IDC HPC Market Update July 2012 Title of Presentation Client IDC HPC Market Update July 2012 Earl Joseph – ejoseph@idc.com Steve Conway – sconway@idc.com Chirag Dekate – cdekate@idc.com Lloyd Cohen – lcohen@idc.com Charlie Hayes – charles.w.hayes@verizon.net Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

IDC’s HPC Team Earl Joseph Beth Throckmorton Steve Conway IDC HPC research studies, HPC User Forum and strategic consulting Steve Conway Strategic consulting, HPC User Forum, market trends, Big Data Chirag Dekate HPC QView, technology trends, Big Data, innovation awards program Lloyd Cohen HPC market analysis, data analysis and workstations Beth Throckmorton Government account support and special projects Charlie Hayes Government HPC issues, DOE and special studies Mary Rolph HPC User Forum conference planning and logistics

IDC Standard HPC Activities Track all HPC servers sold each quarter Plus researching all HPC products (over 100 vendors) 4 HPC User Forum meetings each year Publish 35 plus research reports each year Visit all major supercomputer sites & write reports Assist in collaborations between buyers/users and vendors Assist governments in HPC plans, strategies and direction Assist buyers/users in planning and procurements Maintain 5 year forecasts in many areas/topics Conduct special research studies

Title of Presentation Client 2012 IDC HPC Research Areas Title of Presentation Client New And Potentially Disruptive Technologies & New Approaches Big data Co-processors All types of flash/SSDs New software solutions Government programs to help bring to market new capabilities Special HPC Research Areas & Reports: A new user-based data tracker product End-user based MCS reports: clusters, processors, accelerators, storage, interconnects, system software, and applications The evolution of government HPC budgets Emerging markets including China, Russia, etc. SMB and SMS research The HPC award program The evolution of clouds in HPC Scaling of software – issues and soultions Worldwide Petascale and Exascale Initiatives S Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Example Research Studies & Surveys EU 2020 HPC leadership plan SMB/SME/SMS special studies DOE skills, talent and hiring plan EU parallel software scaling plan Automotive benchmark analysis Oil/gas HPC usage comparisons NCSA/NSF science underlying applications study Council on competitiveness studies Power and cooling studies Vendor product concept & TAM studies And many private studies

Top Trends in HPC The global economy in HPC is growing again 2010 grew by 10%, to reach $9.5 billion 2011 grew by 8.4% to reach $10.3 billion We are forecasting ~7% growth over the next 5 years Major challenges for datacenters Power, cooling, real estate, system management Storage and data management continue to grow in importance Software hurdles continue to grow GPUs are seeing real tractions in the market The worldwide Petascale Race is in full speed

2011 HPC WW Market Results: By Competitive Segments Title of Presentation Client 2011 HPC WW Market Results: By Competitive Segments C Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

2011 HPC WW Market Results: By System Units Sold Title of Presentation Client 2011 HPC WW Market Results: By System Units Sold C Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

2011 HPC WW Market Results: By Price Bands Title of Presentation Client 2011 HPC WW Market Results: By Price Bands C Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

2011 HPC WW Market Results: By Region Title of Presentation Client 2011 HPC WW Market Results: By Region C Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

HPC in EMEA in 2011

2011 HPC WW Market Results: Processor Packages Shipped Title of Presentation Client 2011 HPC WW Market Results: Processor Packages Shipped C Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

HPC Vendor Revenue Shares Title of Presentation Client Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

HPC Forecasts And Growth Areas Title of Presentation Client HPC Forecasts And Growth Areas Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Title of Presentation Client New HPC Forecasts Title of Presentation Client 2011 was a very strong year and we are projecting 7% yearly growth to 2016 2016 should exceed $14 billion Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

New HPC Forecasts: By Competitive Segment Title of Presentation Client New HPC Forecasts: By Competitive Segment Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

New HPC Forecasts: By Verticals Title of Presentation Client New HPC Forecasts: By Verticals Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

The HPC Market Beyond The Servers Title of Presentation Client The HPC Market Beyond The Servers Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

The Broader HPC Market

IDC Study Results: HPC Talent And Skill Sets

Study Background IDC conducted this study for the Department of Energy (DOE) National Nuclear Security Administration (NNSA) and surveyed three constituencies: DOE national laboratories, major HPC data centers, and academic and training organizations with strong HPC ties. Objective: investigate current/future HPC talent/staffing issues and requirements for reaching petascale, and eventually exascale, computing capability. A total of 41 organizations participated in the study. 7 were DOE government labs, 27 were other major HPC centers and 7 were in academia or other training organizations.

