DescriptionDesigning and building future leadership computer systems is challenging. However, realizing the performance potential of these systems rests almost solely on obtaining additional concurrency from software. Algorithmic innovations and creation of new libraries and tools are essential to obtaining the performance improvements that these platforms promise.
This panel is composed of a diverse group of experts who are leading the creation of new algorithm and software capabilities that will make exascale performance real. Questions addressed by this panel include, How will exascale systems be usable and useful? Can new problem formulations overcome traditional scaling bottlenecks? Can algorithms exploit low precision hardware that is an order of magnitude faster and still obtain robust solutions? Can we raise abstraction layers so applications can leverage higher level common capabilities? What role will tools play in achieving scalability? Can data and visualization workflows adapt to exploit new system capabilities and overcome limitations?