insideHPC: During the upcoming session on Great Success Stories of HPC, you’ll be featuring two case studies: one describing 3-D simulations of supernova explosions, and the other modeling aircraft behavior. What are some of the other application domains where HPC is pushing the envelope for researchers and engineers?
Frank Baetke: HPC is pushing the envelope basically in all areas where complex time-dependent phenomena need to be simulated. Why is that? First, because complex phenomena cannot be described by simple equations that can be solved analytically so that we end up with numerical methods. Second, because those problems are multi-dimensional (3D plus time) leading to fine computational grids that exceed the capacity of small systems often by far.
In terms of what HPC makes possible, imagine a detailed analysis of a flow through a biologic system – for example a pumping heart with attached arteries. Or imagine simulating the circuitry of a brain. A completely different area is finding and exploiting natural resources while minimizing the ecological impact.
insideHPC: To date most of more highly publicized HPC success stories have involved physics-based simulations – the basis of traditional high performance computing applications. Are there other types of applications that are getting less attention?
Frank Baetke: Of course there are other areas like we are trying to model such as the behavior of societies or financial communities for example.
insideHPC: The global pursuit of exaflop supercomputers is a focus of a large segment of the HPC community. Do you think HPC success stories will be tied to the continued advancement of floating point performance or are other capabilities becoming more important to users now?
Frank Baetke: It is naïve to think that “Exaflops” will provide a sudden breakthrough in simulation capabilities in areas where floating point power prevails. With every three orders of magnitude of performance increase, we are hitting a “magic name” – once it was Teraflops, then Petaflops, now we see Exaflops at the horizon and it is obvious that for certain complex phenomena we may need a Zetaflop-system.
But for certain problems, as described above, we may not need floating operations at all but rather integer or logical operations. We are already seeing systems with very specific, application-oriented instruction sets. The HP “Moonshot” architecture is a typical example of such a system. Also note that the future of medicine will be based on personalized genetic data, leading to an immense increase in computational needs.
insideHPC: Is the global high performance computing community – the vendors and users – making its case for government investment in HPC applications and technologies? For example, do you feel government investment in HPC in Europe is adequate?
Frank Baetke: Government investment is taxpayers’ investment and the taxpayers should always ask, “What is the return of my investment?” In the case of HPC systems, I think the return for the society compared to the investments is very high. The technology behind HPC is global and HPC centers providing resources should have access to the best available technology when looking for new systems. Ultimately the key resource will be the skills of the scientists and the researchers. Europe could make access to large HPC resources more attractive but also emphasize the importance of HPC for the society and the economy so that students are more compelled to focus on HPC-related curricula at universities. Unfortunately, there is also a history of subsidized technologies that never became really competitive on a global scale. That money could have been spent better.
insideHPC: How about the vendors themselves? Many companies appear to be devoting less attention to their traditional HPC business, drawn by the faster growth of the big data and cloud markets, which can employ many of the same technologies. At the same time, a number of HPC vendors have been swallowed up by larger companies, many of which have shown little interest in HPC. Do you think this changing HPC vendor landscape is of concern to users?
Frank Baetke: No, not at all! In the past years HPC became a mainstream technology where global and regional vendors have to compete for public and private customers. As the overall usage of HPC is growing there will always be companies with an HPC-focused product portfolio. And if one leaves the arena, another one will step in and try to play. The best a government can do is to enable competition – this will also allow true innovators to find customers as long as the rules are fair and clearly defined.
Note that I have answered the above question as a private person in my role as a session chair at ISC. The opinions above do not necessarily reflect the official position of HP.
Registration is now open for ISC High Performance, which takes place July 12-16 in Frankfurt, Germany.