getting cloudy with remote graphics and gpu compute using g2 instances (cpn210) | aws re:invent 2013

63
© 2013 Amazon.com, Inc. and its affiliates. All rights reserved. May not be copied, modified, or distributed in whole or in part without the express consent of Amazon.com, Inc. GPU Instances on Amazon EC2 John Phillips, Sr. Product Manager, Amazon EC2 November 15, 2013

Upload: amazon-web-services

Post on 10-May-2015

625 views

Category:

Technology


4 download

DESCRIPTION

Amazon EC2 now offers a new GPU instance capable of running graphics and GPU compute workloads. In this session, we take a deeper look at the remote graphics capabilities of this new GPU instance, the tooling required to get started, and a live demo of applications streamed from our West Coast regions. We also explore the benefits of hosting your 3D graphics applications in the AWS cloud, where you can harness the vast compute and storage resources.

TRANSCRIPT

Page 1: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

© 2013 Amazon.com, Inc. and its affiliates. All rights reserved. May not be copied, modified, or distributed in whole or in part without the express consent of Amazon.com, Inc.

GPU Instances on Amazon EC2

John Phillips, Sr. Product Manager, Amazon EC2

November 15, 2013

Page 2: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Agenda

Overview of GPU Instances

Gyuri Ordody with Autodesk: Evolution of CAD on AWS

Teng Lin with Schrodinger: Drug Discovery on AWS

Questions from audience (if there’s time)

Page 3: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Instance Types Today

2006 2007 2008 2009 2010 2011 2012 2013

m1.small

m1.xlarge

m1.large

m1.small

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

cc2.8xlarge

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

hs1.8xlarge

m3.xlarge

m3.2xlarge

hi1.4xlarge

m1.medium

cc2.8xlarge

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

c3.large

c3.xlarge

c3.2xlarge

c3.4xlarge

c3.8xlarge

i2.large

i2.xlarge

i2.2xlarge

i2.4xlarge

i2.8xlarge

g2.2xlarge

cr1.8xlarge

hs1.8xlarge

m3.xlarge

m3.2xlarge

hi1.4xlarge

m1.medium

cc2.8xlarge

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

new

existing Entry into GPU

space

G2 Instances

Page 4: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Why GPUs? Parallel Performance

Product Example CPU Example GPU

Coprocessor

Processing cores 8 2,688

Clock frequency 2.6GHz 732MHz

Memory bandwidth 51.2 GB/s / socket 250GB/s (DDR5)

Peak Gflops (single) 333* 3,950**

Peak Gflops (double) 166* 1,310***

Total Memory >>4GB 6GB

* 256-bit AVX addition + 256 AVX multiplication /cycle/core ** 32-bit FMA /cycle/core *** 64-bit FMA /cycle/2core

Page 5: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

cg1.4xlarge

2 x NVIDIA GF104 GPU (Fermi / Tesla)

Intel Xeon X5570

16 vCPUs, 22.5 GiB of RAM

2 x 840 GB storage

10 Gbps NIC

Page 6: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Customer Feedback

Page 7: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

g2.2xlarge

1 NVIDIA GK104 GPU (Kepler / GRID)

2.6 GHz Sandy Bridge CPU w/ Turbo enabled

8 vCPUs, 15 GiB of RAM

60GB SSD storage

EBS-Optimized up to 1Gbps

Frame Capture and Encoding APIs

Page 8: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

$0.65 per hour

Page 9: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Why remote graphics in Amazon EC2?

Accessibility

Quality of service

Business agility

Collaboration

Data security

And…

Page 10: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

AWS Under Your Desk

Page 11: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

© 2013 Amazon.com, Inc. and its affiliates. All rights reserved. May not be copied, modified, or distributed in whole or in part without the express consent of Amazon.com, Inc.

GPU Instances on AWS:

Desktop Apps Are The New Web Apps

Gyuri Ordody, Autodesk

November 15, 2013

Page 12: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

About Autodesk

Autodesk started more than 30 years ago,

with 16 employees and one software title.

12

Page 13: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Today: an industry leader in design software for the building, manufacturing,

infrastructure and entertainment industries

About Autodesk

13

Page 14: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Architecture, Engineering and Construction

Image courtesy of Castro Mello Architects Image courtesy of Castro Mello Architects

14

Page 15: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Image courtesy of Brimrock Group Inc. and Mechanix Design Solutions Inc.

