innovative solutions for energy exploration and data processing
TRANSCRIPT
Innovative Solutions for Energy Exploration and Data Processing, Nov 2013 REWRITING THE RULES WITH GPUS
Outline
Ultra-compact GPU configurations
Leveraging GPUs for energy exploration
Best practices for GPU deployments
Outline
Leveraging GPUs for energy exploration
Increasing need for energy exploration
0
50
100
150
200
250
300
350
400
450
500
0
12.5
25
37.5
50
62.5
75
87.5
100
112.5
125
1970 1975 1980 1985 1990 1995 2000 2005 2010
Spare Capacity
Oil P
roducti
on (
million b
opd)
World Capacity
Projected Growth in Supply and
Demand
World Demand
Increasing need for energy exploration
0
50
100
150
200
250
300
350
400
450
500
0
12.5
25
37.5
50
62.5
75
87.5
100
112.5
125
1970 1975 1980 1985 1990 1995 2000 2005 2010
E&
P E
xpendit
ure
s ($
billion) Spare
Capacity
Oil P
roducti
on (
million b
opd)
Oil P
rice $
/bbl
World Capacity
Projected Growth in Supply and
Demand
World Demand
Oil Price ($/bbl)
E&P Expenditure
Use simulations to reduce risk
Image courtesy of Acceleware
An unlikely hero…
Unreal Engine 4 ©Epic Games
Computer graphics require billions
of parallel computations
Millions of triangles Millions of pixels
Why are so many parallel operations needed?
Millions of triangles Millions of pixels
Why are so many parallel operations needed?
Input triangle Tessellate Projection Rasterize Shade Transform vertices
Image plane
Camera
E&P workflow
Seismic
Acquisition Seismic Imaging
Interpretation Reservoir
Characterization Petrophysics Well
Planning Drilling Reservoir Simulation Economics
Images courtesy of Schlumberger
Quadro™ Visualization
Seismic
Acquisition Seismic Imaging
Interpretation Reservoir
Characterization Petrophysics Well
Planning Drilling Reservoir Simulation Economics
Images courtesy of Schlumberger
Seismic
Acquisition Seismic Imaging
Interpretation Reservoir
Characterization Petrophysics Well
Planning Drilling Reservoir Simulation Economics
Images courtesy of Schlumberger
Tesla™ Computation
Seismic
Acquisition Seismic Imaging
Interpretation Reservoir
Characterization Petrophysics Well
Planning Drilling Reservoir Simulation Economics
Images courtesy of Schlumberger
NVIDIA Visual Computing
Transforming E&P workflow
E&P workflow
Seismic
Acquisition Seismic Imaging
Interpretation Reservoir
Characterization Petrophysics Well
Planning Drilling Reservoir Simulation Economics
Images courtesy of Schlumberger
Seismic Imaging
Compute requirements
Chart courtesy of TOTAL
Seismic Imaging
1995 2000 2005 2010 2012 2015 2020
100 GF 1 TF
10 TF
10 PF
100 TF
100 PF
1 PF
1 EF
1990
Paraxial WE
approximation
Kirchhoff Migration
Post SDM, PreSTM
Full WE Approximation
WEM
RTM
FWI
Elastic
Imaging
Why augment infrastructure with GPUs?
4x to 6x more throughput
4x Perf/W advantage
More revenue
Same cost basis
Sharper image
Higher throughput
Better drilling decisions
12% cost reduction
Why augment infrastructure with GPUs?
