Formation du Club des Affiliés du LAAS-CNRS, Toulouse, 22 mars 2016

Frédéric Parienté, Tesla Accelerated Computing, NVIDIA

GPU EN CALCUL SCIENTIFIQUE

2

ENTERPRISE AUTOGAMING DATA CENTERPRO VISUALIZATION

THE WORLD LEADER IN VISUAL COMPUTING

3

FIVE THINGS TO REMEMBER

Time of accelerators has come

NVIDIA is focused on co-design from top-to-bottom

Accelerators are surging in supercomputing

Machine learning is the next killer application for HPC

Tesla platform leads in every way

4

“It’s time to start planning for the end of Moore’s Law, and it’s worth pondering how it will end, not just when.”

Robert Colwell Director, Microsystems Technology Office, DARPA

5

TESLA ACCELERATED COMPUTING PLATFORMFocused on Co-Design from Top to Bottom

ProductiveProgrammingModel & Tools

Expert Co-Design

Accessibility

APPLICATION

MIDDLEWARE

SYS SW

LARGE SYSTEMS

PROCESSOR

Fast GPUEngineered for High Throughput

0,0

0,5

1,0

1,5

2,0

2,5

3,0

2008 2009 2010 2011 2012 2013 2014

NVIDIA GPU x86 CPUTFLOPS

M2090

M1060

K20

K80

K40

Fast GPU+

Strong CPU

6

0

25

50

75

100

125

2013 2014 2015

ACCELERATORS SURGE IN WORLD’S TOP SUPERCOMPUTERS

100+ accelerated systems now on Top500 list

1/3 of total FLOPS powered by accelerators

NVIDIA Tesla GPUs sweep 23 of 24 new accelerated supercomputers

Tesla supercomputers growing at 50% CAGR over past five years

Top500: # of Accelerated Supercomputers

7

70% OF TOP HPC APPS ACCELERATEDTOP 25 APPS IN SURVEYINTERSECT360 SURVEY OF TOP APPS

GROMACSSIMULIA AbaqusNAMDAMBERANSYS MechanicalMSC NASTRANSPECFEM3D

ANSYS FluentWRFVASPOpenFOAMCHARMMQuantum Espresso

LAMMPSNWChemLS-DYNASchrodinger

GaussianGAMESS

ANSYS CFXStar-CDCCSMCOMSOLStar-CCM+BLAST

= All popular functions accelerated

Top 10 HPC Apps90%

Accelerated

Top 50 HPC Apps70%

Accelerated

Intersect360, Nov 2015“HPC Application Support for GPU Computing”

= Some popular functions accelerated= In development= Not supported

8

370 GPU-Accelerated Applications

www.nvidia.com/appscatalog

9

TESLA BOOSTS DATACENTER THROUGHPUT$500M Datacenter, 4x increase in ROI

1000 Jobs Per Day

70%GPU-Accelerated

Nodes

30%CPU Nodes

3800 Jobs Per Day

100%CPU Nodes

70% of Applications5x Faster with GPU

10

U.S. Dept. of EnergyPre-Exascale Supercomputers

for Science

NEXT-GEN SUPERCOMPUTERS ARE GPU-ACCELERATED

NOAANew Supercomputer for Next-Gen

Weather Forecasting

IBM WatsonBreakthrough Natural Language

Processing for Cognitive Computing

SUMMIT

SIERRA

11

MACHINE LEARNINGHPC 1ST CONSUMER KILLER-APP

MICROSOFT OPEN-SOURCE

DMTK

GOOGLE OPEN-SOURCETENSORFLOW

YOUTUBE CLICK-TO-BUY ADS GOOGLE PHOTO

MICROSOFT CORTANAFACEBOOK MESSENGER FACIAL RECOGNITION

12

TESLA PLATFORM LEADS IN EVERY WAY

PROCESSOR INTERCONNECT

ECOSYSTEMSOFTWARE

13

TESLA PLATFORM FOR HPC

14

“Approximately a third of HPC systems operating today are equipped with accelerators and nearly half of all newly deployed systems have them.”

