450 likes | 769 Views
The Role of High-Performance Computing in Industry: Customer Perspectives and Market Analysis. Prepared for: NSA Prepared by: Debra Goldfarb Vice President, Workstations and High-Performance Systems December 7, 1998.
E N D
The Role of High-Performance Computing in Industry: Customer Perspectives and Market Analysis Prepared for: NSA Prepared by: Debra Goldfarb Vice President, Workstations and High-Performance Systems December 7, 1998
The Role of High-Performance Computing in Industry: Customer Perspectives and Market Analysis • Agenda • Project overview • Key findings • Executive summary • Business dependence on HPC • Future demand characteristics • Reliance on Japan for HPC • Future applications • Government role • Market Analysis detail • Conclusions Presentation Focus
Project Overview • HECC commissions IDC to undertake a study focusing on the current and projected use and importance of the most powerful high-performance computers in the industrial sector. • The study had three major objectives. • Gain an understanding of the industry’s current and future requirements for computers to help solve the most demanding problems. • Identify interdependencies between high-end HPC and business, economic, and societal advancement. • Analyze the dynamics in the HPC market, identifying future growth potential and future risks.
Voice of Industry “The high performance computing industry in the United States today appears to be almost as if someone hit the pause button. We’re seeing a reduction in innovation at the processor level and a tendency to follow a single manufacturer or Intel. If we lose this advanced HPC platform we stand on, innovation begins to slow down.”
Voice of Industry “It is absolutely unthinkable that the U.S. should give up its leadership of HPC. Innovation and technology leadership are what U.S. industry is famous for. We’re the innovators. We’re the leaders. Our HPC technology leadership helps drive our IT industry leadership. It’s enabling other industries to be competitive and to let something like this go, it just doesn’t make any sense …”
Voice of Industry “You can’t buy your infrastructure from someone else. You really have to own it—create it… It’s got to be protected… What if all your medicine came from another country, or all of your food came from another country? Or how about this question...What if all your oil was coming from another country--and then all of a sudden they decided that the prices were too cheap--so that they would have to reduce supplies to increase price. I would hope our politicians would know the answers to those questions. Aren’t those pretty obvious? You’ve got to own some of your future.”
Executive Summary 7 HPC Industry in Crisis • High-end HPC users: • Are very concerned about the state of U.S. HPC industry • See U.S. leadership in HPC as highly related to U.S. industrial leadership • Perceive U.S. economic and societal advancement dependent upon high-end HPC • Believe government needs to advance the technology and help provide access to the highest capability tools • User community feels they have limited ability to influence government or industry; they feel isolated
Executive Summary (Cont.) • Primary high-end HPC benefit: Provides critical time-to-solution advantage that enables market leadership • Primary barrier: Lack of business and engineering management priority for investment in high-end HPC • Industrial HPC users will buy Japanese if no U.S. alternative exists despite recognition of significant negative impacts
Key Findings - cont’d • Scientists and engineers with a requirement for specialized scientific computing are: • Very concerned about the viability of the U.S. high-end HPC industry and the potential negative impact on their competitive position • Key industries with concern: automotive, aerospace, pharmaceutical, chemical • Those pursuing MPP/scalable architectures are less concerned • Petroleum, electronics, aerospace (some applications), entertainment, financial
Key FindingsBusiness Dependence on High-End HPC 10 • Consensus that high-end HPC is fundamental in driving innovation into product development, creating process efficiencies, and in advancing scientific understanding. • Primary benefit: Provides critical time-to- solution advantage that enables being “first to market.” HPC Linked to Global Competitive Position
Value to Industry The value of HPC in analyzing compounds is very dramatic. It would cost about $50,000 per compound to analyze in a lab. Computationally, we can do it for about $3,000. Turnaround time in a lab is about nine months; on a computer, it is less than one week. If you can advance a drilling date for a pipeline connection by days, and then look at the impact of that change on the capital investment curve, it can be huge. Changing the slope of that curve by a few percent, over a 10-20-year period will have a tremendous dollar impact on us. Reducing cycle time is one of the key business drivers for us.
Value to Industry • “Solving the simulation problem well is easily worth a billion dollars or more a year to our company.” • “A die for a stamped metal part can cost $1 million. In addition, the costs for stopping an assembly line due to a broken die or to problems associated with the design can be as much as $100,000 per minute. Through simulation and modeling, engineers can identify design problems prior to casting the die. The pay back on this ability alone has been enormous.”
