Innovation without Limits SGI High Performance Computing

1
Innovation without LimitsSGI High Performance
Computing
Silicon Graphics, Inc.
Nyuli Gábor Silicon Computers Kft.
2
Silicon Graphics
Industry Leading Compute, Storage and
Visualization Solutions

• Focus on the technical computing market
• Over 20 years of experience in high performance
  computing innovation
• Technology designed to enable the most
  significant scientific and creative breakthroughs
  of the 21st century
• Products and services are mission critical to
  government and defense, science and research,
  manufacturing, energy and media industries

Images courtesy of Dr. Arthur W. Toga, The
Laboratory of Neuro Imaging American Museum of
Natural History Industrial Light and Magic,
Lucasfilm Ltd Image courtesy of Parametric
Technology Corporation.  
3
Focus Is on Engineers, Scientists and Creative
Professionals
Defense Security
Science
Media
Manufacturing
Energy
Images courtesy of EPL Productions, Leonard
Wikberg III, EnSight image, Landmark, and Georgia
Public Broadcasting
4
SGI Core Technology Offerings

• Linux OS and Intel Itanium 2 Processor-based
  platform
• Scales up and out
• Programming models and libraries
• System partitioning and resource management

High-PerformanceComputing

• I/O performance
• File systems
• Data management
• Networking
• Distributed data access

AdvancedVisualization
Storage

• Unmatched visualization
• Leverages industry and open
• standard components
• Single-user and collaborative
• Large-model visualization
• Real-time and distributed visualization

Delivering technology that enables significant
scientific and creative breakthroughs
5
SGI in High Performance Computing
and BEYOND
2000s
1990s
1980s

• The industry's most scalable architecture from
  2 to 10,000 processors
• Shared memory scales up to 24 terabytes
• Based on Intel processors and the Linux
  operating system
• Flexible, modular design that independently
  scales CPU, memory, I/O
• Complete solution stack for High Performance
  Computing

6
SGI Altix Market Momentum
HPCWire Awards For 2004 Best Price/performance Mos
t Innovative Hardware Technology Most Innovative
Visualization Technology Most Innovative
Storage Best Govt/industry collaboration Most
Innovative Implementation
Dresden
62,000 Processors Sold!
LRZ
IFS
CSCS
IMS
Yukawa
Processors Sold
3,300 Processors Sold!
GFDL APAC
JAERI
DoDMod
NASA
NCSA
800Processors Sold!
Ford
Total
GFDL
SARA
SGI Altix 3000 BX2 Announced
SGI Altix 330, RASC Announced
SGI Altix 3000 Server Announced 1/03
SGI Altix 350 Server Announced
FY03
FY04
FY05
FY06
7
SGI Altix Server High-Performance Architecture

• NUMAflex global shared memory and ultra-high
  bandwidth interconnect
• Modular, expandable architectureprocessors, I/O,
  memory
• SGI ProPack software feature for Linux OS
  optimizations for HPC

SGI Altix Family

• All processors operate on one large
  shared-memory space
• Industrys highest bandwidth interconnect at
  6.4GB/second
• High performance, low cost, easy to deploy

8
SGI Altix Servers
Application Complexity

• Mix of applications
• Unpredictable workloads
• Growing job size

SGI Altix 3700 Bx2 NEW SGI Altix 4000
Supercomputers
100s
SGI Altix 350 Departmental servers (432 CPUs)
Scale up (processors per node)
SGI Altix 330 Low-cost workgroup (116 CPUs)
SGI Altix Clusters Large node clusters
10s
2s
Scale out (Total number of processors)
10s
1,000s

• Number of users
• Number of jobs

Single-job, single-user
Processing Capacity
9
SGI Altix High-end Servers

• Small-footprint packaging for high density,
  reduced costs
• Up to 512 processors under one OS, single system
  image
• Up to 2,048 total processors per NUMAlink
  system
• Best-in-class NUMAlink 4 routers with new
  topologies
• Water-cooled door option for larger
  configurations
• NEW SGI Altix 4000 extends scaling up to 8192
  core NUMAlinkd systems and 128TB of global
  shared memory in a functional blade design

10
SGI Altix 3700 Bx2 Building Blocks
Linux Operating Environment
11
Altix Customers with Memory gt500GB

• 98 systems with gt500GB memory
• 39 with gt 1TB
• Benefits of Large Shared Memory
• Massive In-core Computation, Memory-Mapped I/O
  Lead to Breakthrough Results
• Unmatched Ease of Scaling from Efficient MPI,
  Trivial Load Balancing and No Data Replication
• Up to 512P, 6TB under one OS Means Ease of
  Development and Administration
• Only publicly referenced customers are listed.

