1
Meeting Manufacturing’s Need for
Production Supercomputing
Tony DeVarco
Director of Virtual Product Design
Manufacturing Solutions
2©2016 SGI
3©2016 SGI
Challenges Manufacturing Companies Are Facing
Image	Courtesy	of	ANSYS		
•  The need to increase engineering productivity.
•  Speed up product development time.
•  Reduce physical prototyping to save time and costs.
•  Inefficient use of expensive ISV licenses.
•  Need to maintain multiple geographically distributed engineering
facilities.
•  Distributed workstations cannot interact with computer resources and
data in the corporate data centers.
•  Need to replace workstations with low-cost, remote clients.
4©2016 SGI
Pre/Post
Model and Mesh
Creation and
Visualization
CEM
Computational
Electromagnetic
s
CFD
Computational
Fluid Dynamics
CSM*
Explicit
Workload Scheduling
SGI® Management Center and SGI® Performance Suite
SGI® Scale-up and Scale-out Computing Solutions
Workload Scheduling Tool
Linux OS
Nastran
ANSYS® Mechanical
ABAQUS/Standard
ADINA
Permas
LS-DYNA
PAM-CRASH
RADIOSS
ABAQUS
Madymo
ANSYS® FLUENT
OpenFOAM®
StarCCM+
PowerFLOW
CFD++
Pam-Flow
FEKO
ANSYS® HFSS
CST
FMSlib
ANSA
Hypermesh
Gambit
Patran
Altiar Hyperworks
ANSYS EKM
SimManager
ABAQUS/CAE
d3View
SDM
Simulation Data
Management
CSM*
Implicit
Nastran
ANSYS® Mechanical
ABAQUS/Standard
ADINA
Permas
Altair RADIOSS
LS-DYNA
PAM-CRASH
ABAQUS
Madymo
Altair AcuSolve
ANSYS® FLUENT
OpenFOAM®
StarCCM+
PowerFLOW
ANSYS® HFSS
Altair FEKO
CST
ANSYS Maxwell 3D
Altair Hypermesh
ABAQUS/CAE
ANSA
Gambit
Patran
Altiar Hyperworks
ANSYS EKM
SimManager
d3View
SGI® Vizserver®
SGI Hardware 3rd Party Software CAE Segments
* CSM Is Computational Structural Mechanics
SGI Software SGI Services
SGIServices
SGI® Solution Environment and
CAE Application Segments
5
SGI and ANSYS 

Scaling Fluent 17.2 to a New
Record of 145,152 cores
6©2016 SGI
SGI ICE XA System Used
SGI and ANSYS
Scaling Fluent 17.2 to a New Record of 145,152 Cores
•  A great example using a balanced production supercomputer is illustrated in a
recent setting of a new commercial CFD benchmark for the widely-used ANSYS
application.
•  Specifically, ANSYS and SGI application engineers worked together to achieve
a new world record, scaling ANSYS Fluent® on SGI® ICE™ XA, which is one of
the world’s fastest commercial distributed-memory supercomputer platforms.
7©2016 SGI
0	
20000	
40000	
60000	
80000	
100000	
120000	
140000	
160000	
0	 20000	 40000	 60000	 80000	 100000	 120000	 140000	 160000	
Speedup	
Cores	
Combustor	830M	--	Fluent	17.2	
ICE-XA	--	E5-2697v4	--	2.30	GHz	 Linear	Scaling	
Gas	Combustor	830M	cell	Model,	Courtesy	of	ANSYS	
In the test, SGI ran Fluent on 145,152 CPU cores, which is
over 16,000 more cores than the previous record. The benefit
of scaling to this level is that the total wall clock time to
complete a simulation can be significantly reduced. In this
case, a single simulation was completed in 13 seconds. In
contrast, the same simulation run on 1,296 cores took 20
minutes.
