Stephen S. Pawlowski knows a thing or two about Moore's Law. He joined Intel, the company co-founded by Gordon Moore, in 1982, and has seen the number of transistors its microprocessors hold double about 20 times since then—a factor of nearly three million.
He has a keen sense of where these processors fit into the HPC. niche He's an Intel Senior Fellow, chief technology officer for the Datacenter and Connected Systems Group (DCSG), and general manager for the Intel Architecture Group and DCSG Pathfinding at Intel. He oversees architectural consistency across all of Intel's architectures, and implements initiatives such as security and manageability across Intel Core and Intel Atom product lines.
Pawloski also been the director of the Corporate Technology Group's Microprocessor Technology Lab, the same kind of position Moore had when he made his famous observation.
At the International Supercomputing Conference (ISC'13
) on June 18, he will give a keynote speech titled Moore's Law 2020
. It points out why Moore's Law has been so influential in the past and why the new era of "Moore Squared" will shoot HPC into the exascale era and beyond.
In anticipation of that talk, HPCwire asked him a few questions about the future of Moore's Law and the inexorable move toward exascale computing.
HPCwire: What new software, hardware and integration features or products are needed to successfully reach exascale computing?
Stephen Pawlowski: The breakthroughs are in resiliency, scalability, programmability and power efficiency: How to make systems robust against real time errors, how to have high performance interconnect that makes scaling to the dimensionality necessary for exascale tractable, and how to do all that is required at a fraction of the performance per watt we see today. These are the key questions.
I don’t think these are new products, nor new features per se. They are improvements in the way existing products do their tasks. This in itself is a great source of innovation.
HPCwire: Sometimes it seems that exascale is a distant reality that keeps receding in to the future. How soon do you think we'll get there and why?
Pawlowski: There will certainly be real world applications on exascale systems in this decade. Perhaps sooner. Perhaps it's destiny. Burgelman wrote that “Strategy is Destiny,” and if that’s so, then many nations have strategies to remain ahead in the computation capability game.
But I also think that Economics is Destiny. The economic competitive imperative will drive industry to thirst for exascale capability, if not by the end of the decade, then very very soon there after.
In 2024, there will be many dozens of exascale resources serving academic and commercial interests, and that will be just the tip of the iceberg. The return on investment in computing looks very strong indeed for many years to come. That, as much as strategy, will cause the challenges at the heart of exascale to be brought to solution.
HPCwire: What significant trends do you predict will emerge between ISC'13 and SC'13?
Pawlowski: In the next five months? The industry doesn’t change that fast! The major inflections that are coming have been signaled by the participants for years. I don’t think we’ll see any real major inflections in that time, but we will see those already-signaled changes start to become real.
Specifically, we will see an increased interest in data analytics and other technologies that will make supercomputing computer science a more practical application for more engineers and scientists. This will reinvigorate the technology of storage and networking systems, both in hardware and software.
Certainly the funding agencies are signaling this by the merging of Big Data and Exascale initiatives. HPC and Big Data live together to generate return on investment. Computation without data is just telling us what we already should have known. Data without computation is search. Interesting, but it only touches the tip of the iceberg of insight.
HPCwire: What can you tell us about your keynote?
Pawlowski: Moore’s law has been the heartbeat of the electronics industry for over 30 years. In the technical computing industry, Moore’s law plays a part in enabling the largest supercomputers on the planet.
But real applications require a balanced system, which requires applying the technology of Moore’s law to not only computation, but to managing the complexity of the system, the storage and the networking technology that comprise it.
It’s the combination of these two Moore’s law functions that enables a growth rate in supercomputing that's faster than Moore's law and gives us confidence in achieving our objectives ahead. This combination also drives innovation into the hands of hundreds of thousands of engineers and scientists who use much smaller systems, but whose work is just as economically significant.
HPCwire: Any other comments you'd like to add?
Pawlowski: I'm looking forward to coming to ISC again. It has been a while since I've attended this important worldwide gathering. I look forward to the exchange of ideas that will help grow the industry.