Get documents about "The Complete IDC Intelligence Solution
Document Sample
IDC Technical Market
Overview
www.idc.com/hpc
IDC is Focused on IT
38 years of experience in IT market research
Founded in 1964
700+ analysts worldwide
Quality primary research and methodologies
More than 370,000 surveys/year
Only global/local IT research company
50 offices in 42 countries
IDC Research Areas
Enterprise
Systems Technical
Global HPC
Research
Workstations
Services
Software
Internet
Channels
Verticals
Personal
Systems
Networking and Consumer and Small
Telecommunications Business
Components and
Peripherals
Overview: The New Realities
Technical markets join the mainstream
Scalable systems and leverage product strategies
Hardware alone is becoming less important
Capability market transition
Shake up in vendors and technologies
Bio-Sciences markets accelerate
Major opportunities developing for IT suppliers
Clusters – Not Just for Breakfast Anymore
Real interest in clusters appearing at all levels
GRIDS are becoming mainstream for many sites
2002 vs. 2001 Highlights
Overall -- Revenue down 7.2% to $4.7B
Capability -- Revenue up 24.3%, to $1,003M
Enterprise -- Revenue down 25.8%, to $785M
Divisional -- Revenue down 27.0%, to $1.062B
Departmental -- Revenue down 5.4%
Still largest market segment at $1.850B
High-end HPC -- Revenue down 4.1% to $1.8B
High-end HPC = Capability + Enterprise
Rest of market down – 9.0% to $2.9B
IDC’s Market Segmentation
Technical Capability
Systems configured and purchased to solve the largest most
demanding problems
Technical Enterprise
Systems purchased to support technical applications in
throughput environments selling for $1 million or more
Technical Divisional
Systems purchased for throughput environments selling from
$250,000 to $999,000
Technical Departmental
Systems purchased for throughput environments selling for
less than $250,000
WW Technical Computing Market
$9,000
Capability
$8,000 Enterprise
Divisional
$7,000 Departmental
$6,000
$5,000
$4,000
$3,000
$2,000
$1,000
$0
1997 1998 1999 2000 2001 2002
Overall -- Revenue down 7.2% to $4.7B (from $5.06B)
2002 Industry/Application Segments:
Percentage By Revenues
30%
25%
20%
15%
10%
5%
0%
E
M
.
D
D
A
m
er
eo
se
e
e
ci
A
nc
ED
&
ag
CA
he
&
th
-S
en
G
C
SR
na
O
an
C
io
C
D/
ef
C
B
Fi
M
A
D
D
C
ch
Te
IDC Application Segment Definitions
1. Biological Sciences -- applications include genomics, proteomics, pharmaco-genomics,
pharmaceutical research, bioinformatics, drug discovery, bio-analytic portals and ASP
type service providers, agricultural research. Computational techniques range from data
base searching and management to molecular modeling, to computational chemistry.
These applications are used in commercial, academic and institutional research.
System that are specifically targeted for these applications should be included, systems
purchased for general scientific and R&D environments should be counted in the
Science and R&D segment.
2. Chemical engineering applications include molecular modeling, computational
chemistry, and chemical analysis. All chemistry applications that are not directly related
to bio-sciences research and development. These applications are used in commercial,
academic and institutional research.
3. Classified and Defense applications include surveillance and signal processing,
encryption, C3I (command, control, communications and intelligence), defense
research, and other national security applications.
4. Digital Content Creation (DCC) and Distribution applications include 2D and 3D
animation, film and video editing and production, and multimedia authoring for both CD
and web pages that utilize sophisticated graphics content. This category also includes
servers used for image rendering, content management, and distribution of finished
products. The increasing use of digital content in areas such as film, TV, commercial
animation, advertising, product styling, and industrial design has expanded the market
to include both low end systems and extremely high performance computers.
5. Economic and financial modeling includes both trader and non-trader tasks, such as
econometric modeling, portfolio management, stock market and economic forecasting,
and financial analysis.
IDC Application Segment Definitions
6. Electronic design and engineering analysis covers all electrical/electronic tasks,
including schematic capture, logic synthesis, circuit simulation, and PCB routing.
7. Geoscience and geo-engineering applications include earth resources-related work
such as seismic analysis, oil services, atmospheric modeling, geo signal analysis, and
various applications in meteorology industries. These applications are used both in
institutional research and commercial enterprises. Geoscience and geo-engineering is
also referred to as GIS, and includes components of mapping and image processing.
