National Institute for

birdsowlΛογισμικό & κατασκευή λογ/κού

2 Δεκ 2013 (πριν από 3 χρόνια και 8 μήνες)

71 εμφανίσεις


NICS is a collaboration between the
University of Tennessee and ORNL


Awarded the NSF Track 2B
-
Kraken (1PF)


Remote Data Analysis and Visualization

Nautilus (Sean Ahern)


Experimental GPGPU system


Keeneland (Jeff Vetter)


National Institute for

Computational Sciences

Cray XT5 system


October 2009


8,256 two
-
socket nodes


16,512 six
-
core AMD Istanbul processors


99,072 cores (2.6 GHz)


129 TB memory


1,030 teraflops

NICS User Accounts and Projects


#

Date

235

305

382

481

515

586

638

747

902

985

1074

1157

1448

1504

1562

1633

1670

1707

1808

2024

2124

2255

35

49

78

90

92

107

125

149

170

187

201

236

254

268

290

308

315

327

371

400

441

462

0
500
1000
1500
2000
2500
Aug-08
Sep-08
Oct-08
Nov-08
Dec-08
Jan-09
Feb-09
Mar-09
Apr-09
May-09
Jun-09
Jul-09
Aug-09
Sep-09
Oct-09
Nov-09
Dec-09
Jan-10
Feb-10
Mar-10
Apr-10
May-10
# of Users
# of Projects
HPSS Usage


#

Date

24

29

40

42

51

61

68

86

96

107

113

119

128

139

158

169

171

183

193

202

361

372

380

422

481

506

729

945

1088

1462

1527

1572

1773

1992

2005

2111

2194

2248

2442

2518

72

90

138

164

240

327

391

475

592

702

794

943

1,234

1,311

1,523

1,836

2,115

2,470

2,883

3,098

0
500
1000
1500
2000
2500
3000
3500
Aug-08
Sep-08
Oct-08
Nov-08
Dec-08
Jan-09
Feb-09
Mar-09
Apr-09
May-09
Jun-09
Jul-09
Aug-09
Sep-09
Oct-09
Nov-09
Dec-09
Jan-10
Feb-10
Mar-10
# of Users
# of Files (thous)
Total TB stored
Kraken Job Mix (March, 2010)

0
1,000,000
2,000,000
3,000,000
4,000,000
5,000,000
6,000,000
1
4
8
16
24
32
40
48
56
60
63
127
255
511
1023
2047
4095
8191
8256
Nodes

Wallclock

Hours

Nodes

CPU
-
Hours

Kraken Utilization (weekly)

39

71

64

50

39

55

49

53

58

65

63

37

34

35

47

66

78

70

77

92

90

92

74

81

87

84

83

83

77

85

86

90

96

94

0
10
20
30
40
50
60
70
80
90
100
10/5
10/12
10/19
10/26
11/2
11/9
11/16
11/23
11/30
12/7
12/14
12/21
12/28
1/4
1/11
1/18
1/25
2/1
2/8
2/15
2/22
3/1
3/8
3/15
3/22
3/29
4/5
4/12
4/19
4/26
5/3
5/10
5/17
5/24
Kraken XT5 Utilization
Nautilus Versions: all SGI
Ultraviolet, running SLES 11 OS


P0 (half
-
rack)


128 Cores


256 GB RAM


1 GPU


P1 (1 rack)


256 Cores


1 TB RAM


4 GPUs


Final System (4 racks)


1024 Cores


4 TB RAM


16 GPUs


Nautilus Delivery Schedule

Remote Data Analysis &
Visualization Events


RDAV resources are currently in the allocations
system, and several requests have been made.


J
oint visualization class with TACC at the
Petascale

Programming Environments and Tools
classes in early July.


A

tutorial on Nautilus usage for visualization,
data analysis, and workflow management will be
taught at TeraGrid'10



Keeneland



An NSF
-
Funded Partnership
to Enable Large
-
scale Computational
Science on Heterogeneous Architectures


NSF Track 2D System of
Innovative Design


Georgia Tech


University of Tennessee, Knoxville


UT National Institute for
Computational Sciences


ORNL


Exploit graphics processors to
provide extreme performance
and energy efficiency


Deploy two GPU clusters


Initial Delivery


2010


Final Delivery


2012


NVIDIA, HP, Intel,
Qlogic


Software tools, application
development


Operations, user support


Education, Outreach, Training
for scientists, students,
industry



FERMI


capable of over 1 TFs single
precision and over 500 GFs
double precision


Includes error correction in
memory


Includes new level
of cache

10

NVIDIA’s new Fermi GPU

Jeffrey
Vetter
Jack
Dongarra
Richard
Fujimoto
Thomas
Schulthess
Karsten
Schwan
Phil
Andrews
Troy
Baer
Kathlyn
Boudwin
Mark
Fahey
Jim
Ferguson
Ursula
Henderson
Doug
Hudson
Ron
Hutchins
Patricia
Kovatch
Bruce
Loftis
Nathaniel
Mendoza
Jeremy
Meredith
Terry
Moore
Tracy
Rafferty
Don
Reed
Jim
Rogers
Philip
Roth
Arlene
Washington
Sudha
Yalamanchili
Keeneland will enable transformational science for those
applications currently limited by node level parallelism
and memory bandwidth


Node
-
level extreme fine
-
grained
parallelism and memory
bandwidth from GPUs can
transform applications that
cannot benefit directly from
scaling up


Recent applications successes
on GPUs:


Molecular modeling (NAMD, VMD,
OpenMM
, GROMACS, AMBER)


Materials modeling (DCA++,
QMCPACK, LAMMPS)


Combustion (S3D)


GPUs are setting a new trajectory
for HPC architectures by
providing very high energy
efficiency and density

11

NAMD

DCA++

S3D