Every spectrum of a star is like a tapestry woven from thousands of fine absorption lines. These lines are left by atoms in the star's atmosphere, and from their pattern astronomers learn the temperature, composition, and motion of the star. But such a tapestry is hard to 'weave': old programs did it sequentially, thread by thread, taking nearly an hour for a single star.

Payne Zero moves this process onto a graphics processing unit (GPU), which can perform many operations simultaneously — like a loom with millions of needles. Now all spectral lines are calculated at once. The atmospheric model, which sets the tension and color of the threads, is accelerated on a multi-core CPU by a factor of 6–7, and a trained assistant picks initial conditions, reducing the number of iterations. The result: a full optical spectrum of the Sun is ready in 14 seconds — 54 times faster than before.

This speed makes it possible, for the first time, to embed direct physical calculations into the study of millions of stars. In a test on 1600 stars from the APOGEE survey, the code measured the abundances of 12 elements — from carbon to nickel — in a single pass and clearly separated the stars into two chemical groups, as expected. No more relying on approximate surrogates: each star's spectrum is woven anew from atomic 'threads', calibrated using reference stars like the Sun. This approach promises to recalibrate the atomic data itself and reveal the detailed chemical evolution history of our Galaxy.