Central stars of nova systems

To investigate the performance gains for a realistic stellar model at higher temperatures, we ran a complete NLTE model calculation for a set of parameters expected for the primary component of a nova system in a quiet post-outburst phase. The parameters appropriate for these white dwarf systems are $\hbox{$\,T_{\rm eff}$}=21,000\k$, $\log(g)=8.0$ and solar abundances. We use the following NLTE species: H I-II, He I-III, C I-IV, N I-VI, O I-VI, Mg II, Ca II, S II-II, Si II-III, Ne I, and Fe II with a total of 2,826 NLTE levels and 26,874 NLTE lines. All NLTE lines are treated with detailed Voigt profiles. In addition, we include 621,920 LTE background lines and the calculation uses a total of 93,619 wavelengths points. On a SGI Power Indigo 2 a single model iteration with this setup requires about 9.5h CPU time and the full model run (10 iterations) needs about 2 days CPU time. On 10 processors of the 2 (using a standard load-distribution not optimized for the model parameters), a single iteration takes about 1.4h wall-clock time and the full model calculation needs about 7.3h wall-clock time, which is less than the CPU time for a single iteration on the SGI Power Indigo 2. This shows that the parallel speed-ups that we can achieve in realistic calculations are very significant, even with a small number of CPU's. With about 7.5h wall-clock time for a complete model iteration, substantial grids of these models can be constructed in relatively short time, thus making detailed analysis of observed spectra with the best input physics feasible.