Abstract
This thesis addresses the computational challenges associated with the non-Local Thermal Equilibrium (non-LTE) radiative transfer problem in stellar atmospheres. By exploring the potential of Graphics Processing Units (GPUs) to accelerate the calculation for the formal solution. It delves into both the theoretical framework of non-LTE and the principles governing the operation of GPUs, laying a foundation for their intersection within this study. The core of this investigation involves implementing and optimising the non-LTE calculation on GPUs. Even though this work did not involve a complete solution. of the non-LTE problem. This work provides a quantifiable assessment of performance improvements by conducting a comparative analysis with the conventional CPU-based approach. The outcomes demonstrate a highly promising acceleration, with observed speedups reaching 82.7 times on specific GPU architectures when contrasted with traditional CPU calculations. Other GPU environments yielded accelerations of approximately 50 times, underscoring the broad applicability and advantage of GPU processing across different hardware configurations. Additionally, the thesis evaluates the energy efficiency of the GPU algorithms, revealing a significant reduction in power consumption compared to CPU implementations. These results highlight GPUs as potent tools for the computationally demanding non-LTE problem, thereby unlocking more rapid and energy-sustainable avenues for spectral analysis. The implications of this work extend beyond raw performance metrics, suggesting an advancement in solar physics research through improved computational means.
Supervisor: Professor Tiago Pereira, Institute of Theoretical Astrophysics, UiO
Intern. assessor: Professor Øystein Elgarøy, Institute of Theoretical Astrophysics, UiO
Extern. assessor: Professor Jorrit Leenaarts, Institute for Solar Physics University of Stockholm