12
SGI Altix 4000 PlatformExtending Performance
Leadership

• Superior Performance Density up to nearly 3X
  density improvement over Altix 3700 Bx2
• Plug and solve Versatility 8 standard
  functional blade choices, high-speed NUMAlink
  backbone supports top performance for any HPC
  application
• Best Performance on Cluster Applications SGI
  NUMAlink interconnect beats the competition with
  sub-microsecond latency

13
SGI Altix 4000 PlatformStep into
Multi-Paradigm Computing

• Multi-Paradigm Computing Taking HPC to the next
  level with integrated computational resources
  that can be seamlessly accessed as application
  workload requires.
• SGI Altix 4000 Resources Tightly Integrated with
  Peer I/O Direct, high-speed connection of
  integrated graphics, RASC/FPGA and other future
  processing elements to global memory address
  space
• Takes HPC to the Next Level beyond the limits
  of Moores Law and parallelism

14
One Giant Leap for NASA
Deployment 20 systems, each gt512p, 1 TB memory,
1 Linux kernel gt440 Tbytes CXFX SAN
storage gtSAN with gt1 petabyte managed by DMF
The Challenge gt Deliver a world-class
supercomputing platform to NASA in support of
their world-class scientific and engineering
problems gt Provide the environment needed for
NASA to perform simulations to support Space
Shuttle Return to Flight The Solution gtDeploy
a 10,240 processor supercomputer in a record 15
weeks. Ranks as 3rd most powerful computer in
world.
15
SGI Altix 350 Powerhouse Midrange Server

• Provides superior performance and scalability
  from 1 to 32 processors
• Best price/performance midrange server
• Ideal for departmental application servers,
  technical database, throughput clusters
• One Linux instance to manage
• Cluster to thousands of processors using
  industry-standard interconnects
• Independently scale CPU, memory, I/O
• Investment protection and leverage current assets
• Allocate budget and resources to ongoing needs
• Right-size systems for the ultimate
  price/performance

16
SGI Altix 330 Entry Midrange Price/Performance
Leader
SGI Altix Family
Combined With
NUMAflexArchitecture
Linux OS
Intel Itanium 2Processor
Ultra-high Density Package
HPC Heritage
Great Price Point
SGI HPC Expertise
Ultra-dense, Ultra-affordable, Ultra-powerful
Advanced Developer Tools
Scalable Architecture
17
SGI Altix Midrange Powerful, Affordable,
Flexible

• Advanced scalability
• Altix 350 scales to 32 processors, 384GB of
  memory
• Altix 330 scales to 16 processors, 128GB of
  memory
• Expand-on-demand modularity for perfect
  right-sizing cost effectiveness
• Best in class price/performance
• Altix 330 provides affordable entry point,
  development platform
• Based on award-winning SGI NUMAflex
  architecture for superior performance and
  scalability from 1 to

18
Accelerating Research at the University of
Minnesota Supercomputing Institute

• Deployment
• Added a 256P Altix 3700 system and 18P SGI Altix
  3700 system featuring Intel Itanium 2
  Processors with 512GB of memory to existing SGI
  Altix infrastructure
• Added a 128P (8 x 16P) SGI Altix 350 cluster with
  Voltaire InfiniBand interconnect
• Integrated new cluster with SGI InfiniteStorage
  SAN Solution with CXFS Shared Filesystem
  software and 3TB fibre channel RAID

• The Challenge
• gtProvide computational power for researchers in a
  broad range of disciplines, including
  computational genetics, biology, and atmospheric
  sciences
• Cost efficiently acquire a supercomputing class,
  shared memory computing solution
• The Solution
• gt The 256P SGI Altix system is helping to break
  new ground in a broad range of disciplines
• The University obtained a 128P cluster with lower
  interconnect and fabric management costs than
  with other 2P or 4P node systems

Image Courtesy of Dr. Shuxia Zhang at the
Minnesota Supercomputer Institute.
Shared memory architecture is widely
acknowledged as ideal for large-scale scientific
computing tasksIt undoubtedly will lead to
significant improvements in our already high
productivity. DONALD TRUHLAR, SUPERCOMPUTING
INSTITUTE DIRECTOR AND PROFESSOR OF CHEMISTRY
19
SGI RASC Technology

• Reconfigurable Application Specific Computing
• Extends SGIs leadership in high performance
  computing
• Delivers unmatched performance, scalability and
  bandwidth for data-intensive applications
• Over 100X Improvements in mission-critical
  application acceleration
• Solution stack comprised of industry-leading
  tools and libraries

20
RASC Technology Demonstrated Application
Speed-up
1 Based on internal testing
21
SGI Altix and DatabasesA Wealth of Options
22
Altix-ok itthon

• Országos Meteorológiai Szolgálat (Altix 3700, 144
  CPU)
• Szegedi Tudományegyetem (Altix 3700, 48 CPU)
• ELTE Szerves Kémia Tsz. (Altix 350, 16 CPU)
• Collegium Budapest (Altix 350, 16 CPU)
• MTA Atomenergia Kutatóintézet (Altix 350, 12 CPU,
  96GB RAM)
• MTA SZBK Enzimológia Intézet (Altix 350, 12 CPU)
• Semmelweis Egyetem Biofizika Intézet (Altix 350,
  6 CPU)
• Duna Televízió (Altix 3300)
• Nemzeti AudioVizuális Archívum (Altix 350)
• Silihost (Altix 350)

23
Summary

• SGI proven leader in High Performance Computing,
  with industry-leading technology and expertise
• SGI Altix platform based on SGI NUMAflex
  architecture, with ability to scale from 1 to
  10,000 processors and terabytes of memory
• Unified Altix family of servers, supercomputers,
  and clusters provides the ability to "start
  small" and grow to meet any processing
  requirements
• Investment protection, with an open
  standards-based system that supports all major
  HPC applications

24
Thank You
Thank You