4x to 6x more throughput
4x Perf/W advantage
More revenue
Same cost basis
Sharper image
Higher throughput
Better drilling decisions
12% cost reduction
Solution to challenges: just add GPUs
Application Code
Cooperation
GPU CPU
Use GPU to Parallelize
Compute-Intensive Functions
Perform other tasks simultaneously
GPUs: Two year heartbeat
2012 2014 2008 2010
DP G
FLO
PS p
er
Watt
Kepler
Tesla
Fermi
Maxwell
Volta
FP64
CUDA
32
16
8
4
2
1
0.5
2016
Five years of NVIDIA GPUs in seismic
1992
NVIDIA Founded
... 2006
1st Generation
Tesla M870
CUDA Released
Five years of NVIDIA GPUs in seismic
1992
NVIDIA Founded
... 2006
1st Generation
Tesla M870
CUDA Released
2008
2nd Generation
Tesla M1060
Port RTM
App Ports
KDM
KTM
WEM
2007
Five years of NVIDIA GPUs in seismic
1992
NVIDIA Founded
... 2006
1st Generation
Tesla M870
CUDA Released
2010
3rd Generation
Tesla M2090
“Fermi”
Port FWI
2008
2nd Generation
Tesla M1060
Port RTM
App Ports
KDM
KTM
WEM
2007
Five years of NVIDIA GPUs in seismic
1992
NVIDIA Founded
... 2006
1st Generation
Tesla M870
CUDA Released
2010
3rd Generation
Tesla M2090
“Fermi”
Port FWI
2008
2nd Generation
Tesla M1060
Port RTM
4th Generation
Tesla K10
“Kepler”
Port Elastic
2012
App Ports
KDM
KTM
WEM
2007
Industry trends and challenges
GPU-based servers Consolidate workloads
Improve datacenter operations
Image courtesy of Schlumberger
Industry trends and challenges
GPU-based servers Consolidate workloads
Improve datacenter operations
Regional field studies Wide azimuth
Ocean bottom sensors Time lapse surveys
Images courtesy of Schlumberger, Shell
Industry trends and challenges
GPU-based servers Consolidate workloads
Improve datacenter operations
Collaborative workforce
Globally disperse Ever-changing environments
Regional field studies Wide azimuth
Ocean bottom sensors Time lapse surveys
Images courtesy of Schlumberger, Shell, and Paradigm
Remote collaboration architecture
Multiple users on virtualized
server platform
Server platform
VT-d capable chipset
NIC
Public or
Private
Cloud
Notebook PCs running
Client
GRID K2 GRID K2
Remote collaboration architecture
Multiple users on virtualized
server platform
Server platform
VT-d capable chipset
Virtual Machine
Quadro Driver
Win7
App
NIC
Public or
Private
Cloud
Notebook PCs running
Client
Remote Graphics
Multiple clients connected
to same remote session
GRID K2 GRID K2
Hypervisor
Remote collaboration architecture
Multiple users on virtualized
server platform
Server platform
VT-d capable chipset
Virtual Machine
Quadro Driver
Win7
App
Virtual Machine
Quadro Driver
Linux
App
NIC
Public or
Private
Cloud
Remote Graphics
Notebook PCs running
Client
Remote Graphics
Multiple clients connected
to same remote session
GRID K2 GRID K2
Hypervisor
Transformational hybrid architecture
Compute Intensive Apps
GPU Cluster Seismic Imaging
Electromagnetic Imaging Reservoir Simulation
Image courtesy of Schlumberger
Transformational hybrid architecture
Compute Intensive Apps
GPU Cluster Seismic Imaging
Electromagnetic Imaging Reservoir Simulation
Interactive Apps
GPU Cluster Adapt to client needs
Scalable rendering Fast access to data Automatic updates
Data
Storage
High-speed
network
Image courtesy of Schlumberger
Transformational hybrid architecture
Compute Intensive Apps
GPU Cluster Seismic Imaging
Electromagnetic Imaging Reservoir Simulation
Interactive Apps
GPU Cluster Adapt to client needs
Scalable rendering Fast access to data Automatic updates
Data
Storage
High-speed
network
Datacenter
Client Environments
Workstations Laptops
Quad-HD displays Immersive rooms
Tablets Smart phones
Image courtesy of Schlumberger
Depth-correct transparency rendering
Depth-correct transparency rendering
Depth-correct transparency rendering
Depth-correct transparency rendering
Depth-correct transparency rendering
Depth-correct transparency rendering
Energy Exploration challenges
Need to improve operational efficiency
Use simulations to reduce risk
GPUs deployed in datacenters
Better drilling decisions
Transforming E&P workflow
Leverage GPUs for collaboration
Need compact, reliable operation
Must monitor/manage GPUs
Section re-cap
Bill Carter, PresidentNovember 2013
R Associates Inc.
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement.
RAI GPU Solutions
Slide 2
• Kirchhoff Time Migration (KTM)
• Kirchhoff Depth Migration (KDM)
• Wave Equation Migration (WEM)
• Spectral Inversion
• Reverse Time Migration (RTM)
• Tesla GPUs have been deployed:
Schlumberger
Chevron
TOTAL
Petrobras
Repsol
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement.
GPU Drivers in Seismic Exploration
Slide 3
• Choose Correct Hardware
• Ensure Space, Power and Cooling
• Assemble and Physical Deployment
• Head Node Installation
• Compute Node Installation
• Management and Monitoring
• Run Benchmarks and Applications
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement.
