An Instrumentation Request for Upgrading Euler, a CPU-GPU Heterogeneous Cluster Supporting Research and Educational Activities in Multi-Physics Modeling and Simulation
Abstract
This instrumentation request seeks funding to upgrade an existing heterogeneous computing platform called Euler. Euler is a mosaic of architectures and computational paradigms combining fine- and coarse-grain parallel computing, as supported by GPU cards (NVIDIA), multi- & many-core CPU chips (Intel Xeon Phi, Intel Xeon and AMD Bulldozer), and hybrid CPU/GPU solutions (AMD Kaveri, NVIDIA Jetson). In staying faithful to the idea of using Euler as an enabler of Òresearch in computingÓ asset, we propose to use the grant funds to update Euler with (i) 12 new nodes, each with one Intel Xeon Phi Knights Landing bootable chip; and, (ii) a new file server, to replace the under provisioned/obsolete server in use today. In justifying (i), we point out that the Knights Landing chip represents a revolutionary design in several respects: it is an x86 chip with very high bandwidth owing to the presence of a 16 GB of on-package high bandwidth memory; it has outstanding support for vectorization; and, it supports a mechanism to expose the chip for programming in a variety of ways via a logical partitioning of the 72 cores in several topologies, each with its own set of advantages and disadvantages. Insofar (ii) is concerned, the current file server draws on hardware purchased in 2012. Euler supports a research program that represents the collective effort of one faculty, one senior researcher, two visiting scientists, four graduate students, and eight undergraduate students. Euler is the very cornerstone of this teamÕs research effort whose mission is to advance the state of the art in Computational Dynamics. To that end, we operate at the intersection of Mechanical Engineering Ð for handling applications in ground vehicle mobility, terramechanics, granular dynamics, robotics; Applied Mathematics Ð for developing numerical methods for sparse linear algebra, large scale optimization, differential algebraic equations and differential variational inequalities; and Computer Science Ð for efficiently leveraging GPU, multi/manycore, and heterogeneous computing. Euler is a hardware asset with which we challenge the conventional wisdom in computational engineering by investigating approaches that look beyond MooreÕs Law. We seek breakthroughs that capitalize on heterogeneous computing and the use of scalable numerical methods to increase solution accuracy, problem size, and simulation speed. From an educational perspective, Euler is an asset that has been used over the last five years by hundreds of students from more than 10 departments at University of Wisconsin-Madison to further their education in advanced computing. These are students who are taking the graduate class ME/CS/ECE/EP-759: High Performance Computing for Engineering Applications. For the first time, in 2017 the class will also be offered simultaneously online to practicing engineers from companies such as Lockheed Martin, IBM, Intel, Johnson Controls, and John Deere. Euler has also been used to promote the adoption of advanced parallel computing via one-day workshops at ASME conferences (2009 through 2016), and open collaboration avenues with international colleagues (Darmstadt University, Politecnico di Milano, Polytechnic Institute of Bucharest). Finally, Euler comes into play in all our outreach and technology transfer initiatives. In Madison and at the US Army TARDEC location in Warren, we have trained specialists from US Army TARDEC and ERDC in the use of the in-house developed open source software Chrono, which has been designed to leverage extensively the computational paradigms exposed by Euler. Lastly, since 2013, we have organized an event tied to the topic of advanced computing and vehicle mobility (Machine-Ground Interaction Consortium), in which we have engaged more than 200 colleagues from industry.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 14, 2019
- Source ID
- W911NF1810476
Entities
People
- Dan Negrut
Organizations
- Army Contracting Command
- United States Army
- University of Wisconsin–Madison