Digital Prototyping

15

Page 16: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Media & Entertainment

16

Page 17: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

17

CAD Evolution

IBM PC 5150 with keyboard and green monochrome monitor (5151), running MS-DOS 5.0 © Boffy

b

Page 18: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Lines, Arcs, Circles Features, Shapes, Blocs Intelligent Objects

18

CAD Evolution

Page 19: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Image courtesy of Hunt Construction Group and SHoP

19

Huge Datasets

Page 20: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Simulation, Analysis

20

Page 21: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

21

Page 22: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Design = Visualization

• High-end desktop

workstation – CPU (Xeon multicore)

– RAM (16GB)

– GPU (DirectX 9-11)

– Fast Disk

22

Page 23: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Design = Collaboration

23

Page 24: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

+

Design = Collaboration

24

Page 25: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Design Graph +

Design = Collaboration

25

Page 26: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Strategies

• Create new cloud services on server clusters – Write or rewrite from scratch

• Move desktop technology to headless server

technology – EC2 instances and Amazon S3 as

backend – Recreate UI functionality in the browser

• Deploy existing desktop apps in the cloud – Reuse engine and GUI

26

Page 27: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

27

Page 28: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Collaboration – Using the AWS Cloud

• Access it anywhere

• Access using any device

• Seamless collaboration

• Editing from anywhere

• Data close to application

28

Page 29: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Desktop Apps Home / Office

Player

Application Remoting Overview

29

Bitmap / Video

Keyboard,

mouse, USB

Internet

EC2 Instance

Page 30: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

EC2 Instance

Home / Office

Player

Application Remoting Overview

Bitmap / Video

Keyboard,

mouse, USB

Internet

30

Desktop Apps

Page 31: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

EC2 Instance

Home / Office

Player

Application Remoting Overview

Bitmap / Video

Keyboard,

mouse, USB

Internet

31

Desktop Apps

Page 32: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Autodesk Online Application - Architecture

Internet

User

Client

32

Region 1

Controller

EC2 Instance

Region 2

Controller

EC2 Instance

Region N

Controller

EC2 Instance

Page 33: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Autodesk Online Application - Architecture

Internet

User

Client

33

Application Settings

Default User Data

SimpleDB

User Data

Session Data

Region

App Servers

EC2 Instances

App Server AMI

Connection

Controller

EC2 Instance

S3

Custom Scaling

Page 34: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Autodesk Desktop Apps

EC2 Instance

Home / Office 1

Application Remoting – Instance sharing

Internet

34

Home / Office n

Internet

Page 35: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Data Sets

Desktop Applications – Data Exchange N GB/exchange

35

Page 36: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Data references

Cloud data exchange – Predictable Data Traffic

Data references

N kb/exchange

N kb/exchange

36

Page 37: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Data references

Cloud data exchange – Predictable Data Traffic

N kb/exchange

37

Data

references +

Video Stream

1 GB/hr

Page 38: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

A360 and the AWS Cloud

Supporting

Infrastructure Identity

EC2 + RDS

Storage

S3 DynamoDB

Simulation

EC2 EMR

Analysis

EC2 EMR RDS

38

Page 39: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Application + Cloud Integration

User

Ec2 + RDB

Autodesk

Identity

Service

S3 +

DB

Autodesk

Storage

Service

AWS Cloud

Autodesk

Desktop Apps

Internet

Client

39

Page 40: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Application + Cloud Integration

User

EC2 + RDB

Identity

S3 +

DB

Storage

AWS Cloud

Internet

Client

Autodesk

Desktop

Apps

EC2

No

GPU

Autodesk

App

Player

Autodesk

Desktop Apps

40

Page 41: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Application + Cloud Integration

User

EC2 + RDB

Identity

S3 +

DB

Storage

AWS Cloud

Internet

Client

Autodesk

Desktop

Apps

EC2

Autodesk

App Player

GPU!

41

Page 42: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Autodesk Apps in the Cloud Without Amazon EC2 - GPU Instances

42

Page 43: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Autodesk Apps in the Cloud With Amazon EC2 - GPU Instances

43

Page 44: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Application + Cloud Integration

User

EC2 + RDB

Identity

S3 +

DB

Storage

AWS cloud

Internet

Client

Autodesk

Desktop

Apps

EC2

Autodesk

App Player

GPU!

44

Page 45: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Application + Cloud Integration

User

EC2 + RDB

Identity

S3 +

DB

Storage

AWS cloud

Internet

Client

Autodesk

Desktop

Apps

EC2

* GPU!

45

Page 46: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Application + Cloud Integration

Identity

Storage

AWS cloud

Autodesk

Desktop

Apps

EC2

Instances

Search

Translation

Simulation Analysis

GPU!

Supporting

Infrastructure

Internet

User

Client

*

46

Page 47: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Demo: Running Design Apps

In a Browser

47

Demos

• Live

• YouTube: http://www.youtube.com/watch?v=lU85EjvTyz0

Page 48: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

© 2013 Amazon.com, Inc. and its affiliates. All rights reserved. May not be copied, modified, or distributed in whole or in part without the express consent of Amazon.com, Inc.

