If high performance computing is such an innovation accelerator, why don’t more companies embrace the technology? Although HPC is growing rapidly in relation to other IT realms, it still only represents about three percent of the computing market. Last week, the Council on Competitiveness released two new studies that looked at why technical computing users on desktop systems are not graduating to HPC servers. While previous Council reports explored why HPC users often failed to advance beyond entry-level HPC platforms, these latest reports focus on why some desktop-only companies have never taken that first step into HPC.
According to Council on Competitiveness Vice President Suzy Tichenor, “If we’re going to be an innovation-based economy, we really have to make sure there is widespread access to the use of high performance computing. What these studies show is that current access is limited.”
Tichenor compares the current perceptions of HPC to that of quality control 25 years ago. In the late 80s, the Japanese achieved global dominance in several industries due to the implementation of better quality practices. The ensuing products, everything from automobiles to consumer electronics, gave Japan a global competitive advantage for years to come. Prior to that, investing in quality was considered an expense that didn’t deliver much return. The Japanese proved that the ROI was really there. Eventually, their quality programs gained wide acceptance and were copied by other industrialized nations.
In the broader market, high performance computing is still perceived as exotic technology. Of course, not every prototyping or modeling application needs more power than a desktop can deliver. Some companies with less demanding software will just be able to ride the performance curve on the desktop as these systems incorporate more powerful processors. But the Council studies found that the majority of the firms surveyed — 57 percent — do have problems that are beyond the capabilities of current PCs or workstations. And about a third of the firms surveyed have a desire to move up the computing ladder onto HPC platforms.
The studies looked at firms doing desktop-only technical computing in a range of vertical industries. The analysis revealed the main barriers that prevented users from moving desktop applications to HPC systems, which included the lack of:
- application software for parallel platforms;
- human talent and expertise to apply technical computing applications to HPC; and
- an ROI model for increasing HPC investment.
In fact, these barriers also exist for entry-level HPC users looking to move to more powerful systems, but for desktop users, the barriers were magnified. Unlike companies that have already invested in HPC, firms using desktop-based technical computing have little if any experience with high performance computing systems. In some cases, the lack of expertise means that even if HPC-ready software is available for their particular application, they might not be aware of it. As a result, they couldn’t build an ROI case for its use. “These barriers are really all interlinked,” notes Tichenor.
Many desktop-only firms are part of the supply chain for larger manufacturers. So while the individual components may be modeled on the desktop, the systems these components end up in need to be modeled on supercomputers. According to Tichenor, established HPC users like GE, Boeing, and Pratt & Whitney want to move toward full system simulations. But, she says, if they can’t import the computing models from their parts suppliers, they’re not going to be able to achieve that.
It’s quite likely that most modeling and prototyping is still performed on the desktop — maybe not in total HPC cycles, but almost certainly in number of individual application seats. “You have a situation here where you have a very small group of high-end users and a very large group of desktop and entry-level users, and not very many in the middle,” explains Tichenor. “What you really want to do is move everyone forward.”
The Council thinks an “enabling function” is going to be required to get most of these companies over the barriers. That might take the form of a public-private partnership, where universities or government labs get involved with commercial organizations to provide resources and guide them through some of the pitfalls of high performance computing. The DOE’s INCITE program, which first offered national lab HPC resources to commercial users in 2005, is an example of this approach. But INCITE is for the elite supercomputing crowd. What is needed is something at the other end of the spectrum. It’s also possible that commercial firms with HPC experience will help their desktop-only partners get up to speed. Pratt & Whitney has done this to some extent with a few select suppliers. ISVs can help potential HPC users over the ROI barrier with better license pricing, access models (utility computing), and other incentives to encourage customers to scale up their applications.
As of today, the United States leads in HPC infrastructure, expertise, and software development. But according to Tichenor, now is the time to take advantage of the window of opportunity. Despite the barriers, there is global access to HPC and some of the fastest growth in this area is taking place in India and China. In general, globalization favors countries that can harness human resources most efficiently. In the case of high performance computing, this means developing a community of highly skilled engineers and scientists that can apply this technology. Considering that the U.S. has become a net importer of technological expertise, the nation’s HPC leadership is in danger.
“There’s a real opportunity to make modeling and simulation a national best practice,” says Tichenor. “I think that’s something we have to start thinking about as a country. High performance computing is really the next productivity platform.”