8. Imaging applications handle input from some type of camera, typically a video camera,
and require specialized image preprocessors to prepare the image data for
manipulation by a computer system. The medical and earth science industries have
traditionally been the most common users of image technology, for such applications as
body scanning, medical imaging, terrain modeling, and remote-sensed (LANDSAT or
satellite-derived) data analysis.
9. Measurement and Control applications include measurement analysis, and the front
end of automated test equipment (ATE). Like the simulation portion of the market,
measurement and control typically requires real-time response.
10.Mechanical Design and Drafting includes design applications such as mechanical CAD;
2D, 2-1/2D, and 3D design and drafting; 3D wireframe; and civil engineering design.
Design and drafting applications require graphics capability, but are less compute-
intensive than design engineering and analysis applications. CAD tasks are typically
done by designers and drafters. Users are found primarily in discrete manufacturing
industries such as automotive, aerospace, heavy machinery, and consumer goods.
IDC Application Segment Definitions
11.Mechanical Design Engineering and Analysis are those tasks generally accomplished
by engineers - not drafters. This segment of technical computing covers such analytical
tasks as finite element modeling and analysis, mechanical CAE, civil engineering,
structural analysis, computation fluid dynamics, and solid modeling. Like CAD
applications, these CAE tasks are used to design automobiles, commercial aircraft,
running shoes, ski equipment, sail boards, beer bottles, and other everyday items.
12.Other commercial applications are those that have not otherwise been classified; such
as data analysis, office automation, WP, accounting, payroll, inventory, and customer
service. This category is used to account for additional sales of workstations, and for
special sales of servers from companies with a strong primary focus on technical
markets. It is not intended to include commercial server sales by companies selling into
a broad range of horizontal markets.
13.Scientific research and R&D applications are typically compute-intensive, and often
require high-performance graphics. Programs can be exceedingly large and complex.
These users are less bound by strict economic constraints than those performing
applications in products environments. Therefore scientists and researchers are among
the most CPU-hungry users in the technical computing market. Users are found in
universities, national laboratories (such as Argonne and Lawrence Livermore), pure
research institutions (such as MCC and MCNC), and for-profit corporations. They
conduct basic and applied research in areas such as physics (including Nuclear),
computational fluid dynamics, scientific visualization, thermodynamics, wind tunnel
testing, laboratory data analysis, and scientific signal analysis.
IDC Application Segment Definitions
14.Simulation applications include operator-in-the-loop simulators in such areas as:
factory operations analysis and operator training, flight training, nautical training,
nuclear plant analysis and operator training, robotic simulation, utility or power
distribution analysis, and related training (computer-aided instruction (CAI)).
15.Software engineering users generally require less compute power than other technical
users. These applications are typically performed by low-end and midrange systems,
particularly low-cost technical workstations, and in some cases even PCs. Software
engineering includes those tasks such as CASE, expert/knowledge systems, algorithmic
research. Perhaps the most critical requirements for software developers is windowing
capability. The need for high-end graphics capability is relatively low, since two
dimensions are generally sufficient for the conceptualization and visualization of a logic
system. However, the compilation and test of programs, can benefit from access to
powerful networked compute resources to speed the overall development process.
16.Technical Management and support applications involve tracking, documenting and
controlling the product life cycle chain, and the scientific research process. Such
management and support is typically done by a file or data server on a network. Tasks
include: product data management, maintenance records management and analysis,
revision control, configuration management, network management, and project
management.
17.Technical Other applications not otherwise specified by the above definitions.