Decisions to Drive GPU Clusters
5017GR-TF
GPU Servers
1017GR-TF
6037R-72RFT+
7047GR-TPRF
GPU BladeSBI-7127RG-E
43
6
4
2
4
1U UP – Value
2
1U/2U DP, Scalable, High Density 3U & Above – Powerful
1027GR-TQF1027GR-TQFT
2027GR-TRFH2027GR-TRFHT
1027GR-TRF1027GR-TSF1027GR-TRFT
2027GR-TR22027GR-TRT2
22
3
GPU FatTwinF627G3-FT+F627G3-F73+F627G3-F73PT+F627G2-FT+F627G2-F73+F627G2-F73PT+
2
1027GR-TRF+1027GR-TRFT+
3
1027GR-72R21027GR-72RT21027GR-TR21027GR-TRT2
n # of GPU per node
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement.
Slide 4
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement.
Motherboard: X9DRG-HF+II/-HTF+II
1U Chassis: 118GH-R1K65B
Processor SupportDual Intel® Ivy Bridge EP (Socket R) series CPU
Memory Capacity 16 DIMM, Max of 512GB Reg. ECC DDR3 up to 1600MHz
Expansion Slots 3 PCI-e x16 Gen 3 for double width GPU cards1 x8 (in x16 slot) LP card
I/O ports 1 VGA, 2 Gbit LAN, 2 USB 2.0, and 1 IPMI dedicated LAN port.
Drive Bays4 hot-swap 2.5” drives bays
System Cooling 10 counter rotating fans with optimal fan speed control1 air shroud
Power Supply1600W Platinum level efficiency redundant power supply
1
2
3
4
6
7
Key Features
Support up to 3 double width GPU cards (K10/K20M/K20X/K2/K1/Xeon Phi)
16 DIMM, up to 512GB memories SAS 2208 controller Platinum level 1600W power supply
1
23
4
6
7
5
5
DP GPU Server: SYS-1027GR-
Slide 5
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement. Slide 6
Motherboard: X9DRG-HF+II/-HTF+II
2U Chassis: 218GH-R1K65BProcessor SupportDual Intel® Ivy Bridge EP (Socket R) series CPU
Memory Capacity 16 DIMM, Max of 512GB Reg. ECC DDR3 up to 1600MHz
Expansion Slots 4 PCI-e x16 Gen 3 for double width GPU cards1 x4 (in x16 slot) and1 x8 (in x16 slot) LP card
I/O ports 1 VGA, 2 Gbit LAN, 2 USB 2.0, and 1 IPMI dedicated LAN port.
Drive Bays10 hot-swap 2.5” drives bays
System Cooling 5 counter rotating fans with optimal fan speed control1 air shroud
Power Supply1600W Platinum level efficiency redundant power supply
1
2
3
4
6
7
Key Features
Support up to 4 or 6 double width GPU cards (K10/K20M/K20X/K2/K1/Xeon Phi)
16 DIMM, up to 512GB memories Platinum level 1600W power supply Optional rear fans to support 300W GPU cards Optional two by 8 rear I/O add on card slots
1
23
4
6
7
5
5
Key Application
Computational Finance HPC/Oil and gas Weather and Climate Analysis
DP GPU Server: SYS-2027GR-TR2/TRT2
GPU Blade: SBI-7127RG-E (20 GPUs)
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement.
Slide 7
Intel® Ivy Bridge Qty:8 DDR3 DIMM Slots
Greater TCO
No Cables
System Mgmt.
System Maintenance
Enhanced 3000W redundant power supply
Highest GPU Blade Density20 GPU + 20 CPU per 7U
Integrated BMC – IPMI 2.0, KVM-over-IP, Remote Virtual MediaDual Port QDR/FDR IB
or 10GbE Support
2 Kepler
1x SSD or1x SATA-DOM
The RAI Advantage
© Copyright 2013 R Associates Incorporated (RAI). The information contained herein is subject to
change without notice, and is RAI CONFIDENTIAL. For use only under Confidentiality Agreement. Slide 8
• Experience
• Breadth of Partnerships
• Breadth of Solutions
• Dedicated Engineering
• Tested Segment-specific Solutions
• Small, Nimble, Flexible
• Innovation in our solutions
• Commitment to You!!
Thank you
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Bright Cluster
metrics
13
•
•
•
•
•
•
•
•
•
•
•
–
•
–
•
•
Booth #1725
Last week via GTC Express!