Wall time, per-stage timing, GPU utilisation and cost for the /dft/magnetic/mae (TB2J) route. Cases are L1₀ FePt (2 and 16 atoms) and ferrimagnetic GdCo₅. Runs compare A100 and CPU workers, the ELPA, genelpa and cusolver solvers, and MPI process layouts, before and after the TB2J kernel rewrite. Every run in a case must give the same MAE. MAE values here still use TB2J's former 5.1 eV band cut; the accuracy dataset has the corrected values. Costs use Modal list prices.
| Column | Range | Mean | Filled |
|---|---|---|---|
| bench_case | 3 distinct | — | 100% |
| formula | 2 distinct | — | 100% |
| n_atoms | 2–16 | 7.143 | 100% |
| kmesh | 3 distinct | — | 100% |
| config | 6 distinct | — | 100% |
| hardware | 3 distinct | — | 100% |
| abacus_device | 2 distinct | — | 100% |
| abacus_solver | 3 distinct | — | 100% |
| process_layout | 4 distinct | — | 100% |
| tb2j_kernel | 2 distinct | — | 100% |
| tb2j_band_window | 1 distinct | — | 100% |
| wall_s | 57.6–560.7 | 294.1 | 100% |
| tb2j_scf_s | 25.3–413.4 | 134.3 | 100% |
| soc_step_s | 9.1–75.4 | 37.91 | 100% |
| tb2j_band_energy_s | 17.1–348.2 | 121.2 | 100% |
| est_cost_usd | 0.006–0.3857 | 0.1282 | 100% |
| mae_mev_per_fu |
| 1.037–3.169 |
| 2.478 |
| 100% |
| gpu_util_scf_pct | 0–23.9 | 9.757 | 50% |
| gpu_util_tb2j_pct | 0–4 | 1.629 | 50% |
| note | 3 distinct | — | 100% |