2002 Revenue Share By Technologies
2002 By Processors
Sparc PA-RISC
20% 18%
Alpha
14%
2002 By Memory Architecture
Power
23%
Intel
MIPS Custom 13% SM-xbar
SMP-Bus 15%
6% 6% 72%
SMP-Bus
MP SM-xbar
8% MP
NUMA NUMA
Other
2% Other
3%
Technical Market Forecasts
Our current forecast model demonstrates a
reset pattern returning to year over year
growth beginning in 2003
Revenue Growth – 6.1% CAGR to 2007
– Growth rates relative to the peak year of
2000 are about 2.0% CAGR
– Over $6.3 billion by 2007
– Bio & Life Sciences over $2.6 Billion by 2007
Unit Growth -- 11.8% CAGR
– 108,450 shipments in 2007
Key Forecast Assumptions
Economic recovery
– HPC economies return to a growth mode
– No major disruptions from terrorist events
Biosciences major market driver
– Bio-sciences revenue growth – 16.6%
– Traditional HPC revenue growth – 1.4%
HPC resiliency
– Research and Development necessary for product cycles
– Government spending is long terms expected to increase
– Longer sales cycles
Market Dampers
– Price/Performance reset
– Clusters provide a new sources of price competition
– Grid model increase overall efficiencies
A Few Next Generation Applications
Automotive
Full fidelity crash
Integration of design, engineering, manufacturing and test
Full durability -- 150,000 miles, no failures
Pharmaceuticals
Properties modeling
Cell membrane modeling
Health Sciences -- Genome and:
Physiological modeling
Vascular simulation
Virtual surgical planning
Chemicals
Developing new biological compounds
Process modeling
Petroleum
Seismic and reservoir visualization
Improved substrate modeling
Forward analysis of oil-bearing structures
New Application Development Areas
New Application Areas in 2003:
48% Will Invest In Current Applications
14.0%
11.5%
12.0%
10.0%
8.0%
5.8% 5.8%
6.0%
3.8%
4.0%
1.9% 1.9% 1.9% 1.9%
2.0%
0.0%
D
cs
s
r
m
D
e
e
he
ic
as
CF
siv
&
ati
e
om
t
Ch
aB
R
ea
m
en
n
/W
ps
t
or
l
nt
Da
Ge
na
Ap
nf
aI
tio
ate
oI
t
ity
Da
ta
Bi
im
pu
un
Cl
m
m
Co
m
Co
Worldwide Technical Market Forecast:
Revenues ($,000)
10,000,000
Capability
9,000,000 Enterprise
Divisional
8,000,000 Departmental
7,000,000
6,000,000
5,000,000
4,000,000
3,000,000
2,000,000
1,000,000
0
2001 2002 2003 2004 2005 2006 2007
The overall market will grow at a CAGR of 6.1%
to reach $6.3 Billion by 2007
Worldwide Technical Market Forecast:
System Shipments
120,000
Capability
100,000 Enterprise
Divisional
80,000 Departmental
60,000
40,000
20,000
0
2001 2002 2003 2004 2005 2006 2007
Technical Cluster and Grid
Market Overview
www.idc.com/hpc
Definitions: Clusters and Grids
A set of independent computers combined into a
unified system through systems software, and
networking technologies
Independent components
Standard interconnects
Clusters vs. Grids: Different ends of same spectrum
Clusters:
– Dedicate components -- All components are exclusively
“owned” and managed as part of the cluster
Grid:
– Virtual system -- Configured from components that are
generally managed and used both as part of the grid and
as independent systems
IDC Cluster Market Model Definitions
BRIGHT CLUSTERS
Complete configurations offered by system’s suppliers
The vendor is responsible for assembly, delivery and
support of the cluster product
DIM CLUSTERS
End-user assembly of individual systems sold as
technical servers or workstations
DARK CLUSTERS
End-user assembly of components purchased from a
number of sources
INVISIBLE CLUSTERS
Re-purposed clusters build from systems that were
purchased for other reasons
These systems do not add measurable market revenue
Cluster Market Model Topology: View #1
Technical
Cluster Servers
Bright Dim Dark Invisible
Clusters Clusters Clusters Clusters
Traditional End New End End Grid
Vendors Users VARs Users Users Intersect
View #2: An Alternative View Of Grids
< - - - - - - - Grids - - - - - - - - >
Other Devices:
-- PCs
-- Workstations
-- Commercial Servers
HPC -- Handhelds
Server Market -- Instruments
Traditional: Clusters: Other Servers:
Servers -- Bright -- Proprietary
-- SMPs -- Dim -- FPGA
-- Vectors -- Dark -- Other Custom
-- Invisible
2002 HPC Cluster Market Revenues: $1.6B
$700
$600
$500
$ Millions
$400
$300
$200
$100
$0
Bright Dim Dark
Market Drivers for Clusters
