Graphical processing units (GPUs) have reached new heights of popularity with the advent of AI models in recent years. For parallelizable calculations, they provide unmatched computational power hundreds of times faster than traditional central-processing-unit (CPU)-based computing. This white paper explores use cases for GPUs in actuarial modeling and discusses applicability, benchmarks, and a cost-performance overview. In recent years, there has been a mass adoption of a new wave of generative pre-trained transformer (GPT) AI models, which currently require GPUs for training and efficient execution of user queries, exploiting computations orders of magnitude faster than what is achievable on CPUs. Discussion points in the paper include:
- CPU versus GPU
- Understanding GPU performance
- Suitability for GPU processing
- Architecture fundamentals
- Actuarial problems and GPUs
- Efficient modeling
- Cost analysis of GPU performance