Key Findings #1 -- There is a serious shortage of qualified candidates for HPC positions today 93% of the HPC centers said it is "somewhat hard" or "very hard" to hire staff with the requisite skills. Most of the centers (56%) fell into the "very hard" category. The DOE labs also have a shortage of qualified people. This shortage spans the government, university, and industrial sectors and is global in scope. It results largely from the post-Cold War period of slowdown in the worldwide HPC market (ca. 1990-2001) and the HPC industry's failure to shed the "old technology" image it acquired during that period – while Internet, PC and gaming ventures seized the "new technology" label.

Key Findings #2 -- The candidate shortage affects some job categories more than others The job categories that the HPC center respondents consider hardest to find are: Scientists with HPC capabilities, parallel programmers, algorithm developers, and system administrator with high-end computing experience. The DOE labs' list of hard-to-find skills mirrors these same categories. The skills most badly needed today include: A combined understanding of a scientific discipline and computational science and/or computer science; parallel programming and code optimization, especially for scaling to large processor/core counts; algorithm development; and understanding of parallel file systems

Key Findings #3 -- The candidate shortage affects the ability of HPC centers to staff up fully But it has not yet seriously affected the quality of their support staff. Only two of the 27 centers (7%) characterized their support staff size as "very adequate." When "somewhat adequate" responses are added to that figure, the total was still only a 41% minority of the centers. But the vast majority (93%) of the centers rated support staff quality as either "very adequate" (44%) or "somewhat adequate" (48%).

Key Findings #4 -- The best sources of job candidates are universities & other HPC data centers The most fruitful source of qualified candidates for HPC positions are university graduates in mathematics, engineering, or the physical sciences (cited by 63% of the respondents). A smaller but substantial percentage of the respondents (48%) pointed to university graduates in computer science. The most productive non-academic source of qualified HPC job candidates (48% of the respondents) was employees of other HPC data centers. Followed by "employees in related positions/areas at our own site" (41%) and employees of HPC software or hardware vendors" (37%). 22% of the respondents had found qualified candidates ”through on-line posting and placement resources," such as monster.com and other websites.

Key Findings #5 -- Qualified job candidates from non- HPC centers are hard to find, but often work out well The least productive hunting ground for qualified HPC positions was "employees at commercial or general IT data centers (non-HPC centers)." Despite this, the majority of respondents (89%) reported that they have hired people from non-HPC, commercial environments to work in their HPC centers. And the number of these hires that worked out well outnumbered the failures by a five-to-one ratio (20:4) – although this does not mean that the successes were immediate or complete.

Key Findings #6 -- The cost of transferring an employee from a non-HPC technical job is not especially burdensome But it can take two years or more to become productive in the HPC environment. HPC centers generally do not characterize the transfer cost as unreasonably large compared with the cost of getting any other new employee up to speed on the job. More important is how much relevant experience the person brings to the HPC position. Ease-of-transfer from non-HPC technical disciplines depends on which other discipline the individual is coming from and which HPC position the person is heading into. This can make a major difference in determining how long it takes until the person is performing to expectations in the HPC environment. The estimates vary from three months to two years, with many responses in the 6-12 month range.

Key Findings #8 -- The most important task is preparing people to exploit highly parallel HPC systems There was remarkably strong agreement that the number one disruption is already far along – the shift to systems that are based on large numbers of multi-core processors with inadequate per-core bandwidth. These systems pose enormous challenges for existing algorithms and applications. Although the HPC community as a whole has been delaying the inevitable, sooner or later the sustained-to-peak performance ratio on these systems will become so low that the HPC community will need to take action and rewrite algorithms and application codes, presumably using more efficient programming models.

Key Findings #9 -- Surrounding HPC systems is a complex of interrelated issues that need to be tackled together These issues include not just extreme parallelism, but also petascale/exascale computing, HPC system heterogeneity, HPC system architectural balance, HPC system reliability, and HPC system and data center power and cooling. Any one of these inflection points could become a disruptive factor that hampers progress in HPC. There is no easy or inexpensive way to get past these hurdles. Instead, substantial thought and investment will be needed.