Absolute node level performance
Moore’s Law at work
Price/Performance as throughput engines
Single problem per node computing
Maximizes management flexibility
Best solution for highly parallel applications
Embarrassingly parallel problems
Important enough to justify special system
Technology exploration phase
Highly visible technology requires investigation
Broader market is experimenting with clusters
Potential for dramatic changes in market structure
Linux
Breaks O/S binding to vendor
Provides for greater sharing of software
HPC Cluster Market Forecast by Source
Segment: Revenues ($ Millions)
$3,000,000
Dark
$2,500,000 Dim
Bright
$2,000,000
$1,500,000
$1,000,000
$500,000
$0
2001 2002 2003 2004 2005 2006
HPC Cluster Market Forecast by Source
Segment: System Shipments
35,000
30,000 Dark
Dim
25,000 Bright
20,000
15,000
10,000
5,000
0
2001 2002 2003 2004 2005 2006
Operating System Trends
And Observations
www.idc.com/hpc
2002 HPC Server Market Revenues: $4.7B
$4,000
$3,500
$3,000
$ Millions
$2,500
$2,000
$1,500
$1,000
$500
$0
UNIX Linux NT Other
WW Technical Computing Market by OS
$9,000
$8,000 UNIX Linux
$7,000 NT Other
$6,000
$5,000
$4,000
$3,000
$2,000
$1,000
$0
1997 1998 1999 2000 2001 2002
Overall -- Revenue down 7.2% to $4.7B (from $5.06B)
2002 HPC Cluster Market Revenues: $1.7B
$1,000.00
$900.00
$800.00
$700.00
$ Millions
$600.00
$500.00
$400.00
$300.00
$200.00
$100.00
$0.00
Linux UNIX NT Other
WW Technical Cluster Market by OS
$2,000
$1,800 Linux Unix
$1,600 NT Other
$1,400
$1,200
$1,000
$800
$600
$400
$200
$0
1997 1998 1999 2000 2001 2002
In Conclusion
www.idc.com/hpc
Major Market Trends
• Commodity processors & systems now dominate
Custom processors & systems aren’t material in
market size, but provide unique capabilities
Hardware alone is becoming less important
• Capability market transition
Shake up in vendors and technologies
• Clusters and Linux now part of the mainstream
• GRIDs and IA-64 are emerging technologies
• Bio-Sciences and cluster markets accelerate
• Long term growth is projected in all segments
Short term economic situation is less clear
Questions?
Please email:
hpc@idc.com
Or check out:
www.idc.com/hpc
Goals of the hpc@idc User Forum
Assist HPC users in solving their ongoing computing
and business problems
A forum for exchanging information, identifying
areas of common interest, and developing unified
positions on requirements
Provide members with a continual supply of information
on:
Uses of high end computers, high end best
practices, market dynamics, computer systems and
tools, vendor activities and strategies
Provide members with a channel to present their
achievements and requirements to outside interested
parties
The Technical Agenda:
Initial HPC User Issue Areas
Better Metrics: “Sizing” Computers
Sustained vs. Peak Performance
Develop a position paper on HPC architecture
requirements
Learn What Other Users Are Doing
Collaboration Among Users
Leverage Technology
Leverage HPC Vendors
And Academia
And ISVs
Map Applications to Architectures
The Economics Of Supercomputing
Previous HPC User Forum Meetings
• October 1999: First planning meeting
• November 1999: User Forum public launch
• April 17, 2000: Second planning meeting
• July 24, 2000: Third planning meeting
• September 18 & 19, 2000: First full User Forum
meeting, Richmond, Virginia
• April 11 & 12, 2001 Dearborn, Michigan
• April 23 & 24, 2002 in Santa Fe, New Mexico
• September 10 & 11, 2002 in Portland, Maine
• April 8 & 9, 2003 in Sundance, Utah
• May, 2003 in Bristol, UK and Annecy, France
• September 16 & 17, 2003 in Princeton, NJ
• October, 2003 in London and Paris
HPC User Forum Update
110 attended the last meeting in Princeton
96 was the previous high point (Sundance)
Membership is now over 145
Check out the web site at:
www.idc.com/hpc
Web Site
IDC Balanced HPC Performance
Rankings
www.idc.com/hpc
Why IDC Started This Activity
HPC users and vendors listed it as their MOST pressing
issue
Impacting the health of the industry
– And the quality of the products available
Vendors felt that it was driving a number of major
purchases in the US
– “So why not just design cheap computers with very high
processor peak numbers?”
