gpu-accelerated design optimization on the cloud

38
GPU-Accelerated Design Optimization on the Cloud Krishnan Suresh Associate Professor Mechanical Engineering

Upload: others

Post on 25-May-2022

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: GPU-Accelerated Design Optimization on the Cloud

GPU-Accelerated

Design Optimization

on the Cloud

Krishnan Suresh

Associate Professor

Mechanical Engineering

Page 2: GPU-Accelerated Design Optimization on the Cloud

Design Optimization

2

Reduce weight

subject to constraints

(GE/GrabCAD)

A structure subject to loading

Design Optimization

Page 3: GPU-Accelerated Design Optimization on the Cloud

Domains

3(OptiStruct)

(Generico)

Page 4: GPU-Accelerated Design Optimization on the Cloud

4

Big Players, Big $

� ANSYS

� Abaqus

� Altair/OptiStruct

� Nastran

� SolidWorks

� &

$10 billion investment annually technavio.com

Page 5: GPU-Accelerated Design Optimization on the Cloud

Design Optimization on the Cloud

5

Browser driven design optimization

Page 6: GPU-Accelerated Design Optimization on the Cloud

Design Optimization on the Cloud

6

Client

• No software/hardware investment

• Pay as you go

• Anywhere, anytime

• &

Service Provider

• Easier maintenance

• Larger market

• &

Page 7: GPU-Accelerated Design Optimization on the Cloud

3D-Printing

Democratization of

fabrication

Page 8: GPU-Accelerated Design Optimization on the Cloud

Design Optimization to 3D Printing

8

Democratization of design

Page 9: GPU-Accelerated Design Optimization on the Cloud

Catch?

9

Page 10: GPU-Accelerated Design Optimization on the Cloud

Design Optimization

DesignSpace

Finite Element Analysis(FEA)

Optimal?

ChangeDesign

No

10^5 ~ 10^7 dof

Solve Kd = f

K: Sparse SPD

100’s of iterations!

10

Page 11: GPU-Accelerated Design Optimization on the Cloud

Design Optimization Cost

11

A naïve port

to cloud will not work!

� OptiStruct (commercial code)

� Xeon E5 2697, 92 GB

� 20 hours!

Page 12: GPU-Accelerated Design Optimization on the Cloud

Cloud Based Design Optimization

Fast Limited

Memory FEA

Pareto

Optimization

GPU

Acceleration

WebGL

Page 13: GPU-Accelerated Design Optimization on the Cloud

Fast Limited

Memory FEA

Cloud Based Design Optimization

Page 14: GPU-Accelerated Design Optimization on the Cloud

FEA Bottleneck: Kd = f

DesignSpace

Finite Element Analysis(FEA)

Optimal?

ChangeTopology

No

10^5 ~ 10^7 dof

Solve Kd = f

K: Sparse SPD

100’s of iterations!

14

==

Page 15: GPU-Accelerated Design Optimization on the Cloud

15

Kd = f (GTC)

� Fine-grained Parallel Preconditioners

� CULA

� MAGMA

� Accelerating Iterative Linear Solvers

� Efficient AMG on Hybrid GPU Clusters

� Preconditioning for Large-Scale Linear Solvers

� &

Page 16: GPU-Accelerated Design Optimization on the Cloud

� Exploit mesh congruency

� Exploit physics behavior

� K constantly changing

� &

Design Optimization

Kd = f

Page 17: GPU-Accelerated Design Optimization on the Cloud

17

Exploit Mesh Congruency (GTC 2014)

Kd f=

Model DiscretizeAssemble/

Solve

Post-process

Mesh-aware SpMV Acceleration: Congruence

Page 18: GPU-Accelerated Design Optimization on the Cloud

Element Congruency

18

62350 elements

2780 distinct

95.5% congruent

Observation: Large-meshes contain many similar elements!

Elements are ‘rigid-body/scaling’ congruent

⇒ Identical element stiffness Ke

Only store Ke of distinct elements

Page 19: GPU-Accelerated Design Optimization on the Cloud

Implication: SpMV

19

( )1

Classic: N

e

i

Kd K d=

≡ ∑

: Sparse Matrix-Vector Multiplication (SpMV)

Critical operation in ALL iterative solvers

Kd

( )1

Assembly-free: N

e e

i

Kd K d=

≡∑

Only store Ke of distinct elements + Assembly Free

Page 20: GPU-Accelerated Design Optimization on the Cloud

Experiment

20

106 Elements

1 Distinct element

0

200

400

600

800

1000

Assembled AF-CPU AF-GPU

SpMV; Kd (msec)

770

37

- Same number of FLOPS!