Source: ACCELERATED COMPUTING: A TIPPING POINT FOR HPC Intersect360 Nov 2015

15

TESLA FOR SIMLUATION

LIBRARIES

TESLA ACCELERATED COMPUTING

LANGUAGESDIRECTIVES

ACCELERATED COMPUTING TOOLKIT

16

Tesla Accelerates Discoveries

Using a supercomputer powered by the Tesla Platform with over 3,000 Tesla accelerators, University of Illinois scientists performed the first all-atom simulation of the HIV virus and discovered the chemical structure of its capsid —“the perfect target for fighting the infection.”

Without GPU, the supercomputer would need to be 5x larger for similar performance.

17

TESLA K80World’s Fastest Accelerator for HPC & Data Analytics

0 5 10 15 20 25 30

Tesla K80 Server

Dual CPU Server

# of Days

AMBER Benchmark: PME-JAC-NVE Simulation for 1 microsecondCPU: E5-2698v3 @ 2.3GHz. 64GB System Memory, CentOS 6.2

CUDA Cores 4992

Peak DP 1.9 TFLOPS

Peak DP w/ Boost 2.9 TFLOPS

GDDR5 Memory 24 GB

Bandwidth 480 GB/s

Power 300 W

GPU Boost Dynamic

Simulation Time from 1 Month to 1 Week

5x FasterAMBER Performance

18

0x

5x

10x

15x

K80 CPU

TESLA K80: 10X FASTER ON REAL-WORLD APPS

CPU: 12 cores, E5-2697v2 @ 2.70GHz. 64GB System Memory, CentOS 6.2GPU: Single Tesla K80, Boost enabled

Quantum ChemistryMolecular Dynamics PhysicsBenchmarks

19

TESLA K80 BOOSTS DATA CENTER THROUGHPUT

ACCELERATING KEY APPS 1/3 OF NODES ACCELERATED, 2X SYSTEM THROUGHPUT

100 Jobs Per Day 220 Jobs Per Day

CPU-only System Accelerated System

0x

5x

10x

15x

QMCPACK LAMMPS CHROMA NAMD AMBER

K80 CPU

CPU: Dual E5-2698 v3@2.3GHz 3.6GHz, 64GB System Memory, CentOS 6.2GPU: Single Tesla K80, Boost enabled

Speed-up vs Dual CPU

20

TESLA FOR VISUALIZATION

IRAY

TESLA ACCELERATED COMPUTING

INDEXOPTIX

VISUALIZATION TOOLS FOR HPC

21

VISUALIZE DATA INSTANTLY FOR FASTER SCIENCE

TraditionalSlower Time to Discovery

CPU Supercomputer Viz Cluster

Simulation- 1 Week Viz- 1 Day

Multiple Iterations

Time to Discovery = Months

Tesla PlatformFaster Time to Discovery

GPU-Accelerated Supercomputer

Visualize while you simulate/without

data transfers

Restart Simulation InstantlyMultiple Iterations

Time to Discovery = Weeks

Flexible

Scalable

Interactive

Days

Data Transfer

22

VISUALIZATION-ENABLED SUPERCOMPUTERSSimulation + Visualization

CSCS Piz Daint NCSA Blue Waters

Galaxy Formation

ORNL Titan

Molecular Dynamics Cosmology

23

GROWING ADOPTION IN CLIMATE & WEATHER

MeteoSwiss Deploys World’s First Accelerated Weather Supercomputer

2x higher resolution for daily forecasts

14x more simulation with ensemble approach for medium-range forecasts

NOAA Chooses Tesla To Improve Weather Forecast Research

Develop global model with 3km resolution, five-fold increase from today’s resolution

