Overview
Norne is a real Norwegian Sea oil field, released by Equinor and the University of Stavanger as an open benchmark. It is the credibility test between a clean Cartesian box (like SPE-9) and a genuine field study: 44,431 active cells on a faulted corner-point grid, 36 wells / 504 perforations, wet-gas PVT, water- alternating-gas (WAG) injection, and ~9 years (3,312 days) of production history.
Running Norne at all exercises nearly every part of the simulator — corner-point geometry, sealing and partially-sealing faults, non-neighbor connections, region transmissibility multipliers, endpoint scaling, and gas hysteresis. Matching OPM Flow on it is the goal.
Reference: OPM Flow 2021.10, run on the same deck and the same CPU cores.
What it takes to run Norne
The full geometry and physics stack was validated cell-by-cell against OPM’s own exported arrays:
| Component | Detail |
|---|---|
| Corner-point grid | NEWTRAN transmissibilities; vertical trans cross-checked vs OPM TRANZ |
| Faults & multipliers | 977 fault faces / 60 MULTFLT; 23,986 MULTX/Y/Z; 7,449 MULTREGT |
| Wells | 36 wells, COMPDAT connection factors, group controls, RESV voidage |
| Physics | wet-gas PVT, 3-phase relperm with endpoint scaling, gas-trapping hysteresis |
| System size | 44,431 active cells, 123,991 connections — 5× larger than SPE-9 |
Accuracy
BirSim reproduces the full 3,312-day history against OPM Flow on the
mass-conservation invariants (cumulative oil / gas / water production) and on field
pressure. The corner-point transmissibilities match OPM’s exported TRANZ to the
unit factor, so the geometry the two simulators solve on is the same.
| Quantity | Validation |
|---|---|
| Cumulative production (FOPT / FGPT / FWPT) | Full-history match |
| Field pressure (FPR) | Tracks OPM over 9 years |
| Vertical transmissibility | Cell-by-cell vs OPM TRANZ (unit factor) |
Performance vs OPM Flow
Norne’s dominant field-scale cost is the Newton-iteration count. A campaign of targeted, answer-preserving levers — a per-Newton linear tolerance, OPM-style relaxed-CNV convergence, and a predictive timestep controller — cut the full-history wall from the first running build to production without a solver rewrite.
Full history, 2 cores, same host
| Metric | BirSim v1 | BirSim now | OPM Flow | now vs OPM |
|---|---|---|---|---|
| Wall time | 11,916 s | 1,199 s | 445 s | ~2.0× |
| Newton (accepted) | 7,074 | 4,670 | 1,425 | 3.3× |
| Wall per Newton | 1.68 s | 0.19 s | 0.31 s | 0.6× (faster) |
| Timesteps | 5,015 | 2,867 | 358 | 8.0× |
BirSim’s per-Newton cost is already faster than OPM (0.19 s vs 0.31 s); the remaining ~2× gap is the number of Newton iterations, a genuine field-scale robustness floor rather than a parallelization deficit.
Multi-CPU (MPI) scaling
A balanced graph partitioner took the distributed run from a naive 20,374 s → 524 s on 12 ranks. At matched core counts, clean MPI-vs-MPI, BirSim lands at ~2× OPM across every core count — the scaling is healthy; the residual is the Newton-count wall.
| Configuration | Wall time |
|---|---|
| MPI, 12 ranks (naive partition) | 20,374 s |
| MPI, 12 ranks (balanced partition) | 524 s |
Why it matters
SPE-9 proves a simulator is correct on a clean grid. Norne proves it survives a real one — faulted corner-point geometry, non-neighbor connections, region multipliers, and a multi-well WAG schedule, over nine years of history. BirSim matches OPM Flow on the physics and closes to ~2× on runtime at matched cores, with per-iteration cost already ahead.
Reference
Norne is distributed via the Open Porous Media (OPM) project with the permission of Equinor and the Norne license partners. Deck path, environment flags, and command lines are available on request.