Key Findings #10 -- Expanding the markets for HPC could make HPC a more attractive career path HPC architectures are becoming more narrow-purpose, making this even harder. Multiple HPC center respondents said that expanding the markets for HPC could help address the technology challenges by providing vendors with more incentive to innovate for HPC users. . This could spark investment and make HPC a more vibrant, attractive market for young people to enter. But others point out that the recent evolution of HPC hardware architectures has been in the opposite direction, making them useful for fewer and fewer types of problems. More balanced architectures could increase the systems' breadth-of-applicability

Key Findings #11 -- To grow the HPC talent pool, the best hope is to encourage universities The goal here is to integrate their HPC training more fully into scientific and engineering curricula. Universities need to expand coursework in computational and integrate computational science methods into the requirements for science degrees. Because HPC-based modeling and simulation has firmly established itself as the third branch of scientific inquiry, respondents say HPC should be as much a part of scientific education curricula as the two more established methods. Another strong recommendation is for government, academia, and industry to collaborate to increase the number of internship and fellowship opportunities. To make a major improvement , the U.S. government needs to make this a priority and invest accordingly.

The IDC HPC Innovation Award Program Title of Presentation Client The IDC HPC Innovation Award Program Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

We Are Collecting A Large Set Of HPC ROI Examples Title of Presentation Client We invite users to submit their examples at: www.hpcuserforum.com/innovationaward/ Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

New HPC Innovation Award Program: www. hpcuserforum New HPC Innovation Award Program: www.hpcuserforum.com/innovationaward/

HPC Award Program Goals Title of Presentation Client #1 Help to expand the use of HPC by showing real ROI examples: Expand the “Missing Middle” – SMBs, SMEs, SMSs -- by providing examples of what can be done with HPC Show mainstream and leading edge HPC success stories #2 Create a large database of success stories across many industries/verticals/disciplines To help justify investments and show non-users ideas on how to adopt HPC in their environment Creating many examples for funding bodies and politicians to use and better understand the value of HPC  to help grow public interest in expanding HPC investments For OEMs to demonstrate success stories using their products Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Users Have to Submit the Value of the Accomplishment Title of Presentation Client Users are required to submit the value achieved with their HPC system, in any of 3 broad categories: a) Dollar value of the HPC usage e.g. made $$$ in new revenues, saved $$$ in costs, made $$$ in profits, etc. b) Scientific or engineering accomplishment e.g. discovered how xyz really works, develop a new drug that does xyz, etc. c) Value to society as a whole e.g. ended nuclear testing, made something safer, provided protection against xyz, etc. … and the investment in HPC that was required (in order to calculate the ROI) Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Title of Presentation Client The Winners At ISC’12 Title of Presentation Client We recognize these sites for their excellence in applying HPC to solve key business and scientific problems Site (Alpha Ordered) Person Success Area Org Size Beijing Genomics Institute Lin Fang ROI & Eng Large GE Global Research Richard Arthur Bus Impact & Eng Large DOD HPCMP CREATE John West ROI & Society Large Aon Benfield Securities Peter Phillips, Amir Mohammed Bus Impact & Eng Small Mary Bird Perkins Cancer Research / LSU Wayne Newhauser Sci/Eng/ROI & Society Large Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Title of Presentation Client The Trophy For Winners Title of Presentation Client Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Title of Presentation Client In Summary Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Why HPC Is Projected To Grow Title of Presentation Client It has become a competitive weapon For companies, universities and governments Global competitiveness is driving R&D and better product designs Even small companies can use HPC to gain market share Governments view HPC leadership as critical For national pride, but more importantly for economic prosperity It use to be 1 large supercomputer – now its multiple ones There are very critical HPC issues that need to be solved Global warming, alternative energy, safe NE, financial disaster modeling, healthcare, homeland security, … And 3D movies and large scale games are fun At the same time, “live” science and “live” engineering costs have escalated And time-to-solution is months faster with simulations Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

But There are Still Major Customer Pain Points Title of Presentation Client Software is the #1 roadblock Better management software is needed Parallel software is lacking for most users Many applications will need a major redesign Clusters are still hard to use and manage System management & growing cluster complexity Power, cooling and floor space are major issues Third party software costs RAS is a growing issue Storage and data management are becoming new bottle necks Lack of support for heterogeneous environment and accelerators ROI is becoming a requirement, especially as system costs escalate Multi-core issue: hard to use(apps don’t scale well (or no apps at all) to take advantage of the processing capability), memory bandwidth limit, Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.

Title of Presentation Client Questions? Please email: hpc@idc.com Or check out: www.hpcuserforum.com Copyright 2007 IDC. Reproduction is forbidden unless authorized. All rights reserved.