Improved resolution requires 100x computational complexity

24

U.S. TO BUILD TWO FLAGSHIP SUPERCOMPUTERSPowered by the Tesla Platform

100-300 PFLOPS Peak

10x in Scientific App Performance

IBM POWER9 CPU + NVIDIA Volta GPU

NVLink High Speed Interconnect

40 TFLOPS per Node, >3,400 Nodes

2017

Major Step Forward on the Path to Exascale

25

IBM POWER CPUMost Powerful Serial Processor

NVIDIA NVLinkFastest CPU-GPU Interconnect

NVIDIA Volta GPUMost Powerful Parallel Processor

ACCELERATED COMPUTING DELIVERS 5X HIGHER ENERGY EFFICIENCY

80-200 GB/s

26

CORAL: BUILT FOR GRAND SCIENTIFIC CHALLENGES

Fusion EnergyRole of material disorder, statistics, and fluctuations in nanoscale materials and systems

CombustionCombustion simulations to enable the next gen diesel/bio-fuels to burn more efficiently

Climate Change Study climate change adaptation and mitigation scenarios; realistically represent detailed features

Nuclear EnergyUnprecedented high-fidelity radiation transport calculations for nuclear energy applications

BiofuelsSearch for renewable and more efficient energy sources

AstrophysicsRadiation transport – critical to astrophysics, laser fusion, atmospheric dynamics, and medical imaging

27

TESLA PLATFORM FOR MACHINE LEARNING

28

THE BIG BANG IN MACHINE LEARNING

“ Google’s AI engine also reflects how the world of computer hardware is changing. (It) depends on machines equipped with GPUs… And it depends on these chips more than the larger tech universe realizes.”

DNN GPUBIG DATA

29

Tesla Revolutionizes Machine Learning

GOOGLE BRAIN APPLICATION – DEEP LEARNING

BEFORE TESLA AFTER TESLA

Cost $5,000K $200K

Servers 1,000 Servers 16 Tesla Servers

Energy 600 KW 4 KW

Performance 1x 6x

30

THE AI RACE IS ON

31

NVIDIA GPU THE ENGINE OF DEEP LEARNING

NVIDIA CUDAACCELERATED COMPUTING PLATFORM

WATSON CHAINER THEANO MATCONVNET

TENSORFLOW CNTK TORCH CAFFE

32

CUDA BOOSTS DEEP LEARNING

5X IN 2 YEARS

Perf

orm

ance

AlexNet training throughput based on 20 iterations, CPU: 1x E5-2680v3 12 Core 2.5GHz. 128GB System Memory, Ubuntu 14.04

Caffe Performance

K40K40+cuDNN1

M40+cuDNN3

M40+cuDNN4

0

1

2

3

4

5

6

11/2013 9/2014 7/2015 12/2015

33

39%

45%

55%

62%66%

72%75%

79%83%

86%87,5%

30%

40%

50%

60%

70%

80%

90%

100%

72%74%

84%

88%

93%

96%

65%

70%

75%

80%

85%

90%

95%

100%

2010 2011 2012 2013 2014 2015

ImageNet Accuracy

NVIDIA GPU

AMAZING RATE OF IMPROVEMENTPedestrian Detection

CALTECH

65%

70%

75%

80%

85%

90%

95%

100%

11/2013 6/2014 12/2014 7/2015 1/2016

Accu

racy

CV-based DNN-based

Object Detection

KITTI

Image Recognition

IMAGENET

Top Score

NVIDIA DRIVENet

34

CUDA FOR DEEP LEARNING DEVELOPMENT

TITAN X GPU CLOUDDEVBOX

DEEP LEARNING SDK

cuSPARSE cuBLASDIGITS NCCLcuDNN

35

FACEBOOK’S DEEP LEARNING MACHINEPurpose-Built for Deep Learning Training

2x Faster Training for Faster Deployment

2x Larger Networks for Higher Accuracy

Powered by Eight Tesla M40 GPUs

Open Rack Compliant

Serkan PiantinoEngineering Director of Facebook AI Research

“Most of the major advances in machine learning and AI in the past few years have been contingent on tapping into powerful

GPUs and huge data sets to build and train advanced models”

36

DESIGNED FOR AI COMPUTING AT LARGE SCALE

Built on the NVIDIA Tesla Platform

• 8 Tesla M40s deliver aggregate 96 GB GDDR5 memory and 56 teraflops of SP performance

