Conjugate gradient solvers on Intel Xeon Phi and NVIDIA GPUs - INSPIRE

DataBETA

Conjugate gradient solvers on Intel Xeon Phi and NVIDIA GPUs

Nov 17, 2014

6 pages

Part of Proceedings, GPU Computing in High-Energy Physics (GPUHEP2014) : Pisa, Italy, September 10-12, 2014, 157-162

Contribution to:

GPU Computing in High-Energy Physics, 157-162

Published: Jun, 2015

e-Print:

1411.4439 [physics.comp-ph]

DOI:

10.3204/DESY-PROC-2014-05/28

View in:

ADS Abstract Service

reference search6 citations

Citations per year

Abstract:

Lattice Quantum Chromodynamics simulations typically spend most of the runtime in inversions of the Fermion Matrix. This part is therefore frequently optimized for various HPC architectures. Here we compare the performance of the Intel R © Xeon Phi TM to current Kepler-based NVIDIA R © Tesla TM GPUs running a conjugate gradient solver. By exposing more parallelism to the accelerator through inverting multiple vectors at the same time, we obtain a performance greater than 300 GFlop / s on both architectures. This more than doubles the performance of the inversions. We also give a short overview of the Knights Corner architecture, discuss some details of the implementation and the effort required to obtain the achieved performance

Note:

7 pages, proceedings, presented at 'GPU Computing in High Energy Physics', September 10-12, 2014, Pisa, Italy

lattice field theory
quantum chromodynamics
numerical calculations
multiprocessor: graphics
programming

References(16)

Figures(5)

[1]

The melting and abundance of open charm hadrons

A. Bazavov
(
- Iowa U.
)
,
H.-T. Ding
(
- Hua-Zhong Normal U., LQLP and
- CCNU, Wuhan, Inst. Part. Phys.
)
,
P. Hegde
(
- Hua-Zhong Normal U., LQLP and
- CCNU, Wuhan, Inst. Part. Phys.
)
,
O. Kaczmarek
(
- Bielefeld U.
)
,
F. Karsch
(
- Bielefeld U. and
- Brookhaven
)

et al.

- Phys.Lett.B 737 (2014) 210-215
•
e-Print:
- 1404.4043
•
DOI:
- 10.1016/j.physletb.2014.08.034

[1]

Additional Strange Hadrons from QCD Thermodynamics and Strangeness Freezeout in Heavy Ion Collisions

A. Bazavov
(
- Iowa U.
)
,
H. -T. Ding
(
- Hua-Zhong Normal U. and
- Hua-Zhong Normal U., LQLP
)
,
P. Hegde
(
- Hua-Zhong Normal U. and
- Hua-Zhong Normal U., LQLP
)
,
O. Kaczmarek
(
- Bielefeld U.
)
,
F. Karsch
(
- Bielefeld U. and
- Brookhaven
)

et al.

- Phys.Rev.Lett. 113 (2014) 7, 072001
•
e-Print:
- 1404.6511
•
DOI:
- 10.1103/PhysRevLett.113.072001

[1]

Strangeness at high temperatures: from hadrons to quarks

A. Bazavov
(
- Brookhaven
)
,
H. -T. Ding
(
- Brookhaven and
- Columbia U.
)
,
P. Hegde
(
- Taiwan, Natl. Taiwan U.
)
,
O. Kaczmarek
(
- Bielefeld U.
)
,
F. Karsch
(
- Brookhaven and
- Bielefeld U.
)

et al.

- Phys.Rev.Lett. 111 (2013) 082301
•
e-Print:
- 1304.7220
•
DOI:
- 10.1103/PhysRevLett.111.082301

[1]

Freeze-out Conditions in Heavy Ion Collisions from QCD Thermodynamics

A. Bazavov
(
- Brookhaven
)
,
H.T. Ding
(
- Brookhaven
)
,
P. Hegde
(
- Brookhaven
)
,
O. Kaczmarek
(
- Bielefeld U.
)
,
F. Karsch
(
- Brookhaven and
- Bielefeld U.
)

et al.

