BLADE - The Allen Telescope Array CUDA-Accelerated Real-Time DSP Library

Explore a conference talk on BLADE, the Allen Telescope Array CUDA-Accelerated Real-Time DSP Library. Delve into the development of a GPU-based software for processing data from a radio interferometer, focusing on beamforming and high-resolution spectrograms. Learn about overcoming hardware bottlenecks, implementing optimization techniques, and building a distributed pipeline capable of processing massive data streams in real-time. Gain insights into software-defined telescopes, GPU-enabled processing backends, and the modular interface of BLADE that applies advanced optimization techniques automatically.

Introduction
Agenda
Island Telescope Array
Antennas
Compute Cluster
RF Stock Mart
Satellite Downlink
New Science
Bing Farming
Hybrid Processing
BLADE
Blade Components
WeightVector
spanner
custom class
new method
shape class
profiling
transpositions
structs
configuration
input and output
setup and forget
kernel compilation
Big Farmer
High Spectrum Resolution Spectrogram
Pipeline
Pipeline initialization
BladeRunner
Synchronous Architecture
Production Pipeline
Summary
Contact
Any questions