (mumax3: marker-file timestamps around run(), its own relaxation excluded). The 600³ cube uses the distributed FFT with the native NCCL all-to-all; the step counts of jaxmag and mumax3 agree within 3 % (stability-limited at 1 nm), so the gain is the single stray-field evaluation per step instead of six.

Cross-code comparison on rented GPUs (vast.ai, 12–13 October 2026)

One right-hand-side evaluation = exchange + full stray field (FFT, no extrapolation) on a 100 × 100 × 100 cube of 2 nm cells (1 M cells); minimum over repeated calls, ms. magnum.np 2.2.0 and neuralmag 1.0.0 measured with M. Steinbauer's timing suite, jaxmag with the same measurement, using the FFT demag path pushed on Oct 13 (pad- and crop-as-you-go, real kernels, no z transform for single-layer meshes). mumax3 3.12 (float32 only) is timed separately, see the note. All seven cards are complete (H100: 203 of 216 suite processes, every mesh covered).

B200, all five large meshes, float32, ms per evaluation:

A100 SXM4 80 GB, all five large meshes, float32, ms per evaluation (mumax3: per torque evaluation):

All meshes, per-step times, memory and dollars per evaluation: slide deck of the campaign.

08 · About

Who runs this, and who pays

jaxmag is developed in the Physics of Functional Materials group at the University of Vienna by Dieter Suess and contributors, and released as open source.