- Phys.Rev.Lett. 109 (2012) 192302
•
e-Print:
- 1208.1220
•
DOI:
- 10.1103/PhysRevLett.109.192302

[2]

Highly improved staggered quarks on the lattice, with applications to charm physics

HPQCD and
UKQCD

Collaborations

•

E. Follana
(
- Glasgow U.
)

et al.

- Phys.Rev.D 75 (2007) 054502
•
e-Print:
- hep-lat/0610092
•
DOI:
- 10.1103/PhysRevD.75.054502

[2]

Scaling studies of QCD with the dynamical HISQ action

MILC

Collaboration

•

A. Bazavov
(
- Arizona U.
)

et al.

- Phys.Rev.D 82 (2010) 074501
•
e-Print:
- 1004.0342
•
DOI:
- 10.1103/PhysRevD.82.074501

[3]

A Framework for Lattice QCD Calculations on GPUs

e-Print:
- 1408.5925
•
DOI:
- 10.1109/IPDPS.2014.112

[3]

Scaling lattice QCD beyond 100 GPUs

QUDA

Collaboration

•

R. Babich
(
- Boston U., Ctr. Comp. Sci.
)

et al.

•
e-Print:
- 1109.2935
•
DOI:
- 10.1145/2063384.2063478

[3]

Solving Lattice QCD systems of equations using mixed precision solvers on GPUs

QUDA

Collaboration

•

M.A. Clark
(
- Harvard-Smithsonian Ctr. Astrophys. and
- Harvard U.
)

et al.

- Comput.Phys.Commun. 181 (2010) 1517-1528
•
e-Print:
- 0911.3191
•
DOI:
- 10.1016/j.cpc.2010.05.002

[3]

Lattice QCD as a video game

Gyozo I. Egri
(
- Eotvos U.
)
,
Zoltan Fodor
(
- Eotvos U. and
- Wuppertal U. and
- UC, San Diego
)
,
Christian Hoelbling
(
- Wuppertal U.
)
,
Sandor D. Katz
(
- Eotvos U. and
- Wuppertal U.
)
,
Daniel Nogradi
(
- Wuppertal U.
)

et al.

- Comput.Phys.Commun. 177 (2007) 631-639,
- Comput.Phys.Commun. 177 (2007) 631-639
•
e-Print:
- hep-lat/0611022
•
DOI:
- 10.1016/j.cpc.2007.06.005

[4]

Lattice QCD on Intel® Xeon Phi Coprocessors

et al.

- Lect.Notes Comput.Sci. 7905 (2013) 40-54
•
DOI:
- 10.1007/978-3-642-38750-0_4

[4]

Staggered Dslash Performance on Intel Xeon Phi Architecture

Ruizi Li
(
- Indiana U.
)
,
Steven Gottlieb
(
- Indiana U.
)

- PoS LATTICE2014 (2015) 034
•
e-Print:
- 1411.2087
•
DOI:
- 10.22323/1.214.0034

[4]

Code Optimization on Kepler GPUs and Xeon Phi

SWME

Collaboration

•

Yong-Chull Jang
(
- Seoul Natl. U., Dept. Phys. Astron.
)

et al.

- PoS LATTICE2014 (2014) 035
•
e-Print:
- 1411.2223
•
DOI:
- 10.22323/1.214.0035

[5]

HISQ inverter on Intel $\scriptsize{^{\circledR}}$ Xeon Phi $\scriptsize{^{TM}}$ and NVIDIA $\scriptsize{^{\circledR}}$ GPUs

Swagato Mukherjee
(
- Brookhaven
)
,
Olaf Kaczmarek
(
- Bielefeld U.
)
,
Christian Schmidt
(
- Bielefeld U.
)
,
Patrick Steinbrecher
(
- Bielefeld U.
)
,
Mathias Wagner
(
- Indiana U.
)

- PoS LATTICE2014 (2015) 044
•
e-Print:
- 1409.1510
•
DOI:
- 10.22323/1.214.0044

[6]

System Software Developers Guide

Intel R. Xeon Phi Coprocessor

[7]

6 GPUHEP

J.D. McCalpin