Users expressed that it impacted their ability to acquire
the best computer
– Without major questioning by upper management
Users and Vendors asked IDC, because we are
independent and have a history in HPC
Philosophical Notes – Our Approach
Using more metrics can provide a more balanced
picture
A good single metric doesn’t currently exist
– One with data on many configurations
There already exists many useful metrics
… the problem can be solved
An adaptive and evolutionary approach
Add better metrics as they become available
Adding price, I/O and ease-of-use
Base Data Set
• Vendor Name • System SPECfp _Rate _base2000
• Computer Model • System SPECint _Rate _base2000
• Number of CPUs • Single System Node Bandwidth
• Total/ Max. Memory • System Wide Memory Bandwidth
• CPU Clock • STREAM TRIAD (Node)
• CPU Peak Gflop/s • STREAM TRIAD (System)
• System Peak Gflop/s • Number of CPUs
• # of Single System Nodes • Total System Memory &
• US List Price ($M) Interconnect Bandwidth
• FCS (Year) • Processor Capability Rating
• IDC Market Segment • Memory Capability Rating
• One Computer or Cluster • Scaling Capability Rating
• CPU Linpack Rmax • Overall IDC Rating
• Total System Linpack • Price/Performance
Example Balanced Rating List
Processor Memory Scaling
Number Capability Capability Capability Overall IDC
Vendor Model of CPUs Rating Rating Rating Rating
# 0 to 100 0 to 100 0 to 100 0 to 100
NEC Earth Simulator 5,120 11,867 99,328 10,240 40,478
IBM ASCI White,SP Power3 8192 9,354 3,149 2,099 4,867
HP LANL AlphaServerSC45 P-2 4,096 7,046 3,473 1,243 3,921
HP PSC AlphaServer SC45 3,016 5,270 2,557 915 2,914
HP CEA DAM AlphaSvr SC45 2,560 4,502 2,176 777 2,485
SGI ASCI Blue Mountain 6,144 4,509 1,230 1,368 2,369
IBM SP Power3 375 MHz 16 way 3328 3,800 1,279 759 1,946
IBM eServer pSeries p690 Turbo 928 2,887 1,657 508 1,684
NEC SX-6/192M24 192 511 3,725 346 1,527
HP LANL AlphaServerSC45 P-1 1,536 2,779 1,302 466 1,516
IBM NERSC SP Power3 375 NH2 2,528 2,895 972 648 1,505
IBM eServer pSeries p690 Turbo 800 2,488 1,429 438 1,452
IBM eServer pSeries p690 Turbo 768 2,389 1,372 420 1,393
IBM eServer pSeries p690 Turbo 768 2,389 1,372 420 1,393
Fuj itsu VPP5000/100 100 317 3,413 404 1,378
NEC SX-5/128M8 128 397 3,356 282 1,345
IBM eServer pSeries p690 Turbo 704 2,190 1,257 385 1,277
Intel ASCI Red 9,632 793 1,027 1,926 1,249
IBM eServer pSeries p690 512 1,350 1,524 253 1,042
NEC SX-6/128M16 128 341 2,483 230 1,018
Example List – Price/Performance
Number of Overall IDC Price/
Vendor Model CPUs Rating Performance
# 0 to 100 List/ IDC Rate
HP rx2600 @900MHz 2 3 4.6
NEC Earth Simulator 5,120 40,478 4.9
HP rx2600 @1GHz 2 4 5.8
HP i2000 (Itanium)@733 1 1 9.1
HP J6000 (PA-8600) 2 2 12.0
HP RX4610 (Itanium)@733 4 3 12.3
HP i2000 (Itanium)@800 2 1 13.0
HP RX4610 (Itanium)@733 2 2 13.8
SGI Origin 300 500 128 98 14.6
SGI Origin 300 600 8 6 15.1
HP J6700 (PA-8700) 2 2 15.2
SGI Origin 300 500 8 5 16.8
HP RX4610 (Itanium)@800 4 4 17.4
HP L3000 (PA-8600) 4 3 17.7
SGI Origin 300 600 32 23 18.0
SGI Origin 300 600 128 94 18.1
SGI Origin 300 600 64 47 18.2
HP RX4610 (Itanium)@800 2 2 18.4
HP rp5470 (PA-8700)@650MHz 4 4 18.8
SGI Origin 300 600 16 12 19.3
HP L3000 (PA-8600) 2 2 19.4
HP rp5470 (PA-8700)@650MHz 2 2 19.8
SGI Origin 300 500 64 40 20.1
SGI Origin 300 500 384 233 20.2
In Summary:
Why Worry About HPC Metrics?
HPC Users and vendors identified it as one of their
top issues
The range in prices as a ratio to peak
performance is over eight times
– Indicating a need for a better metric for
comparing computers
Price as a ratio to peak performance is
becoming less meaningful each year
– Yet it is broadly used (and abused)
– LINPACK closely tracks peak
Balanced Ratings Conclusion
Despite the strategic and economic importance of
supercomputers, or HPC systems, there have been no
generally available databases or tools for analyzing
these complex technologies along multiple dimensions
User organizations need a source of common system
specification data, and top-level tools for sorting through
various aspects of the technologies to better understand
and explain how different technologies match their
requirements
The IDC Balanced Rating provides one such top-level
analysis and ranking
