Intel® HPC Developer Conference 2017

Systems Sessions

Containers for High-Performance Computing

The presentation describes how to use Docker* and Singularity* containers on Intel® Xeon Phi™ processors with Intel® Omni-Path Architecture, as well as integration with the Intel® MPI Library. Compare performance results obtained using containers with bare metal results.

Piotr Uminski, Intel

Presentation (PDF)




Administration of QoS for Lustre*

Administration of QoS for Lustre*
(23 min)

We present our QoS Planner for Lustre* and introduce how we faced different requirements of HPC administrators. Besides usability and its integration into existing batch systems, the performance, reliability, and flexibility of the QoS planner opened many program and design challenges.

Tim Suess, Johannes Gutenberg University

Presentation (PDF)


Designing an Optimal Deep Learning Solution at Dell Technologies*

Designing an Optimal Deep Learning Solution at Dell Technologies*
(18 min)

There are a multitude of factors that go into successfully implementing a deep learning solution. These encompass hardware, software, data ingestion and processing, choosing a model, and so on. This talk outlines our design criteria, the challenges faced, and our resolutions while architecting an optimal deep learning platform using Intel technologies.

Nishanth Dandapanthula, Dell Technologies*

Presentation (PDF)


Hyperacceleration of Apache Spark* with Intel® Arria® 10 FPGAs

Hyperacceleration of Apache Spark* with Intel® Arria® 10 FPGAs 
 

Today’s big data and machine learning workloads are already being throttled by tomorrow’s platforms. Acceleration is the new key to enabling these workloads. Bigstream has developed a hyperacceleration platform that automatically accelerates these workloads on CPUs, GPUs, and FPGAs without changing application code.

Balavinayagam Samynathan, Bigstream

Presentation (Interactive PPT)


The Benefits of Intel® Memory Drive Technology for Scientific Applications

This session focuses on the best practices for using Intel® Memory Drive Technology in a diverse set of HPC codes, demonstrating strong and weak points of hybrid memory approach. Become familiar with hybrid memory approach using DRAM and 3D XPoint™ Memory Media based on Intel Optane™ technology.

Vladimir Mironov, Lomonosov Moscow State University
Yuri Alexeev, Argonne National Laboratory, Leadership Computing Facility
Andrey Kudryavtsev, Intel
Igor Chernykh, Institute of Computational Mathematics and Mathematical Geophysics SB RAS

Presentation (PDF)






Hybrid HPC and HPC as a Service

Infrastructure as a service (IaaS) has been playing a prominent role in enterprise IT for some time. It enables the automation and self-service capabilities many organizations are looking for in their IT services in order to streamline operations and increase efficiency. This session discusses the collaboration between UberCloud*, Hewlett Packard Enterprise (HPE) and Intel with the Living Heart Project.

Wolfgang Gentzsch and Burak Yenier, UberCloud*

Presentation (PDF)