Product Development Process Concept Prototype/ Manufacturing Final Development Physical Testing Design Preparation/Delivery Without/ HPC With HPC Time Savings HPC-Supported Simulation Time to Market HPC’s Key Impact Reduction in Time to Market
Key FindingsFuture High-End HPC Demand Characteristics “Current Science/Eng. Problems Need More!” • Today’s HPC architectures do not meet tomorrow’s performance requirements. “We’ve been actually fairly saddened by the fact that there hasn’t been much done in the last few years with the high-end. The high-end is drying up. It’s unfortunate. The biggest barrier to the high-end of the market from our perspective is that they’re simply not available. That’s the biggest barrier. We can’t buy a machine with enough power. Cost would of course be a consideration, but we would evaluate the cost based on the power of the machine and what it could do for us.”(Automotive)
Key FindingsFuture High End HPC Demand Characteristics (Cont.) • Users are alarmed with the projected design points of future systems (designed around mid-range problems and not the most demanding problem set). • ASCI-class systems are not considered a viable solution • Experimental • Theoretical vs. actual performance gap • High risk • Limited, if any software
Key Findings Future High-End HPC Demand Characteristics (Cont.) • Majority of respondents are faced with scientific / engineering problems demanding 10 to 1,000x performance. “Vehicle synthesis, multi-system simulations — where you have multiple links and interdependencies to systems. We would use this for optimization as well as exploratory problems where we are trying to push new limits.” (Automotive) “10X would allow us to make fifty million compounds accessible; that is to compute them, store them, etc. That’s four times what’s previously known. So that’s interesting but it’s not as interesting as forty times or fifty times. 40X would allow you to ask and answer a lot of questions about this molecular diversity.”(Pharmaceutical)
Key FindingsReliance on Japan for High-End HPC Open, but Concerned • Most said that barring government restrictions, they would buy the best tool available for their needs — regardless of the source. • However … most recognized significant implications to the loss of U.S. leadership in high-end HPC. “If Japanese manufacturers maintain their investment in supercomputers, given the trade sanctions imposed today, we will suffer economically.”
Key FindingsUser Concerns: Japan as HPC Leader Significant threat to U.S. technology leadership • Ecosystem migration -- The scientific / engineering ecosystem includes a tight community of scientists, engineers, supercomputer suppliers, ISVs, and others that enable science and engineering to advance. Supercomputer capability migration to Japan will disrupt these ecosystems; related skills may migrate to Japan.
Key FindingsUser Concerns: Japan as HPC Leader (Cont.) • Security risk -- High risk of losing industrial secrets to Japanese companies. • Equal access -- U.S. scientists will not have early access to leading edge HPC systems. • Opportunity loss -- Japan may gain an edge in breaking into early markets. • Loss of impact -- U.S. scientists and engineers will have less influence on supercomputer design.
Key FindingsHPC Advancement and Business/Economic Growth HPC Impact: Drives U.S. Leadership • Industrial leadership • Auto industry embraced simulation and modeling to regain competitive footing with Japan • Rapidly evolving bioscience sector • Information technology industry leadership • Government investment in advanced HPC technology results in commercial IT product advances downstream • Environment • Ecosphere modeling to evaluate impact of fluorocarbons on atmosphere
Key FindingsHPC Advancement and Business/Economic Growth (Cont.) • Safety • Automotive simulated crash tests increase safety and reduce morbidity and mortality • Health Care • Vascular simulation to understand heart disease and cardiac processes • Virtual surgical planning
Next Generation Applications:Examples Increasing Complexity • Automotive • Full fidelity crash • Virtual test track • Computational steering • Petroleum • Seismic and reservoir visualization • Improved substrate modeling • Forward analysis of oil-bearing structures • Chemical • Developing new biological compounds • Process modeling • Ecospheric modeling
Next Generation Applications:Examples (Cont.) New, Evolving Applications • Pharmaceutical • Properties modeling • High throughput screening • Cell membrane modeling • Human Genome data analysis • Health Sciences • Physiological modeling • Vascular simulation • Virtual surgical planning • Integrated tomography
Key Findings:Leadership Void Industry Needs Government Leadership • General belief that the government should drive investment in advanced HPC systems • Government needs to facilitate widespread access to the most advanced systems “Loss of leadership in basic HPC tools and infrastructure would be disastrous. How can we even think of losing our advanced technology advantage.”