• Leverages world’s leading deep learning platform to tap into frameworks such as Torch and libraries such as cuDNN

Operational Efficiency and Serviceability

• Free-air Cooled Design Optimizes Thermal and Power Efficiency

• Components swappable without tools

• Configurable PCI-e for versatility

37

TESLA M40World’s Fastest Accelerator for Deep Learning Training

0 1 2 3 4 5

GPU Server with4x TESLA M40

Dual CPU Server

13x Faster TrainingCaffe

Number of Days

CUDA Cores 3072

Peak SP 7 TFLOPS

GDDR5 Memory 12 GB

Bandwidth 288 GB/s

Power 250W

Reduce Training Time from 5 Days to less than 10 Hours

Note: Caffe benchmark with AlexNet, training 1.3M images with 90 epochsCPU server uses 2x Xeon E5-2699v3 CPU, 128GB System Memory, Ubuntu 14.04

38

TESLA M4Highest Throughput

Hyperscale Workload Acceleration

CUDA Cores 1024

Peak SP 2.2 TFLOPS

GDDR5 Memory 4 GB

Bandwidth 88 GB/s

Form Factor PCIe Low Profile

Power 50 – 75 W

Video Processing

4x

Image Processing

5x

Video Transcode

2x

Machine Learning Inference

2x

H.264 & H.265, SD & HD

Stabilization and Enhancements

Resize, Filter, Search, Auto-Enhance

Preliminary specifications. Subject to change.

39

TESLA PLATFORM FOR DEVELOPERS

40

10X GROWTH IN ACCELERATED COMPUTING20152008

3 MillionCUDA Downloads

150,000CUDA Downloads

60,000 Academic Papers

4,000Academic

Papers

800Universities Teaching

60Universities

Teaching

54,000SupercomputingTeraflops

77Supercomputing

Teraflops

450,000Tesla GPUs

6,000Tesla GPUs

370CUDA Apps

27CUDA Apps

41

HOW GPU ACCELERATION WORKSApplication Code

+

GPU CPU5% of Code

Compute-Intensive Functions

Rest of SequentialCPU Code

42

COMMON PROGRAMMING MODELS ACROSS MULTIPLE CPUS

x86

Libraries

Programming Languages

CompilerDirectives

AmgXcuBLAS

/

43

GPU ACCELERATED LIBRARIES“Drop-in” Acceleration for Your Applications

Domain-specificDeep Learning, GIS, EDA,Bioinformatics, Fluids

Visual ProcessingImage & Video

Linear AlgebraDense, Sparse, Matrix

Math AlgorithmsAMG, Templates, Solvers

NVIDIA cuRAND

NVIDIANPPNVIDIA CODEC SDK

NVBIO Triton Ocean SDK

NVIDIAcuBLAS, cuSPARSE

AmgX cuSOLVER

developer.nvidia.com/gpu-accelerated-libraries

44

OpenACCSimple | Powerful | Portable

Fueling the Next Wave of

Scientific Discoveries in HPC

University of IllinoisPowerGrid- MRI Reconstruction

70x Speed-Up2 Days of Effort

http://www.cray.com/sites/default/files/resources/OpenACC_213462.12_OpenACC_Cosmo_CS_FNL.pdfhttp://www.hpcwire.com/off-the-wire/first-round-of-2015-hackathons-gets-underwayhttp://on-demand.gputechconf.com/gtc/2015/presentation/S5297-Hisashi-Yashiro.pdf

http://www.openacc.org/content/experiences-porting-molecular-dynamics-code-gpus-cray-xk7

RIKEN JapanNICAM- Climate Modeling

7-8x Speed-Up5% of Code Modified

main() {

<serial code>#pragma acc kernels//automatically runs on GPU

{ <parallel code>

}}

8000+

Developers

using OpenACC

45

Janus Juul Eriksen, PhD FellowqLEAP Center for Theoretical Chemistry, Aarhus University

“OpenACC makes GPU computing approachable for domain scientists. Initial OpenACC implementation required only minor effort, and more importantly,no modifications of our existing CPU implementation.

