GPTune is designed to tune HPC codes (or any other tuning problems) as “black-boxes”, running them for carefully chosen tuning parameter values and building a performance model (i.e. surrogate model) based on the function evaluation results (i.e. run and measure the application on a parallel machine).

GPTune supports single-task, multi-task, multi-fidelity, multi-objective autotuning for a set of integer, real, and categorical tuning parameters. GPTune also provides a historical database which enables transfer learning across different users and different machines.

To improve GPTune, we kindly ask for feedback from our users. All survey forms are optional. You can answer the forms you want to reply to.

GPTune website: https://gptune.lbl.gov
Contact information: gptune-dev@lbl.gov