‘Tis the season for IDC’s annual HPC market forecast, only this time around it needs to consider a global economic recession. In this exclusive interview, HPCwire quizzes Earl Joseph, IDC’s program vice president for HPC, about what’s in store for 2009.
HPCwire: IDC recently revised its forecast for the HPC market. In a nutshell, what is the new forecast?
Earl Joseph: Based on actual numbers for the first three quarters of 2008 and modeled fourth-quarter numbers, IDC estimates that full-year 2008 HPC server revenue will come in at around $9.6 billion. That’s down 4.2 percent from 2007. Our new forecast predicts HPC server revenue will dip about 5.4 percent in 2009, then start modest growth again in 2010 and rebound to 9 percent-plus growth, eventually reaching $11.7 billion in revenue by 2012.
HPCwire: Why did you revise the forecast?
Joseph: We normally do a new five-year HPC market forecast at this time of year. The third-quarter data on HPC servers that we received from the hardware vendors in late 2008 showed that the global economic recession was already throttling down revenue in HPC. That greatly affected our new forecast.
HPCwire: How did you come up with the revised forecast?
Joseph: Each year we go through a specific, careful process to come up with a new five-year forecast. Our HPC team that includes Jie Wu, Steve Conway, Richard Walsh and myself gets together for two full days to go through this process. We looked carefully at the actual data for the prior five years, especially the most recent quarters. We analyzed IDC’s assumptions and projections for the global IT market and the whole server market. Then we first created a Q4 forecast that also provides a full year 2008 forecast. Then we constructed a table of assumptions about the factors that are most likely to influence the HPC server market in 2009 and beyond. With this foundation, we created our five-year forecast for the HPC server market, the competitive price band segments of the market, and so on.
HPCwire: How confident are you in the forecast?
Joseph: We’re fairly confident about the major assumptions and the general trends in the predictions. We all wish that we had a crystal ball that would tell what is going to happen in the overall world economy. It is always harder to forecast during periods of major ups and downs, and this was one of the hardest times for creating forecasts. Our forecasts are intentionally on the conservative side and in recent years the HPC market has consistently beaten our forecasts.
Another thing that gives us confidence is that in late 2008, when the impact of the economic downturn was already starting to be felt, we conducted an extensive worldwide, in-depth study of 110 HPC sites of all sizes and in all major sectors. These were two to three-hour interviews that produced more data points than what would fit into Excel. The topics ranged from HPC systems to processors, storage, interconnects, system software, budgets, TCO, and application workloads, and we’re churning out separate reports on these topics right now. We asked the sites not just about what they’re doing today, but also about their requirements for the next round of HPC purchasing, including the attributes that would command premium pricing. Not one of the 110 government, industrial, and academic sites planned to reduce HPC use in 2009, although we expect them to be more conservative about new spending. We received a good general sense of where budgets and spending are headed directly from the HPC buyers. That gave us some additional confidence when we put together our forecast.
HPCwire: How does HPC compare with the whole IT market?
Joseph: If our baseline assumptions about the HPC market are right, HPC will come out of the recession the same way it went in, as a bright spot in the IT sector. As I mentioned earlier, we expect HPC to start growing again in 2010 and to be on a robust growth path again by the end of our forecast period in 2012.
HPCwire: What will the main effects of the global economic recession be, where HPC is concerned?
Joseph: We expect users to become more conservative about new spending, but most existing plans will go forward, though there may be delays of a quarter or more in some cases. There will be more focus on cost-effectiveness and this will favor clusters and other standards-based solutions. Competition will heat up for new business, and some weaker vendors may close their doors while stronger ones tighten their belts. More sites will apply simulation and analysis to their existing datacenter designs to grow performance with minimal impact on power, cooling, and facility space. And with tighter controls on capital spending, some increases are expected in CAPEX-free HPC cycles delivered via service-oriented grids, or maybe cloud computing in some cases. This will become more appealing to new users and for periodic, overflow work.
HPCwire: Which HPC segments will be most affected by the global economic recession?
Joseph: Some automotive and financial services firms are so hard pressed that we see them shrinking CAPEX even in mission-critical areas, including HPC. North American automakers will generally take more drastic steps than their Japanese and European counterparts in these reductions.
In sharp contrast, HPC is deeply embedded in the R&D process of oil and gas companies, and most of these companies are in good shape financially even with lower energy prices. We don’t expect much in the way of budget cuts and we see HPC growth plans being carried out, although some purchases may be delayed.
Government and academia together make up over 65 percent of the HPC server market. They’ll probably follow historical patterns and react less quickly and less deeply to the economic downturn than the private sector does.
HPCwire: How much of the HPC market does government spending make up? What impact does IDC expect from the Obama Administration?
Joseph: The U.S. Government is the world’s largest HPC customer, and the new U.S. Administration has said it plans to boost spending on science and technology, so that’s a hopeful sign. HPC should also be critical for the alternative energy research that is one of President Obama’s top priorities. But in the U.S. and around the world, HPC will compete for funding with other urgent priorities and not all new HPC initiatives will get funded, or funded fully, in 2009. And in the U.S., the change of Administration could delay funding for new initiatives and for the expansion of existing HPC-related science and technology programs. We expect that some weapons-based HPC work will likely be redirected and that some procurements may be delayed to free up funding for higher priority projects.
HPCwire: Will IDC revisit this forecast?
Joseph: We plan to fully update the forecast once a quarter for a while, after we receive and analyze the results from vendor quarterly sales. We are hopeful that the HPC market will recover faster than we are currently projecting, but it will depend heavily on how long the economic slowdown lasts.
HPCwire: What other major developments do you expect in the HPC market in 2009?
Joseph: Overall, we expect 2009 to be another year of evolutionary change in the HPC market. Incremental advances will ease the pain of dealing with the massive increase in core counts, but they won’t cure the big issues of highly parallel programming, power and cooling costs, software licensing costs, ease of use, and so on. There will likely be a number of exciting new petascale installations in 2009.
The HPC storage market will stay stronger than the server market through the recession period. We expect that “ease-of-everything” solutions will continue to grow in the low-end workgroup segment and start spreading to higher price point systems. The research we did for the Council on Competitiveness showed that HPC use is already a metric for industrial competitiveness in tier 1 firms. In 2009, we think HPC use in these firms’ supply chains will start to become a competitiveness metric.
Standards-based clusters will gain market share in the price-sensitive economy, but more and more HPC sites will experience retrograde performance on some key codes. In a survey we did in the second half of 2008, half of the HPC sites said that within the next 12 months they expected some of their codes to run more slowly on their newest HPC system than on the previous one. That’s a disturbing new trend that’s being driven by escalating core counts and the inability to move data in and out of each core fast enough to keep the cores busy. It’s exacerbated by energy-saving, tuned-down processor speeds that reduce single-threaded performance. Most HPC sites would rather see faster clock rates.