8
An industry example that
highlights the advantages of
scaling CFD workloads on a
SGI® UV™ System
9©2016 SGI
SGI® Origin 2000, 1996
MIPS R10K
SGI Origin 3000, 1999
MIPS R12K
SGI® Altix® 3000, 2003
Intel Itanium
SGI Altix 4700, 2006
Intel Itanium SGI UV 1000, 2010
Intel E7
SGI UV 300, 2014
Intel E7
SGI UV 2000/3000, 2012/2015
Intel E5
Seven Generations of Shared Memory Systems:
1996–2015
10
The Best of DMP and SMP in One Cache-Coherent System with One OS

SGI®
InfiniteStorage™
NVIDIA®
K5200 or M6000
Multiple Users and
Multiple Jobs
Preprocessing,
Mesh Generation
and Model
Decomposition
Running Solvers
Post Processing and
Visualization
UV rack layout for illustration purposes only.
Consolidate Workloads into
One Easy-to-Administer System
11©2016 SGI
Image courtesy London Computational Solutions
•  London Computational Solutions, headed by Mark Taylor,
former Head of CFD at McLaren F1, is working
with Elemental Cars, an advanced track car manufacturer.
•  The goal of this effort is to improve the cornering speed of
its RP1 car by using design elements to create an
aerodynamic downforce that increases the vertical force on
the car’s tires creating more grip with the road.
Elemental Cars
12©2016 SGI
•  “When I was Principal Aerodynamicist for McLaren F1 Racing, and now as
CEO of London Computational Solutions (LCS),” states Mark Taylor, “I
relied on the SGI® UV™ shared memory platform to deliver the
robustness, reliability, and efficiency to scale our CFD simulations to
meet a tight manufacturing deadline and our aerodynamic performance
targets.”
•  “When LCS was presented with the aerodynamic challenge to improve the
cornering speeds for a new British road-legal track car called Elemental
RP1, I knew I could meet their requirements
of a quick turnaround because the running
of our CFD software on SGI UV would
perform as advertised and just work.”
Image courtesy London Computational Solutions
Elemental Cars
13©2016 SGI
Examples of a fifth order accurate simulation of a full automotive car geometry.
Please	have	insideHPC	put	this		
Top	View	animaRon		here	
Please	have	insideHPC	put	this	
Bo*om	View	animaRon		here	
Elemental Cars
14©2016 SGI

SGI: Meeting Manufacturing's Need for Production Supercomputing

  • 1.
    1 Meeting Manufacturing’s Needfor Production Supercomputing Tony DeVarco Director of Virtual Product Design Manufacturing Solutions
  • 2.
  • 3.
    3©2016 SGI Challenges ManufacturingCompanies Are Facing Image Courtesy of ANSYS •  The need to increase engineering productivity. •  Speed up product development time. •  Reduce physical prototyping to save time and costs. •  Inefficient use of expensive ISV licenses. •  Need to maintain multiple geographically distributed engineering facilities. •  Distributed workstations cannot interact with computer resources and data in the corporate data centers. •  Need to replace workstations with low-cost, remote clients.
  • 4.
    4©2016 SGI Pre/Post Model andMesh Creation and Visualization CEM Computational Electromagnetic s CFD Computational Fluid Dynamics CSM* Explicit Workload Scheduling SGI® Management Center and SGI® Performance Suite SGI® Scale-up and Scale-out Computing Solutions Workload Scheduling Tool Linux OS Nastran ANSYS® Mechanical ABAQUS/Standard ADINA Permas LS-DYNA PAM-CRASH RADIOSS ABAQUS Madymo ANSYS® FLUENT OpenFOAM® StarCCM+ PowerFLOW CFD++ Pam-Flow FEKO ANSYS® HFSS CST FMSlib ANSA Hypermesh Gambit Patran Altiar Hyperworks ANSYS EKM SimManager ABAQUS/CAE d3View SDM Simulation Data Management CSM* Implicit Nastran ANSYS® Mechanical ABAQUS/Standard ADINA Permas Altair RADIOSS LS-DYNA PAM-CRASH ABAQUS Madymo Altair AcuSolve ANSYS® FLUENT OpenFOAM® StarCCM+ PowerFLOW ANSYS® HFSS Altair FEKO CST ANSYS Maxwell 3D Altair Hypermesh ABAQUS/CAE ANSA Gambit Patran Altiar Hyperworks ANSYS EKM SimManager d3View SGI® Vizserver® SGI Hardware 3rd Party Software CAE Segments * CSM Is Computational Structural Mechanics SGI Software SGI Services SGIServices SGI® Solution Environment and CAE Application Segments
  • 5.
