Tech Talk: Powering the Future with HPC

When you want to learn about something, you want to learn from one of the best in the field. But who wants one of the best when you can learn from two of the best? Two of the most forward minds in the world of High Performance Computing (HPC), and members of Samsung’s SAIT Systems Architecture Lab, joined together to talk about the current power and future potential of HPC. Samsung’s Senior Vice President and Chief Architect of HPC Dr. Robert W. Wisniewski and Vice President and Chief Hardware Architect of HPC Dr. Samantika Sury joined Dr. Sarah Peach in an HPC-everything Tech Talk. Who wouldn’t want to listen?↗ Or at least read. This has everything you’d want to know, all in one digestible recap. The History and State of HPC Since its beginnings, HPC has always been the problem solver in the world of technology. Difficult questions, and difficult systems, required HPC to find the solution. Originally, this consisted primarily of modeling and simulation applications. It allowed people to test out hypotheses on how things would behave under certain circumstances, and eventually draw conclusions. Scientists and engineers have long used HPC to study everything from the collision of galaxies to sub-atomic particles, and to better design airplane wings and car’s combustion engines. It even affected our everyday lives --- Johnson and Johnson used HPC to model dropping of shampoo bottles so they are less likely to break. Thus, modeling and simulation HPC gave way not only to research findings, but to real-life, tangible products that people use daily. HPC, and its applications, has rapidly changed over the past five years. Increasingly, HPC is being used in combination with Artificial Intelligence (AI) and Deep Learning. These are concepts and topics that sit at the front of the tech world, bringing in interest and demand from all sides. Resulting from the rapid growth of HPC’s capabilities infused with AI is an invigorated interest in what HPC can truly accomplish. On its current trajectory, there’s no telling what HPC could enable in the future. The Future of HPC Actually, let us backtrack. Perhaps we misspoke just a bit – there is some idea of what HPC can accomplish in the future. In the Tech Talk, our Samsung experts hypothesized what HPC will be able to do 5, 10, 15, even 20 years down the line. Their predictions might surprise you! But probably not. Actually, the base answer may sound a bit disappointing – HPC will do the same things it does now, just better, faster, and more efficiently. But the truly fascinating part isn’t what the HPC machines themselves can do, but in the applications that are enabled by that uber-improved efficiency. Faster AI and problem solving unlocks the potential for technology applications that we might even struggle to imagine now. As HPC evolves, the technologies it empowers will evolve with it, giving way to new applications and user experiences. While all this advancement is happening, HPC will continue to become more cost-efficient as well, allowing HPC technology to become even more ubiquitous than it already is. This lower cost to entry will allow greater numbers businesses to truly leverage the power of HPC. But there are many challenges to getting there. Some of the answers will come from what our Samsung leaders are calling memory coupled compute (MCC). Memory and compute are typically two functions that are considered separate, and closing the gap between them will bolster the ability of HPC. The communication currently needed between the two creates latency and reduces energy efficiency, an important aspect of future HPC systems. Think about being given a destination address and directions to get there by someone else versus being able to get there yourself. The closer the memory is to the person or thing performing the actions, the more effective and efficient a process becomes. MCC bridges the gap, moving memory and the compute even closer together, thus improving performance while also reducing system cost and lowering the energy consumed. Samsung’s Forward Perspective To accomplish all that, Samsung needs to look forward. That type of ambition requires more than just a vision – it requires looking at things from a holistic perspective. So, of course, Samsung has its own approach to advancing the field. The Systems Architecture Lab is looking at it from a systems perspective, rather than a component-level perspective. But how is that different? Aren’t components just what makes up a system, which means if components are maximized then the system will be too, right? Not quite. It’s like thinking about a fleet of cars rather than an individual car. With an individual car, you’re just worried about its performance, speed, and efficiency, but when you throw that car into a fleet that is working to get to the same destination together, things change. One singular car can’t go too fast if the others lag behind, and it might have to sacrifice its own speed for the sake of improving the overall efficiency of the group. Additionally, the communication and coordination between the cars become paramount to the fleet’s success, placing emphasis on how the cars work together, not which has the greatest horsepower or mileage. We at Samsung want to get the entire fleet there faster, not just one car. These kinds of tradeoffs from a systems perspective can result in a more optimal, complete HPC system, even if it seems counterintuitive from a purely component-level approach. With a systems approach, while working towards MCC, Samsung hopes to power the HPC of tomorrow; and with it, unlock a whole new range of applications and solutions to better our world. It’s just a matter of time.