Intel® many integrated core (MIC) architecture: portability and performance efficiency study of radio astronomy algorithms

Radio Astronomy demands for HPC power have been rising and are expected to reach exaflop scale by 2020. To address such huge demands for compute, users are testing newer CPU architectures and accelerator architectures such as GPUs and FPGAs. Intel, with the help of HPC Ecosystem partners, has stated a goal to deliver exascale computing capability by 2018. This goal is dependent on solutions based on Intel's multi-core architecture and the Intel® Many Integrated Core (Intel MIC) architecture. In this talk, we explore the Intel MIC platform in the context of 3 radio astronomy applications: Software Correlation, Coherent Beam Forming for distributed arrays of antennas, such as the Murchison Widefield Array (MWA), and FITS to JPEG2000 Conversion and Encoding (SkuareView). These algorithms are highly computational and data intensive and involve mathematical operations like cross correlation, convolution, Fast Fourier Transform and Wavelet Transforms. Our porting efforts to the Intel MIC platform have clearly highlighted the ease of programming, as well as performance efficiency, since the same codebase can be run on both CPU and Intel MIC.