The Neko development team is pleased to announce the release of Neko v1.1, released during WCCM-ECCOMAS 2026 in Munich.
A minor release in name only. The parallel backend saw its biggest overhaul since v1.0: four new gather-scatter backends (native Tofu for Fugaku, MPI-3 neighbourhood collectives, Coarray Fortran, OpenSHMEM), a fused three-component halo exchange, and a new crystal router that pushes initialisation past 100k MPI ranks. The CPU critical path is now threaded, so hybrid MPI+OpenMP is a first-class execution mode.
On the physics side: a compressible Navier-Stokes solver, an ALE framework with multiple independently moving bodies, Lagrangian particle tracking, and new wall and SGS models for ABL simulations.
Also worth noting — unified memory on Apple Silicon and AMD MI300A roughly halves the footprint of mapped data, and case files now get strict type checking plus a JSON schema.
There are breaking changes, including explicit CUDA_ARCH at build time. See the release notes.
Thanks to everyone who contributed!
/ᐠ. ｡.ᐟ\ᵐᵉᵒʷˎˊ˗