Organizers of Supercomputing Frontiers Europe 2018, held in Warsaw, Poland, last month, have posted brief, fascinating video interviews with three prominent keynoters – Dimitri Kusnezov, chief scientist, National Nuclear Security Administration, DoE; Thomas Sterling, professor of intelligent systems engineering and director, Center for Research in Extreme Scale Technologies, Indiana University; and Karlheinz Meier, a leader in the EU Human Brain Project and professor (chair) of experimental physics, University of Heidelberg.
The conference, now in its fourth year, was held in Singapore the first three times; it tackles all things supercomputing spanning technology, international competition, and efforts to transfer advanced technology into productive use by industry. No surprise, AI writ large, the race to exascale, and convergence of AI and traditional HPC were all part of this year’s agenda.
With apologies to those quoted, here are three teaser snippets from their video interviews intended to entice but hardly catch the full scope of their comments (their keynote topics are in parentheses):
- Kusnezov (Precision Medicine as an Accelerator for Next Generation Supercomputing) – “We have looked for ways to partner with private sector to share best practices, share our equipment, share our capabilities and tools to help in advancing their type of competitive edge in the business sectors they are in. [We’ve found] in partnering with the private sector, it’s hit or miss.”
Sterling (Simultac Fonton) – “The US, according to the [NSCI] initiative, wanted to remain the leader in the field. Now in truth, by several metrics, the US is not the leader of the field although it certainly is one of the highest [users] of high performance computing in the world….If there is a race between the U.S. and China in terms of achieving exascale, in all likelihood China has already won.”
- Meier (Neuromorphic computing: From biology to user facilities) – “There is a two-way connection between brain science and computing technology. Brain science can help develop future computing technology. We know a lot about deep learning and we think deep learning is somehow inspired by biology but currently there is very little inspiration from biology. The artificial neural networks that we use don’t have a lot to do with how the brain works.”
The interviews were wide-ranging and worth watching. Links to all three follow
1. Dimitri Kusnezov, DoE
Worried about the global exascale race? Not Kusnezov. “If it’s a race I think that’s great because it means there are many exerting significant energy to advance the field,” said Kusnezov. He spent time distinguishing the DoE’s distinct needs from most HPC users’ communities, also discussed why he thinks HPC in the cloud had yet to gain substantial traction, and talked about DoE desire to influence vendor technology roadmaps. While successful collaboration with DoE supercomputer facilities to foster HPC adoption by industry remains difficult now, Kusnezov singled out oil and gas and tire manufacturers as success stories and expects more to come.
2. Thomas Sterling, Indiana University and CREST
Sterling was informative and entertaining as always. “Simply a prediction on my part. I believe many will disagree [but] I believe intelligent machines that embody principle intelligence – we don’t have these yet – will ultimately go back to the early schools of symbolic computing rather than statistical, probability computing or the training-session-based computing, and will ultimately consume the vast majority of cycles in computing whether it’s small computers or supercomputers larger than can be imagined. So I believe that intelligent computing will dominate computing within our lifetime.”
3. Karlheinz Meier, Human Brain Project and University of Heidelberg
Meier made clear how little we know about how the brains works and how effective neuromorphic computing that more accurately mimics brain function and organization could be game-changing. Duplicating the brain’s efficiency use of energy – 20 watts versus megawatt – has long been an attractive goal and would be a game changer in mobile devices. Fault tolerance is another critical attribute. “We lose about one brain cell per second, 100 thousand per day, and still we kind of function reasonably well. If you would kill a transistor in a processor per second, it would stop processing immediately.”