2015/01/30 by Simon Perkins, Perkins, Simon, Patrick Marais +9
Earth and Planetary Sciences · Environmental Science · Physics and Astronomy · #Computer Vision and Pattern Recognition (cs.CV) #Distributed #FOS: Computer and information sciences #FOS: Physical sciences #Instrumentation and Methods for Astrophysics (astro-ph.IM) #Parallel #Precipitation Measurement and Analysis #Radio Astronomy Observations and Technology #Soil Moisture and Remote Sensing #and Cluster Computing (cs.DC)
paper · pdf · doi:10.48550/arxiv.1501.07719
openalex publication_date 2015/01/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We present Montblanc, a GPU implementation of the Radio interferometer\nmeasurement equation (RIME) in support of the Bayesian inference for radio\nobservations (BIRO) technique. BIRO uses Bayesian inference to select sky\nmodels that best match the visibilities observed by a radio interferometer. To\naccomplish this, BIRO evaluates the RIME multiple times, varying sky model\nparameters to produce multiple model visibilities. Chi-squared values computed\nfrom the model and observed visibilities are used as likelihood values to drive\nthe Bayesian sampling process and select the best sky model.\n As most of the elements of the RIME and chi-squared calculation are\nindependent of one another, they are highly amenable to parallel computation.\nAdditionally, Montblanc caters for iterative RIME evaluation to produce\nmultiple chi-squared values. Modified model parameters are transferred to the\nGPU between each iteration.\n We implemented Montblanc as a Python package based upon NVIDIA's CUDA\narchitecture. As such, it is easy to extend and implement different pipelines.\nAt present, Montblanc supports point and Gaussian morphologies, but is designed\nfor easy addition of new source profiles.\n Montblanc's RIME implementation is performant: On an NVIDIA K40, it is\napproximately 250 times faster than MeqTrees on a dual hexacore Intel E5-2620v2\nCPU. Compared to the OSKAR simulator's GPU-implemented RIME components it is\n7.7 and 12 times faster on the same K40 for single and double-precision\nfloating point respectively. However, OSKAR's RIME implementation is more\ngeneral than Montblanc's BIRO-tailored RIME.\n Theoretical analysis of Montblanc's dominant CUDA kernel suggests that it is\nmemory bound. In practice, profiling shows that is balanced between compute and\nmemory, as much of the data required by the problem is retained in L1 and L2\ncache.\n