Drug Discovery on AWS

Teng Lin, Senior Principal Scientist, Schrödinger

November 15, 2013

Page 49: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Drug discovery and development stages

• It takes $800 Million to $1 Billion and 10 to 15

years to develop a blockbuster

http://www.innovation.org/drug_discovery/objects/pdf/RD_Brochure.pdf

Page 50: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Simple facts about drug discovery

• Each development candidate has a value of

$50-100M

• But the overhead of producing these in

pharmaceutical company is $35-70M – Success rate is only 1 in 3

– Thousands of molecules synthesized

• Pharmaceutical Industry needs to overcome the

innovation deficit in drug discovery process

Page 51: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Schrödinger

• Providing software solutions and services for life

sciences and materials research

Page 52: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Ligand-protein binding

• Altering receptor protein conformation, and

consequently changing biological functions.

• Binding affinity is critical for drug discovery

Yibing Shan etal Journal of the American Chemical Society, vol. 133, no. 24, 2011, pp. 9181–9183.

Page 53: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Free Energy Perturbation (FEP)

• Schrödinger’s FEP product – Can predict binding affinity very accurately

• Key features – Better sampling algorithm

– High quality Force Field

– Perturbation network

– Automated workflow

– GPU support

– Cloud capable

Page 54: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

GPU is significantly faster

• Each edge takes 3 or more days on 96 cores – Slow and unreliable due to cross node communication

– Perturbation network makes it even worse

18.5

109.8

60.8

86.4

4.9

26.3

15.2 21.0

0.0

20.0

40.0

60.0

80.0

100.0

120.0

8 x Intel Xeon X5672 GeForce GTX780 Amazon Tesla M2050 Amazon Geforce GridK520

Sp

ee

d (

ns

/da

y)

DHFR

APOA1

Page 55: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

How can FEP help drug discovery?

• Traditional drug design – Takes weeks or even months to synthesize a compound

– Costs $1,000 to $5,000 per compound

– Synthesize thousands of compounds per project

• In-silico design using FEP – Takes 72 GPU hours (~6 hours per calculation with 12 GPUs)

– Costs about $75, and the price keeps going down

– “want to do 1000 calculations per day”

Page 56: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Why AWS?

• Scalability – Performed virtual screening using 50,000-core on AWS

• Security

• Price per FEP job – It takes us two months to get GTX-780 cluster up running

$16.61

$28.28

$12.67

$32.01

$75.60

$3.78

$17.62

$32.76

$0.00

$20.00

$40.00

$60.00

$80.00

GTX-780 (50%util)

Tesla K20 (70%util)

Spot InstanceCG1

3-yr HEAVYreserved (100%

util) CG1

On-demandInstance CG1

Spot InstanceG2

3-yr HEAVYreserved (100%

util) G2

On-demandInstance G2

Page 57: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

• Next version will be cloud oriented

• Data will be processed and visualized on cloud

FEP on cloud

Client

Mobile Client

Amazon EC2

Web Servers

Traditional ServerCorporate

Data Center

VPN Gateway VPN Connection

VPN

Connection

DB Instance ClusterGPU Cluster

Internet

Gateway

Auto Scaling

Auto Scaling

Page 58: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Retrospective Study

-14

-13

-12

-11

-10

-9

-8

-7

-6

-5

-4

-14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4

y = 1.07x + 0.652; R² = 0.599

(Linear regression to all ~150 ligands

across multiple systems)

Binding Affinity (kcal/mol)

Bin

din

g A

ffin

ity

Pre

dic

tio

n (

kc

al/

mo

l)

Page 59: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

• 9 out of top 10 are active compounds – Probability of achieving result this good is <1%

– “make half as many compounds”

– “save years of time on the project”

• Company X signed a contract after the test

6 3 1

19

88

66

59

48

0

10

20

30

40

50

60

70

80

90

Highest Lowest

Co

un

t

Binding Affinity

Blind test with company X on AWS

Page 60: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Blind test with company Y on AWS

• 8 out of top 10 are the most active compounds – Probability of achieving result this good is <1%

• Company Y wants us to provide a turn key solution

8 1 1

19

10 10

4

1

0

2

4

6

8

10

12

14

16

18

20

Highest Lowest

Co

un

t

Binding Affinity

Page 61: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Prospective FEP with Company Z on AWS

• 1/3 of molecules are active instead of 1/7

• Company Z uses FEP on many projects

10

23

4

19

9

4 4

2 1 1

0 0 0 0 1

2

0 1

0

5

10

15

20

25

Highest Lowest

Co

un

t

Binding Affinity

Non-FEP

FEP predict to be active

FEP predict to be inactive

Page 62: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Summary

• Computer aid drug design plays a critical role in drug discovery

• Combining with GPU computing, accurate modeling tools like FEP will accelerate the drug discovery process

• Cloud is a viable solution for high performance computing, in terms of pricing and scalability

• Amazon is the leader for GPU computing at cloud

Page 63: Getting Cloudy with Remote Graphics and GPU Compute Using G2 instances (CPN210) | AWS re:Invent 2013

Please give us your feedback on this

presentation

As a thank you, we will select prize

winners daily for completed surveys!

CPN210