A Prototype Adaptive Optics Real-Time Control Architecture for Extremely Large Telescopes using Many-Core CPUs

Abstract

A proposed solution to the increased computational demands of Extremely Large Telescope (ELT) scale adaptive optics (AO) real-time control (RTC) using many-core CPU technologies is presented. Due to the nearly 4x increase in primary aperture diameter the next generation of 30-40m class ELTs will require much greater computational power than the current 10m class of telescopes. The computational demands of AO RTC scale to the fourth power of telescope diameter to maintain the spatial sampling required for adequate atmospheric correction. The Intel Xeon Phi is a standard socketed CPU processor which combines many (<64) low power cores with fast (>450GB/s) on-chip high bandwidth memory, properties which are perfectly suited to the highly parallelisable and memory bandwidth intensive workloads of ELT-scale AO RTC. Performance of CPU-based RTC software is analysed and compared for the single conjugate, multi conjugate and laser tomographic types of AO operating on the Xeon Phi and other many-core CPU solutions. This report concludes with an investigation into the potential performance of the CPU-based AO RTC software for the proposed instruments of the next generation Extremely Large Telescope (ELT) and the Thirty Meter Telescope (TMT) and also for some high order AO systems at current observatories.