Alfred Wegener Institute ports the FESOM2 ocean model (about 74,000 lines of Fortran) to C and then GPU-ready C++/Kokkos with Claude Code in a matter of weeks
Domain experts at the Alfred Wegener Institute directed Claude Code, running Claude Opus 4.7 in auto-accept mode, to port the FESOM2 unstructured-mesh ocean and sea-ice model from about 74,000 lines of core Fortran first to a clean C reference (about 20,000 lines, 10 days with commits) and then to C++/Kokkos (about 31,000 lines, 7 days with commits). The key rules were a strictly literal line-by-line translation, two stages separating numerics from parallelism, and a validation ladder per stage. The C port matches the Fortran over five-year runs; the Kokkos port is bit-for-bit identical to the C reference on CPU and statistically close on GPU. A single A100 GPU node runs 1.6–3.7x faster than a CPU node on eddy-rich meshes, and the same source ran on NVIDIA GH200, H100 and AMD MI250X systems. Work was split into short sessions driven by plan, lessons-learned and handoff files.
The numbers (as reported)
| Fortran ported | 73,991 lines in 109 files (core ocean–sea-ice configuration) |
|---|---|
| C port | 20,010 lines, 73 files, 10 days with commits |
| C++/Kokkos port | 30,663 lines, 78 files, 7 days with commits |
| Fidelity, C vs Fortran over five years | SST RMS difference 0.006 °C, SSS 0.002 PSU |
| GPU vs CPU node | 1.6–3.7x faster on meshes up to 7.4 million surface vertices |
| Throughput at 32 nodes | 1.3 SYPD (A100) to 3.5 SYPD (GH200) |
| Bring-up on other GPU systems | Less than a day each |
Facts
- Company / subject
- Alfred Wegener Institute (FESOM2)
- Tools
- Claude Code
- Models
- Claude Opus 4.7
- Codebase
- brownfield
- Work type
- Language port
- Stack
- Fortran, C, C++, Kokkos, CUDA, HIP, MPI, netCDF
- Published
- 2026-06-09
- Last checked
- 2026-11-16
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- Full Fortran ocean model moved to another language and onto GPUs while preserving its physics in a matter of weeks
“while preserving its physics in a matter of weeks”
- A single A100 GPU node runs 1.6–3.7x faster than a CPU node on eddy-rich meshes 1.6–3.7x
- Reaches the 1–2 simulated-years-per-day production band on every machine tested 1.3–3.5 SYPD at 32 nodes
- Each new accelerator system took less than a day to compile and run < 1 day per system
Caveats
Written by the scientists who directed the port. The ported configuration covers the core ocean–sea-ice model (the full distribution is about 269,000 lines) and still lacks restart I/O, other vertical-coordinate variants and coupled-climate physics needed to replace the Fortran in production. Days with commits are not total effort, and no cost or human hours are given. An earlier attempt by the first author had failed. The output system was redesigned rather than ported.
More stories by tool: Claude Code
Related topics: Language portsBrownfield codebases