“

Lines of Code Modified

# of Weeks Required

# of Codes to Maintain

<100 Lines 1 Week 1 Source

Big Performance

0,0x

4,0x

8,0x

12,0x

Alanine-113 Atoms

Alanine-223 Atoms

Alanine-333 Atoms

Spee

dup

vs C

PU

Minimal Effort

LS-DALTON CCSD(T) ModuleBenchmarked on Titan Supercomputer (AMD CPU vs Tesla K20X)

LS-DALTONLarge-scale Application for Calculating High-accuracy

Molecular Energies

46

OPENACC DELIVERS TRUE PERFORMANCE PORTABILITY

Paving the Path Forward: Single Code for All HPC Processors

4,1x 5,2x7,1x

4,3x 5,3x7,1x7,6x

11,9x

30,3x

0x

5x

10x

15x

20x

25x

30x

35x

359.MINIGHOST (MANTEVO) NEMO (CLIMATE & OCEAN) CLOVERLEAF (PHYSICS)

CPU: MPI + OpenMP CPU: MPI + OpenACC CPU + GPU: MPI + OpenACC

Spee

dup

vs S

ingl

e CP

U C

ore

Application Performance Benchmark

359.miniGhost: CPU: Intel Xeon E5-2698 v3, 2 sockets, 32-cores total, GPU: Tesla K80- single GPU NEMO: Each socket CPU: Intel Xeon E5-‐2698 v3, 16 cores; GPU: NVIDIA K80 both GPUsCLOVERLEAF: CPU: Dual socket Intel Xeon CPU E5-2690 v2, 20 cores total, GPU: Tesla K80 both GPUs

47

CUDASuper Simplified Memory Management Code

void sortfile(FILE *fp, int N) {char *data;data = (char *)malloc(N);

fread(data, 1, N, fp);

qsort(data, N, 1, compare);

use_data(data);

free(data);}

void sortfile(FILE *fp, int N) {char *data;cudaMallocManaged(&data, N);

fread(data, 1, N, fp);

qsort<<<...>>>(data,N,1,compare);cudaDeviceSynchronize();

use_data(data);

cudaFree(data);}

CPU Code CUDA 6 Code with Unified Memory

48

GPU DEVELOPER ECO-SYSTEM

Consultants & Training

ANEO GPU Tech

OEM Solution Providers

Debuggers& Profilers

CUDA-GDBNV Visual Profiler

NVIDIA NsightVisual Studio

AllineaTotalView

MATLABMathematica

LabView

Numerical Packages

GPUDirect RDMADatacenter

GPU Manager

Cluster Tools

FFTBLAS

SPARSELAPACK

NPPVideo

Imaging

LibrariesCC++

FortranJava

PythonOpenACCOpenMP

Languages & Directives

49

DEVELOP ON GEFORCE, DEPLOY ON TESLA

Designed for Developers & Gamers

Available Everywhere

developer.nvidia.com/cuda-gpusdeveloper.nvidia.com/devbox

Designed for the Data CenterECC

24x7 RuntimeGPU Monitoring

Cluster ManagementGPUDirect-RDMAHyper-Q for MPI3 Year Warranty

Integrated OEM Systems, Professional Support

51

DEEP LEARNING & ARTIFICIAL INTELLIGENCE

Sep 28-29, 2016 | Amsterdamwww.gputechconf.eu #GTC16

SELF-DRIVING CARS VIRTUAL REALITY & AUGMENTED REALITY

SUPERCOMPUTING & HPC

GTC Europe is a two-day conference designed to expose the innovative ways developers, businesses and academics are using parallel computing to transform our world.

EUROPE’S BRIGHTEST MINDS & BEST IDEAS

2 Days | 800 Attendees | 50+ Exhibitors | 50+ Speakers | 15+ Tracks | 15+ Workshops | 1-to-1 Meetings