3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering

Publikation: KonferenzbeitragAbstractForschungBegutachtung

Abstract

Reconstructing 3D parameter maps of huge volumes entirely on the GPU is highly desirable due to the offeredcomputation speed-up. However, GPU memory restrictions limit the coverable volume. To overcome thislimitation, a double-buffering strategy in combination with model-based parameter quantification and 3D-TGVregularization is proposed. This combination warrants whole volume reconstruction while maintaining thespeed advantages of GPU-based computation. In contrast to sequential transfers, double-buffering splits thevolume into blocks and overlaps memory transfer and kernel execution concurrently, hiding memory latency.The proposed method is able to reconstruct arbitrary large volumes within 5.3 min/slice, even on a single GPU.
Originalspracheenglisch
PublikationsstatusVeröffentlicht - 2019
VeranstaltungISMRM 27th Annual Meeting & Exhibition - Palais des congrès de Montréal, Montréal, Kanada
Dauer: 11 Mai 201916 Mai 2019
https://www.ismrm.org/19m/

Konferenz

KonferenzISMRM 27th Annual Meeting & Exhibition
KurztitelISMRM 2019
LandKanada
OrtMontréal
Zeitraum11/05/1916/05/19
Internetadresse

Fingerprint

Hardware
Data storage equipment
Graphics processing unit

Dies zitieren

Maier, O. (2019). 3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering. Abstract von ISMRM 27th Annual Meeting & Exhibition, Montréal, Kanada.

3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering. / Maier, Oliver.

2019. Abstract von ISMRM 27th Annual Meeting & Exhibition, Montréal, Kanada.

Publikation: KonferenzbeitragAbstractForschungBegutachtung

Maier O. 3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering. 2019. Abstract von ISMRM 27th Annual Meeting & Exhibition, Montréal, Kanada.
Maier, Oliver. / 3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering. Abstract von ISMRM 27th Annual Meeting & Exhibition, Montréal, Kanada.
@conference{d694f9e24c4f4385abc100941a152932,
title = "3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering",
abstract = "Reconstructing 3D parameter maps of huge volumes entirely on the GPU is highly desirable due to the offeredcomputation speed-up. However, GPU memory restrictions limit the coverable volume. To overcome thislimitation, a double-buffering strategy in combination with model-based parameter quantification and 3D-TGVregularization is proposed. This combination warrants whole volume reconstruction while maintaining thespeed advantages of GPU-based computation. In contrast to sequential transfers, double-buffering splits thevolume into blocks and overlaps memory transfer and kernel execution concurrently, hiding memory latency.The proposed method is able to reconstruct arbitrary large volumes within 5.3 min/slice, even on a single GPU.",
author = "Oliver Maier",
year = "2019",
language = "English",
note = "ISMRM 27th Annual Meeting & Exhibition, ISMRM 2019 ; Conference date: 11-05-2019 Through 16-05-2019",
url = "https://www.ismrm.org/19m/",

}

TY - CONF

T1 - 3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering

AU - Maier, Oliver

PY - 2019

Y1 - 2019

N2 - Reconstructing 3D parameter maps of huge volumes entirely on the GPU is highly desirable due to the offeredcomputation speed-up. However, GPU memory restrictions limit the coverable volume. To overcome thislimitation, a double-buffering strategy in combination with model-based parameter quantification and 3D-TGVregularization is proposed. This combination warrants whole volume reconstruction while maintaining thespeed advantages of GPU-based computation. In contrast to sequential transfers, double-buffering splits thevolume into blocks and overlaps memory transfer and kernel execution concurrently, hiding memory latency.The proposed method is able to reconstruct arbitrary large volumes within 5.3 min/slice, even on a single GPU.

AB - Reconstructing 3D parameter maps of huge volumes entirely on the GPU is highly desirable due to the offeredcomputation speed-up. However, GPU memory restrictions limit the coverable volume. To overcome thislimitation, a double-buffering strategy in combination with model-based parameter quantification and 3D-TGVregularization is proposed. This combination warrants whole volume reconstruction while maintaining thespeed advantages of GPU-based computation. In contrast to sequential transfers, double-buffering splits thevolume into blocks and overlaps memory transfer and kernel execution concurrently, hiding memory latency.The proposed method is able to reconstruct arbitrary large volumes within 5.3 min/slice, even on a single GPU.

M3 - Abstract

ER -