Peter Wurmsdobler, Cambridge Consultants Ltd.

This paper presents the following quite demanding Real-Time Linux application. Radar data is generated by bespoke hardware at a rate of 880MB/s and made available to a personal supercomputer at 7.5 kHz through a dedicated PCI express card. This personal supercomputer is equipped with an NVIDA Tesla card providing a theoretical 1TeraFlops processing power and runs Real-Time Linux. The interrupt generated by the PCI express card is processed by in an interrupt service routine which places the radar samples into a ring buffer. After a set number of samples, a thread woken up by the interrupt service routine transfers blocks of pre-processed data to the NVIDIA Tesla card for further quite intensive processing by 240 so-called stream processors, involving subsequent FFT performed on the input data. The result produced by the Tesla card is picked up by the main thread and transferred either to file or another PC for visualisation.