Download Cray Overview 09_08_2008

Cray Innovation
Barry Bolding, Ph.D.
Director of Product Marketing, Cray
[email protected]
September 2008
Cray Utilizes the Best Processor Technologies.
 Cray and AMD have worked together in HPC since 2002.
 Cray currently delivering petascale XT5 systems.
 Cray “Baker” systems driving to multi-petaflops in 2010-2011
 Cray advancing interconnect technologies.
 Cray and Intel formed a strategic partnership in HPC in 2008
 Intel-Cray partnership blends complementary skills
 Intel has world’s best IC process, and very strong IC design
capabilities
 Cray has brings excellence in system architecture, networking and
software expertise in HPC
July 17
HPCUser 2008
Slide 2
“Granite”
The Cray Roadmap
“Baker”+
Realizing Our Adaptive
Supercomputing Vision
“Marble”
“Baker”
Adaptive Systems
Processing Flexibility
Productivity Focus
Cray XT5 & XT5h
Cray XMT
Cray XT4
Cascade Program
Rainier Program
Hybrid Systems
Integrated Infrastructure
High Efficiency
Scalar
Vector
7/6/2017
July
17
Multithreaded
HPCUser 2008
Slide 3
The Cray Roadmap
Socket G34 &
Gemini
Realizing Our Adaptive
Supercomputing Vision
The first Baker delivery
* High Efficiency Cabinet
“Baker”+
“Marble”
“Baker”
Cascade Program
Adaptive Systems
Processing Flexibility
Productivity Focus
Cray XT5 & XT5h
Cray XMT
Cray XT4
“Granite”
Rainier Program
Hybrid Systems
Integrated Infrastructure
High Efficiency
Scalar
Multiple Systems
Shipped
Vector
7/6/2017
July
17
Multithreaded
HPCUser 2008
Slide 4
What does “Green” mean in HPC?
 Green in HPC means more computing in a fixed power
envelope.
 Green in HPC means higher density
 Customers want more computer in less space
 Green is in the Total Cost of Ownership (TCO)
 Customers want fewer dollars diverted away from computing
infrastructure.
 Green should be manageable
 No issues of condensation or leakage
 Easy Serviceability
 Green should be upgradeable
 Less landfill or recycling of components
 Green needs to be innovative
Cray High Efficiency Cabinet with ECOphlex Technology
(PHase change Liquid EXchange(PHLEX))
July 17
HPCUser 2008
Slide 5
ECOphlex
Technology in the
Cray High
Efficiency Cabinet
R134a piping
Exit Evaporators
Inlet Evaporator
July 17
HPCUser 2008
Slide 6
The Cray
High Efficiency Cabinet with
ECOphlex Technology
(PHase change Liquid EXchange(PHLEX))
Cray Momentum
Executing on $250M DARPA HPCS contract
Won majority of TI-08 program, NSF Track-2 system at UT
100TFLOPS350TF
20  50  119  263 TF  ~1 PF
40  124  284 TFLOPS
July 17
HPCUser 2008
Slide 8
Cray: Petascale and Beyond
High Productivity Computing Systems
Critical to National Security
• Develop a new generation of economically viable high productivity computing systems for national
security and industrial user communities (2011)
• Ensure U.S. lead, dominance, and control in this critical technology
Phase III Vendors:
Mission Partners:
Impact:
•
•
•
•
Performance (time-to-solution): speedup by 10X to 40X
Programmability (idea-to-first-solution): dramatically reduce cost and development time
Portability (transparency): insulate software from system
Robustness (reliability): continue operating in the presence of localized hardware failure,
contain the impact of software defects, and minimize likelihood of operator error
Applications:
Weather Prediction
Ocean/wave Forecasting
Ship Design
Climate
Modeling
Nuclear Stockpile
Stewardship
Weapons
Integration
Fill the Critical DoD Need for:
Operational weather and ocean forecasting, weapons design and analysis, airborne
contaminant
modeling,
cryptanalysis
Cray Proprietary Information
Providedintelligence/surveillance/reconnaissance,
Under NDA
July 17
Slide 10
Cray Cascade Project
Cray High Productivity Computing Systems
Scalable, High-Bandwidth Interconnect (Aries)
Granite
Granite
Granite
Granite
Granite
Granite
Marble
Marble
Marble
Marble
Custom
Compute
Node
Custom
Compute
Node
Custom
Compute
Node
Custom
Compute
Node
Custom
Compute
Node
Custom
Compute
Node
x86
Compute
Node
x86
Compute
Node
x86
Compute
Node
x86
Compute
Node
Marble
SIO
Node
Marble
SIO
Node
Local
Local
Local
Local
Local
Local
Local
Local
Local
Local
Local
Local
Memory Memory Memory Memory Memory Memory Memory Memory Memory Memory Memory Memory
Globally Addressable Memory
Support for PGAS and DGAS Data Distributions
Tightly integrated hybrid computing
Configurable network, memory, processing and I/O
Globally addressable memory
Very high performance communication and synchronization
July 17
HPCUser 2008
Slide 11
Thank You!