The MPI+CUDA Gaia AVU-GSR Parallel Solver in perspective of next-generation Exascale Infrastructures and new Green Computing milestones

Date Issued
2022
Author(s)
•
•
Abstract
We ported on the GPU with CUDA the Gaia Astrometric Verification Unit-Global Sphere Reconstruction (AVU-GSR) Parallel Solver. The code aims to find the astrometric parameters of ~10^8 stars in the Milky Way, the attitude and the instrumental settings of the Gaia satellite, and the global parameter 𝛾 of the PPN formalism, by solving a system of linear equations, 𝐀 × 𝒙 = 𝒃, with the LSQR iterative algorithm. The coefficient matrix 𝐀 is large, having ~10^11 × 10^8 elements, and sparse. The CUDA code accelerates ≳ 14 times compared to the current version of the AVU-GSR code, parallelized on the CPU with MPI+OpenMP and in production since 2014. This acceleration factor is ~9.2 times larger than the one obtained with a preliminary GPU porting with OpenACC, equal to ~1.5. We obtained this result by running the codes on the CINECA SuperComputer Marconi100, that has 4 NVIDIA Volta V100 GPUs per node, where the MPI+CUDA application has been recently put in production. This analysis represents a first step to understand the exascale behaviour of a class of applications that follow the same structure of this code, employed in several contexts. In the next months, we plan to run this code on the pre-exascale platform Leonardo of CINECA, with 4 next-generation A100 GPUs per node, to better investigate this behaviour. Computing on highly parallel devices, such as GPUs, might imply a consistent power saving, which might go towards the achievement of a Green Computing milestone.
Series
Report number
164
Uri
Rights
open.access
File(s)
Loading...
Thumbnail Image
Name

Technical_report_Gaia_MPI_CUDA_Valentina_Cesare_et_al.pdf

Size

1.72 MB

Format

Adobe PDF

Checksum (MD5)

b381c55501cc421e27ad7c2212400e1f