    5 SGI and ANSYS
 Scaling Fluent 17.2 to a New Record of 145,152 cores
  • 6.
    6©2016 SGI SGI ICEXA System Used SGI and ANSYS Scaling Fluent 17.2 to a New Record of 145,152 Cores •  A great example using a balanced production supercomputer is illustrated in a recent setting of a new commercial CFD benchmark for the widely-used ANSYS application. •  Specifically, ANSYS and SGI application engineers worked together to achieve a new world record, scaling ANSYS Fluent® on SGI® ICE™ XA, which is one of the world’s fastest commercial distributed-memory supercomputer platforms.
  • 7.
    7©2016 SGI 0 20000 40000 60000 80000 100000 120000 140000 160000 0 20000 40000 60000 80000 100000 120000 140000 160000 Speedup Cores Combustor 830M -- Fluent 17.2 ICE-XA -- E5-2697v4 -- 2.30 GHz Linear Scaling Gas Combustor 830M cell Model, Courtesy of ANSYS In the test, SGI ran Fluent on 145,152 CPU cores, which is over 16,000 more cores than the previous record. The benefit of scaling to this level is that the total wall clock time to complete a simulation can be significantly reduced. In this case, a single simulation was completed in 13 seconds. In contrast, the same simulation run on 1,296 cores took 20 minutes.
  • 8.
    8 An industry examplethat highlights the advantages of scaling CFD workloads on a SGI® UV™ System
  • 9.
    9©2016 SGI SGI® Origin2000, 1996 MIPS R10K SGI Origin 3000, 1999 MIPS R12K SGI® Altix® 3000, 2003 Intel Itanium SGI Altix 4700, 2006 Intel Itanium SGI UV 1000, 2010 Intel E7 SGI UV 300, 2014 Intel E7 SGI UV 2000/3000, 2012/2015 Intel E5 Seven Generations of Shared Memory Systems: 1996–2015
  • 10.
    10 The Best ofDMP and SMP in One Cache-Coherent System with One OS SGI® InfiniteStorage™ NVIDIA® K5200 or M6000 Multiple Users and Multiple Jobs Preprocessing, Mesh Generation and Model Decomposition Running Solvers Post Processing and Visualization UV rack layout for illustration purposes only. Consolidate Workloads into One Easy-to-Administer System
  • 11.
    11©2016 SGI Image courtesyLondon Computational Solutions •  London Computational Solutions, headed by Mark Taylor, former Head of CFD at McLaren F1, is working with Elemental Cars, an advanced track car manufacturer. •  The goal of this effort is to improve the cornering speed of its RP1 car by using design elements to create an aerodynamic downforce that increases the vertical force on the car’s tires creating more grip with the road. Elemental Cars
  • 12.
    12©2016 SGI •  “WhenI was Principal Aerodynamicist for McLaren F1 Racing, and now as CEO of London Computational Solutions (LCS),” states Mark Taylor, “I relied on the SGI® UV™ shared memory platform to deliver the robustness, reliability, and efficiency to scale our CFD simulations to meet a tight manufacturing deadline and our aerodynamic performance targets.” •  “When LCS was presented with the aerodynamic challenge to improve the cornering speeds for a new British road-legal track car called Elemental RP1, I knew I could meet their requirements of a quick turnaround because the running of our CFD software on SGI UV would perform as advertised and just work.” Image courtesy London Computational Solutions Elemental Cars
  • 13.
    13©2016 SGI Examples ofa fifth order accurate simulation of a full automotive car geometry. Please have insideHPC put this Top View animaRon here Please have insideHPC put this Bo*om View animaRon here Elemental Cars
  • 14.