Description: Learn how to distribute PyTorch machine learning tasks across multiple GPUs on multiple nodes to significantly increase data throughput. This seminar will cover everything needed to convert a standard PyTorch model into a fully GPU-parallel model, and to execute these scripts on HPCMP machines. Additional optimizations for decreasing redundant setup, increasing device utilization, and managing memory usage will also be discussed.

CUI

Description: Cray Performance Analysis Tools (CrayPat) is a performance analysis tool offered by HPE-Cray. This seminar will present the basics for using perftools-lite. Perftools-lite tool is a simplified and easy-to-use version of CrayPat that provides basic performance analysis information automatically, with minimum user interaction. Approaches to using this tool to generate reports useful in detecting program hot spots, identifying load imbalances, and measure memory bandwidth will be discussed. After taking this course, the user will gain a general understanding how to use perftools-lite tools to fine tune their applications

Description: Newer HPCs are much more than higher GHz, have more cores, and more memory. The HPCMP now deploys a diverse set of HPCs. Selecting the best one for a given application requires knowing what is available and how they compare. This training collects this information and guides users in selecting an HPC that fits their applications based on HPC hardware and software.

Presenter(s): Dr. Robert Donnelly, GDIT / PET
Location: Webcast
Date & Time: March 15, 2022, 2:00p - 3:30p ET

Distribution Statement D. Distribution authorized to the Department of Defense and U.S. DoD contractors only, Administrative or Operational Use, 15 March 2022.  Other requests for this document shall be referred to the High Performance Computing Modernization Office, 3909 Halls Ferry Road, Vicksburg, MS 39180.

Description: This tutorial covers the numerous opportunities that exist in effectively using the Cray XE/XT system on scientific applications.

Presenters: Jeff Larkin and Alan Minga, CRAY, Inc.,

Date/Location:  July 18, 2012, ERDC

Description: Derived datatypes let you describe non-contiguous or structured memory layouts so MPI can send them directly, which reduces data movement overhead. Virtual topologies let you create a processor mapping that gives structure to a communicator effectively reducing the algorithmic and network overhead of figuring out who talks to whom. Together they make MPI coded faster, cleaner, and more scalable.

The aims in this seminar are two-fold:

1. Discuss and demonstrate how to develop derived types to describe data layouts to automate the packing/unpacking of non-continuous data or nonuniform data types defined by data structures.

2. Discuss and demonstrate to develop virtual topologies that better align with an application’s data communication pattern to simplify and optimize the communication logic between processes.

Description: This class covers how to effectively improve the performance of I/O on Cray systems running the Lustre file system.

Presenters: Jeff Larkin and Alan Minga, CRAY, Inc. 

Date/Location: July 19, 2012, ERDC

Distribution Statement D. Distribution limited to the Department of Defense and U.S. DoD contractors only, for administrative or operational use, 19 July 2012.  Other requests for this document shall be referred to the High Performance Modernization Office, 10501 Furnace Road, Suite 101, Lorton, VA 22079.

Description: Presentations by Intel Technical Consulting Engineers covering Intel's optimizing compiler, VTune Amplifier XE analysis and workflow, vectorization optimization and thread prototyping, tuning MPI application performance, Trace Analyzer and Collector, and data analytics acceleration library (Intel DAAL). 

Presenters: Martyn Corden, Munara Tolubaeva, Gergana Slavova, Shaojuan Zhu, Intel 

Date: August 23, 2016 

Distribution Statement D. Distribution limited to the Department of Defense and U.S. DoD contractors only, for administrative or operational use, 23 August 2016.  Other requests for this document shall be referred to the High Performance Modernization Office, 10501 Furnace Road, Suite 101, Lorton, VA 22079.

Description: To run jobs on an HPC, you use the batch or queue system. PBS is used on most HPCMP HPCs. To use an HPC effectively, you need to use PBS effectively. Doing so involves knowing the basics but also how to specify the hardware you need. There are also different types of queues, including special ones for interactive use and copying files. PBS features like job dependencies and job arrays help you implement a multi-step workflow and execute a large set of tasks. After submitting jobs, there are various tools to track and further control jobs. You can craft job parameters to maximize throughput and reduce queue wait time. If you have urgent compute needs, you should know the process to request time in priority queues and/or use the reservation system. This training provides the skills to not just use HPCs but to optimize your HPC use.

Presenter(s): Dr. Robert Donnelly, GDIT/PET
Location: Webcast
Date & Time: August 31, 2021, 2:00p - 3:00p ET

Distribution Statement D. Distribution authorized to the Department of Defense and U.S. DoD contractors only, Administrative or Operational Use, 31 August 2021.  Other requests for this document shall be referred to the High Performance Computing Modernization Office, 3909 Halls Ferry Road, Vicksburg, MS 39180.

