Javascript must be enabled to continue!
Prospects of GPU Tensor Core Correlation for the SMA and the ngEHT
View through CrossRef
Building on the base of the existing telescopes of the Event Horizon Telescope (EHT) and ALMA, the next-generation EHT (ngEHT) aspires to deploy ∼10 more stations. The ngEHT targets an angular resolution of ∼15 microarcseconds. This resolution is achieved using Very Long Baseline Interferometry (VLBI) at the shortest radio wavelengths ∼1 mm. The Submillimeter Array (SMA) is both a standalone radio interferometer and a station of the EHT and will conduct observations together with the new ngEHT stations. The future EHT + ngEHT array requires a dedicated correlator to process massive amounts of data. The current correlator-beamformer (CBF) of the SMA would also benefit from an upgrade, to expand the SMA’s bandwidth and also match the EHT + ngEHT observations. The two correlators share the same basic architecture, so that the development time can be reduced using common technology for both applications. This paper explores the prospects of using Tensor Core Graphics Processing Units (TC GPU) as the primary digital signal processing (DSP) engine. This paper describes the architecture, aspects of the detailed design, and approaches to performance optimization of a CBF using the “FX” approach. We describe some of the benefits and challenges of the TC GPU approach.
Title: Prospects of GPU Tensor Core Correlation for the SMA and the ngEHT
Description:
Building on the base of the existing telescopes of the Event Horizon Telescope (EHT) and ALMA, the next-generation EHT (ngEHT) aspires to deploy ∼10 more stations.
The ngEHT targets an angular resolution of ∼15 microarcseconds.
This resolution is achieved using Very Long Baseline Interferometry (VLBI) at the shortest radio wavelengths ∼1 mm.
The Submillimeter Array (SMA) is both a standalone radio interferometer and a station of the EHT and will conduct observations together with the new ngEHT stations.
The future EHT + ngEHT array requires a dedicated correlator to process massive amounts of data.
The current correlator-beamformer (CBF) of the SMA would also benefit from an upgrade, to expand the SMA’s bandwidth and also match the EHT + ngEHT observations.
The two correlators share the same basic architecture, so that the development time can be reduced using common technology for both applications.
This paper explores the prospects of using Tensor Core Graphics Processing Units (TC GPU) as the primary digital signal processing (DSP) engine.
This paper describes the architecture, aspects of the detailed design, and approaches to performance optimization of a CBF using the “FX” approach.
We describe some of the benefits and challenges of the TC GPU approach.
Related Results
On the programmability of multi-GPU computing systems
On the programmability of multi-GPU computing systems
Multi-GPU systems are widely used in High Performance Computing environments to accelerate scientific computations.
This trend is expected to continue as integrated GPUs will be i...
Theoretical Foundations and Practical Applications in Signal Processing and Machine Learning
Theoretical Foundations and Practical Applications in Signal Processing and Machine Learning
Tensor decomposition has emerged as a powerful mathematical framework for analyzing multi-dimensional data, extending classical matrix decomposition techniques to higher-order repr...
Enhanced inherent strain modelling for powder-based metal additive manufacturing
Enhanced inherent strain modelling for powder-based metal additive manufacturing
(English) Metal additive manufacturing (MAM), particularly powder bed fusion using a laser beam (PBF-LB), has transformed manufacturing by enabling the production of intricate and ...
CPU AND GPU (CUDA) TEMPLATE MATCHING COMPARISON / CPU IR GPU (CUDA) PALYGINIMAS VYKDANT ŠABLONŲ ATITIKTIES ALGORITMĄ
CPU AND GPU (CUDA) TEMPLATE MATCHING COMPARISON / CPU IR GPU (CUDA) PALYGINIMAS VYKDANT ŠABLONŲ ATITIKTIES ALGORITMĄ
Image processing, computer vision or other complicated opticalinformation processing algorithms require large resources. It isoften desired to execute algorithms in real time. It i...
Sign Problem in Tensor-Network Contraction
Sign Problem in Tensor-Network Contraction
We investigate how the computational difficulty of contracting tensor networks depends on the sign structure of the tensor entries. Using results from computational complexity, we ...
Benchmarking GPU Passthrough Performance on Docker for AI Cloud System
Benchmarking GPU Passthrough Performance on Docker for AI Cloud System
The use of artificial intelligence (AI), which depends only on CPU resources, tends to result in longer execution times or CPU time. Especially when handling large amounts or compl...
Gravitational Waves from Alena Tensor
Gravitational Waves from Alena Tensor
Alena Tensor is a recently discovered class of energy-momentum tensors that proposes a general equivalence of the curved path and the geodesic for the analyzed spacetimes which all...