- Reduced memory

CPU

Mesh-aware Kd

Naïve Kd

One Kd (SpMV)

Page 21: GPU-Accelerated Design Optimization on the Cloud

21

Physics Aware Deflation

Kd f=

Model DiscretizeAssemble/

Solve

Post-process

Page 22: GPU-Accelerated Design Optimization on the Cloud

Physics Aware Deflation

0

TK W KW=%

0K K<<%

Agglomeration/Grouping

Treat each group as rigid body

Deflated CG

Kd f=

Page 23: GPU-Accelerated Design Optimization on the Cloud

Example

23

3.15 million DOF

Page 24: GPU-Accelerated Design Optimization on the Cloud

Fast Limited

Memory FEA

Pareto

Optimization

1. Mesh Congruency

2. AF Deflation

Cloud Based Design Optimization

Page 25: GPU-Accelerated Design Optimization on the Cloud

Design Optimization

K Matrix: Constantly changing

� Update K?

� Update deflation?

( )1

Assembly-free: N

e e

i

Kd K d=

≡∑Skip deleted finite elements

SpMV accelerates further

0

TK W KW=%

0 ( )T

eK K W K W= − ∆% %

K K<<%

Page 26: GPU-Accelerated Design Optimization on the Cloud

Example: Design Optimization

� OptiStruct (commercial)

� Xeon E5 2697, 92 GB

� 20 hours!

� Pareto

� I7 4770, 8 GB

� 42 mins

Page 27: GPU-Accelerated Design Optimization on the Cloud

Framework

Fast Limited

Memory FEA

GPU

Acceleration

Pareto

Optimization

Page 28: GPU-Accelerated Design Optimization on the Cloud

Mesh Aware SpMV on GPU

Page 29: GPU-Accelerated Design Optimization on the Cloud

Deflation on GPU

TW dWµProlongation Restriction

Page 30: GPU-Accelerated Design Optimization on the Cloud

Example: Design Optimization

� OptiStruct (commercial)

� Xeon E5 2697,92 GB

� 20 hours!

� Pareto

� I7 4770,8 GB

� 42 mins

� Pareto

� GTX 480,1.5 GB

� 6 mins

Page 31: GPU-Accelerated Design Optimization on the Cloud

Cloud Based Design Optimization

Fast Limited

Memory FEA

Pareto

Optimization

GPU

Acceleration

WebGL

Page 32: GPU-Accelerated Design Optimization on the Cloud

32

WebGL & Three.js

WebGL

� JavaScript API for 3D graphics in browsers

� www.khronos.org

� Almost all browsers

ThreeJs

� Higher-level library

� www.threejs.org

� Almost all browsers

Page 33: GPU-Accelerated Design Optimization on the Cloud

33

Finally …

Page 34: GPU-Accelerated Design Optimization on the Cloud

34

A Pilot Service

www.cloudtopopt.com

� Entry level server

• E3-1270 V3

• 8 GB

� Limited to 150,000 degrees of freedom

� 300+ users

Page 35: GPU-Accelerated Design Optimization on the Cloud

www.cloudtopopt.com

35

Page 36: GPU-Accelerated Design Optimization on the Cloud

www.cloudtopopt.com

36

Page 37: GPU-Accelerated Design Optimization on the Cloud

37

Plans

� Port to HPC provider

� NSF funding

� Launch startup

www.cloudtopopt.com

Page 38: GPU-Accelerated Design Optimization on the Cloud

Acknowledgements

Praveen Yadav Shiguang Deng Amir M. Mirzendehdel Chaman Singh Alireza Taheri Bian Xiang

Anirudh Krishnakumar Anirban Niyogi Victor Cavalcanti Cameron Gilanshah Yibo Hu Alex Buehler

Funding� NSF

� Air-force

� Luvata

� Autodesk

� Sandia National Lab

[email protected] www.cloudtopopt.com