Accelerating Scientific Discovery V1

Accelerating Scientific Discovery using GPU Clusters http://www.nvidia.com/tesla

2

World’s Fastest Molecular Dynamics Simulation

Sustained Performance of 1.87 Petaflops/s Institute of Process Engineering (IPE)

Chinese Academy of Sciences (CAS)

Simula'on  for  Crystalline  Silicon  Used  for  Photovoltaic  cells  &  Semiconductors  

Used  all  7168  Tesla  GPUs  on    Tianhe-­‐1A  GPU  Supercomputer  

3

World’s First Whole H1N1 Virus Simulation

More accurate & complete model

Furthers understanding of drug interactions

Mole-8.5 GPU Supercomputer

at CAS-IPE

4

ASUCA TeraFlop Scaling (Weather Modeling)

3990 Tesla M2050s

145.0 Tflops SP

76.1 Tflops DP

Simulation on Tsubame 2.0, TiTech Supercomputer

After GPUs

Before GPUs

5

2011 Gordon Bell Prize Winner Tsubame 2.0 GPU Supercomputer

“Peta-scale Phase-Field Simulation for Dendritic Solidification on the TSUBAME 2.0 Supercomputer”

-- Shimokawabe et. al. Science Impact

Developing lightweight material for fuel efficient cars

6

Forecasting Heart Attacks Tsubame 2.0 GPU Supercomputer

Plaque rupture leads to heart attack Forecast where/when plaques form

7

Metagenomics Tsubame 2.0 GPU Supercomputer

BLASTX: Standard CPU Software GHOSTM: GPU-based Software compatible with BLASTX

8

NAMD Scaling on Tsubame 2.0

9

LAMMPS: Billion Atoms Simulation

Test  Pla)orm:    NCSA  Lincoln  Cluster  with  S1070  1U  GPU  servers  a?ached      CPU-­‐only  Cluster-­‐  Cray  XT5  

Billion  Atom  Lennard-­‐Jones  Benchmark  

29  Seconds  

103  Seconds  

288  GPUs  +  CPUs   1920  x86  CPUs  

10

Protein-DNA Docking

Dr. Bo Hong, George Tech Dr. Juntao Guo, UNC Charlotte

Improving Prediction Accuracy of Protein-DNA Docking with GPU Computing, Best Paper Award, IEEE BIBM 2011

11

Strong Scaling LQCD: Chroma & MILC 256 GPUs outperform 8K CPU cores

Chroma 3.41.0 using GCR-DD solver MILC 7.6.3 using mixed-precision CG solver

Guochun Shi (NCSA), Balint Joo (Jefferson Labs), Ron Babich (BU), Mike Clark (Harvard), Rich Brower (BU), Steve Gottlieb (Indiana), “Scaling Lattice QCD beyond 100 GPUs,” SC11, ACM (Nov 2011)

12

Computational Fluid Dynamics Scaling on GPUs

Boise State Univeristy, Jacobsen, Thibault, Senocak 48th AIAA Aerospace Sciences Meeting, January 4-7, 2010

Incompressible Flow Computations, Navier-Stokes 64 Compute nodes with 128 M1060 GPUs

33

69

102

228

428

854

1478

2432

20

200

2000

1 2 4 8 16 32 64 128

GFL

OPS

(lo

gari

thm

ic)

Number of GPUs

2.4 Tflops

128 GPUs

11x Speedup with GPUs

Navier Stokes (Weak Scaling) in GFLOPS

13

Titan at Oak Ridge World’s Top Open Science Computing Research Facility

2x Faster, 3x More Energy Efficient than Current #1 (K Computer)

18,000 Tesla GPUs

20+ PetaFlops

~90% of flops from GPUs

14

NCSA Mixes GPUs into Blue Waters

NCSA  is  excited  about  the  inclusion  of  NVIDIA's  Tesla  GPUs  in  Blue  Waters.    GPUs  provide  extraordinary  capabiliWes  for  numerically-­‐intensive  computaWons  and  a  cost-­‐effecWve,  energy-­‐efficient  way  to  build  tomorrow's  petascale  supercomputers.  

“  

”  Thom  Dunning  Director,  NCSA