What is new? FP32 solver mode DPCPP backend to support Intel GPUs Interpolation based recycling + support of arbitrary fields Ascent in situ visualisation plugin iofld class reading/writing field files including ADIOS2 support Addtional output options (element filter and interpolation on uniform grid / different polynomial-order) Multi session nek-nek including multi-rate time stepping Typed device memory and pool allocation CHT nek-nek support nrsqsub nek-nek handling Improved JIT compilation performance HIP support for SEMFEM Aero forces SEM ops Mesh surface ops Gradient jump penalty method and improved avm reguarlization Linear implicit velocity source term T-mesh support for all scalars Robin boundary conditions RANS ktau variants SST, DDES, IDDES Various bug fixes Good to know GPU aware MPI is disabled by default (to enable set env-var NEKRS_GPU_MPI=1) which may cause a performance regression [reproducibility] variable time step controller restricts dt to 5 significant digits nrsman can be used to display man pages AmgX support was removed (replaced by HYPRE) Field file extension starts with 0-index nrsqsub scripts are moved to HPCsupport First time step may take longer due to initialization overhead Breaking Changes This list provides an overview of the most significant changes in this release, although it may not encompass all modifications. We recognize that several of these updates directly affect the user experience. These adjustments were essential to enhance the stability of the user interface in future iterations. We apologize for any inconvenience this may cause. run build.sh instead of nrsconfig to build the code host mirrored variables like nrs->U, cds->S, mesh->x have been removed send signal (defined in env-var NEKRS_SIGNUM_UPD) to process trigger file nekrs.upd use auto foo = platform->deviceMemoryPool.reserve(nWords) instead of platform->o_mempool.slice[0-15] change count argument of occa::memory::slice, occa::memory::copyFrom, occa::memory::copyTo to number of words instead of bytes use nekrs_registerPtr instead of common blocks NRSSCPTR / SCNRS in usr file and access them using nek::ptr in udf (see examples) use par section [GEOM] for geometry solver instead of [MESH] usr scalars = FOO1, FOO2, ... in [GENERAL] to define additional scalar transport equations use mesh = fluid+solid for T-mesh (conjugate heat transfer) for a scalar e.g. temperature use iofld class () instead of writeFld use planarCopy class instead of velRecycling ::postProcessing functions are now members of nrs_t (except planarAvg) use boundary functions like udfDirichlet for all fields par key GENERAL::checkpointInterval ::setup udf::properties = std::function -> nrs_t::userProperties = std::function udf::uEqnSource = std::function -> nrs_t::userSource = std::function (same hook for all fields) udf::udfconv = std::function -> nrs_t::userConvergenceCheck = std::function udf::udfdif = std::function -> nrs_t::userDivergence = std::function tavg::setup(nrs_t*) -> tavg::setup(dlong fieldOffset, const fields& fields) postProcessing::planarAvg(nrs_t*, const std::string&, int, int, int, int, occa::memory) -> mesh_t::planarAvg(const std::string&, int, int, int, int, dlong, occa::memory o_avg) pointInterpolation_t::setPoints(int, dfloat*, dfloat*, dfloat*) -> pointInterpolation_t::setPoints(const std::vector&, const std::vector&, const std::vector&) Name Changes boundary type names: codedFixed -> udf, value -> Dirichlet, Gradient -> Neumann some member names of bcData changed (see include/app/nrs/bdry/bcData.h) velocityDirichletConditions -> udfDirichlet (same for all fields) velocityNeumannConditions -> udfNeumann (same for all fields) nrs->o_usrwrk-> platform->app->bc->o_usrwrk nrs->_mesh -> nrs->meshT nek::ocopyToNek -> nrs->copyToNek nek::ocopyFromNek -> nrs->copyFromNek nrs->o_FU -> nrs->fluid->o_explicitTerms() cds -> scalar cds->o_FS -> scalar->o_explicitTerms(std::string) cds->o_rho -> scalar->o_transportCoeff(std::string) cds->o_diff -> scalar->o_diffusionCoeff(std::string) cds->o_diff -> scalar->o_diffusionCoeff(std::string) cds->mesh[...] -> scalar->mesh(...) occaKernel -> deviceKernel occaProperties > deviceKernelProperties occa::memory -> deviceMemory nrs->isOutputStep -> nrs->checkpointStep subCyclingSteps -> advectionSubCyclingSteps in par par section VELOCITY -> FLUID VELOCITY par section PRESSURE -> FLUID PRESSURE par section TEMPERATURE -> SCALAR TEMPERATURE par section SCALAR01 -> SCALAR FOO par SCALAR key diffusivity -> diffusionCoeff par SCALARkey density -> transportCoeff comm.mpiRank -> comm.mpiRank() comm.mpiComm -> comm.mpiComm() Known Bugs / Restrictions Code is not fully optimized on CPUs and AMD MI300 GPUs planar avg in 1D is limited to 256 gridpoints per direction locally (per MPI Rank) SEMFEM preconditioner not available when using DPCPP 507 485 258 Thanks to our Contributors @yslan, @kris-rowe, @nandu90, @thilinarmtb We are grateful to all who added new features, filed issues or helped resolve them, asked and answered questions, and were part of inspiring discussions.