Description: CrayPat is a performance analysis tool offered by Cray. CrayPat has a large feature set and two versions: perftools and perftools-lite. perftools-lite is a simplified and easy-to-use version of the CrayPat tool. This seminar will present the basics for using the Cray performance analysis tools: perftools and perftools-lite. Approaches to optimization using CrayPat commands and options will also be discussed. The user should gain a general understanding how to use CrayPat tools to fine tune their applications.

CUI

Distribution Statement D. Distribution authorized to the Department of Defense and U.S. DoD contractors only, Administrative or Operational Use, 29 January 2019. Other requests for this document shall be referred to the High Performance Computing Modernization Office, 3909 Halls Ferry Road, Vicksburg, MS 39180.

Distribution Statement D. Distribution authorized to the Department of Defense and U.S. DoD contractors only, Administrative or Operational Use, 5 February 2019.  Other requests for this document shall be referred to the High Performance Computing Modernization Office, 3909 Halls Ferry Road, Vicksburg, MS 39180.

Description: Mathematical optimization is a critical aspect of DoD research and engineering workflows. Large scale optimization efforts with expensive modeling, simulation and machine learning (ML) tasks can benefit from scalable optimization algorithms. Optuna is a Python library that supports distributed optimization and hyperparameter tuning that can scale across large HPC systems. Global multi-objective and constrained optimization studies can be created with ease. Studies can be monitored in real-time, and results can be visualized immediately using Optuna’s integrated dashboard. In this Webinar, users will learn how to use Optuna at scale for general optimization and ML hyperparameter tuning on HPCMP systems.

Description: The seminar has a preliminary presentation that covers where to find the Sierra Suite on ARL HPCMP systems, how to get access to the ABSierra group, and tools such as the Sandia Analysis Workbench (SAW), Catalyst On-the-Fly visualization, and Zapotec which can be used in conjunction with the Sierra Suite. The main presentation will provide an overview of Sandia’s Sierra mechanics codes including context about Sandia National Laboratories’ mission drivers and partners that use the codes and related codes at the laboratories. Codes include Structural Dynamics, Solid Mechanics, and Thermal Fluids. We will also cover Sandia Analyst Workbench, Inverse Optimization, and Plato (Topology Optimization).

Presenters: Dr. Zachary Wilson (PETTT Engility), Michael Skroch and Lynn Munday (Sandia National Laboratories)

Date/Location: August 22, 2018, ARL

Distribution Statement D. Distribution limited to the Department of Defense and U.S. DoD contractors only, for administrative or operational use, 22 August 2018.  Other requests for this document shall be referred to the High Performance Computing Modernization Office, 3909 Halls Ferry Road, Vicksburg, MS 39180.

Description: This Sierra fundamentals training tutorial provides basic instruction needed to perform solid mechanics, structural dynamics, and thermal mechanical simulations using the Sierra code suite, including explanation of Zapotec and Inverse Optimization.

Presenters: Trevor Hensley, Timothy Shelton, Nathan Crane, Sam Subia, Justin Lamb (Sandia National Laboratories) and Dr. Zachary Wilson (PETTT Engility)

Date/Location: December 3-7, 2018, Doolittle Institute, Eglin AFB, FL

Distribution Statement D. Distribution limited to the Department of Defense and U.S. DoD contractors only, for administrative or operational use, 7 December 2018.  Other requests for this document shall be referred to the High Performance Computing Modernization Office, 3909 Halls Ferry Road, Vicksburg, MS 39180.

Description: Introduction to Singularity Compose for HPC environments. This seminar will cover containerized workflows with Singularity, including setting up compose.yaml, deploying multi-container workloads, managing resource allocation (CPU cores per task), and limitations of the Singularity Compose in relation to Docker Compose. A detailed example is provided showing a generation container creating CSV data and a processing container computing interpolated topographic maps running concurrently on a single node.

Description: Slurm is a batch queue system like PBS and is used on several new HPCs like Jean and Nautilus. This training covers basic use, node selection, and job status and control. Learn how to optimize your use of these new HPCs.

CUI

Description: NEC Vector Annealing is a unique implementation for simulated quantum annealing on NEC’s Vector Engine hardware. This solution can be used to efficiently narrow down the search range of solutions for combinatorial optimization problems by utilizing various constraints imposed in practical situations. The NEC Vector Annealing Service performs matrix calculations with a large capacity memory, ultra-high-speed processing of large-scale combination problems on the vector processor.

This session has been designed to help users understand the concept of quantum annealing, introduce them to the solution stack of NEC Vector annealing, and showcase programming samples to solve combinatorial optimization problems

CUI