Key Findings…Bottom Line The current HPC market situation threatens to disturb and potentially damage the ability of U.S. companies to continue to leverage capability computing and therefore maintain their competitive position in key global industries. • Potentially slow the rate of innovation • Negatively impact societal benefits • Risk of unbalanced access
Capability Market Fundamentals • Capability systems account for less than 1% of total systems sold (175 to 341 systems), and from about 17.4% to 25.5% of revenues ($884 million to $1.147 billion). • Assuming current demand conditions, IDC expects the capability market to grow by about 2.8% or 2.9% (CAGR) in revenue between 1997 and 2002, totaling $1.0 to $1.3 billion by 2002. • This is a mature market with limited growth potential, although given the drain from HPMR systems, the market has remained fairly stable. • New applications • Strong ROI
Capability Market Fundamentals (Cont.) The capability market is: • Split between two competing technologies (vector supercomputers and MPPs) • Divided between three major Japanese vendors (NEC, Fujitsu, and Hitachi) and two (potentially one or zero) major U.S. vendors (IBM, SGI/Cray) • Experiencing a significant shift in underlying market behavior due to competition from midrange products
High-Performance Midrange:Impacts on the Capability Market • The growth of the midrange, while having a positive “overall market” effect, has negatively impacted the capability segment by: • Severing the relationship between capacity and capability computing • Constraining the size and growth of the capability market through the “application pipeline effect” • Creating significant barriers to entry for new technical computing companies • Intense competition blocks new entrants • High R&D costs vs. limited size of market impacts investors and opportunity
Lower capability investment keeps cost high Capability market participation demanding ($200–500M) Capability demand growth flat Reduces capability market High capability market entrycost reducesROI appealof market RAAS offloads capacity; exposes limited capability-only demand DOD, Government reducing capability spending Vendors see poor ROI in capability segment; resist spending Governmentdrivescommodity-based solutions RAAS midrange absorbing more capacity computing Capacity market return becomesmore appealing Industrial capacity demand growing Capacity computing cost drops, increasing demand HPC Industry Dynamics
Worldwide HPC Revenues by Type, 1992–1997 The capability segment represents approximately 1% of units and between 17–25% of revenues.
HPMR HPMR: Gaining Market Share $2,000 $1,800 $1,600 $1,400 $1,200 $1,000 Rest of Market $800 $600 $400 $200 $0 1992 1993 1994 1995 1996 1997 Revenue ($M)
2002 $8.9B Worldwide Technical Market Forecast: Type Share of Revenues 1997 and 2002 1997 $5.1B
Is There a Business Case for Capability Computing? A rule of thumb in this industry: • New generation “capability system” costs between $200 to $500 million to bring to market. • Assuming each of the five vendors brings a new product to market every three years, and that R&D expenses average $300 million per company, then the industry is investing $500 million each year just to develop new innovative products. • This is roughly half of the total yearly revenue, and does not include production, sales, marketing, and other expenses.
Is There a Business Case for Capability Computing? (Cont.) • Given the above analysis, it is difficult to make a pure business case for computer systems manufacturers to invest in researching, developing, and bringing to market a continuing stream of capability oriented high-performance computers …
Business Case (Cont.) • Other factors to consider would include: • Spin-off technology — Technologies developed for supercomputers are eventually incorporated into more general purpose products. • Japanese computer vendors use this rationale consistently. U.S. vendors, on the other hand, tend to invest in R&D that can be more directly tied to existing products with more immediate returns on investment.
Business Case cont’d • Corporate positioning — Computer vendors could view high-performance computers as flagship products to establish them as a “world class” player in the overall computer market. • U.S. systems vendors have moved away from pure technology as a differentiator, and more strongly towards factors such as financial performance market share, and becoming a “solutions provider,” especially for commercial applications.
Business Case cont’d • Support of national programs — Some computer vendors have developed high-performance systems in support of national security, science, industrial or societal goals. • There is no sure path for a vendor between supporting national security/government program efforts and the next generation capability computers that will support industrial requirements.
Industry’s Alternatives • Develop Specialized Solutions • Appeal is the perception of low cost and high control • Reality is high cost, high risk, and poor performance • Wait for More Powerful General Purpose Solutions • Moore’s Law is a compelling proposition • Potentially high risk • This strategy is only effective when all competitors have access to the same class of computers. The U.S. could be at a severe disadvantage in bringing new products to market if a significant disparity in available computer technology developed.
Industry’s Alternatives • Purchase foreign built systems • Risk of equal access • Loss of ecosystem • Control over intellectual property/trade secrets • Risks to competitive position in key industries including IT
A few closing thoughts... • Strengthen the market • Reach out beyond the converted. • Build the message around “business value” • Support the vendor community • Make participation in government programs a financially viable activity (money and leverage) • Government technology initiatives need to balance building increasingly exotic and complex systems with the realities of the industrial data center
Thoughts…cont’d • Build bridges across the HPC user community • Set up an integrated user organization supporting government, academia and industrial users • Use this forum to influence the future direction of the industry, technology and policy • Provide a community to share common issues, challenges and successes