Loading…
Simple
Expanded
Grid
By Venue
BioScience Research Collaborative (BRC) Building [clear filter]
Monday, March 2
 

8:15am PST

Workshop A: Analyzing CPU and GPU Application performance with HPCToolkit
Breakfast 8:15-8:45 AM
Workshop A 8:45-10:15 AM
Break 10:15-10:30 AM
Workshop B 10:30-12:00 PM

Tailoring applications for GPU-accelerated compute nodes is essential to harness the power of current and forthcoming GPU-accelerated platforms. Application developers need effective tools for this purpose.

While NVIDIA and others provide tools for performance measurement and analysis, they fall short in providing what application developers need, especially for programs developed using OpenMP offloading and template-based programming models for GPUs.

This tutorial will (1) introduce new capabilities for performance measurement and analysis of GPU-accelerated codes that are emerging Rice University’s HPCToolkit performance tools and (2) describe how to use them to analyze and tune GPU-accelerated applications.

To support efficient monitoring of accelerated computations, HPCToolkit employs a novel wait-free data structures to coordinate measurement and attribution of performance metrics while a GPU-accelerated computation executes. To help developers understand the performance of accelerated applications as a whole, HPCToolkit tool attributes metrics to rich heterogeneous calling contexts that span both CPUs and GPUs and displays traces that include both CPU and GPU activity time lines. To help developers understand the performance of complex GPU code generated from high-level programming models such as OpenMP or template-based programming abstractions, HPCToolkit constructs sophisticated approximations of call path profiles for GPU computations from flat PC samples collected by NVIDIA GPUs. To support fine-grain analysis and tuning, HPCToolkit uses GPU PC samples to derive and attribute metrics, including measures of GPU latency and throughput at all levels in a heterogeneous calling context.
To make effective use of HPCToolkit for tuning GPU-accelerated applications, one must understand what performance metrics that HPCToolkit can collect, and how to use them to guide analysis and tuning.

This tutorial will show how to use HPCToolkit to measure and analyze the performance of GPU-accelerated programs. We will illustrate the capabilities of HPCToolkit with case studies of various codes and mini-applications.
Speakers
JM

John Mellor-Crummey

Professor of Computer Science and of Electrical and Computer Engineering, Rice University

Monday March 2, 2020 8:15am - 12:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Workshop

8:15am PST

Workshop B: WHPC Roundtable: Growing a Diverse Talent Pool in HPC for Oil & Gas
Breakfast 8:15-8:45 AM
Workshop A 8:45-10:15 AM
Break 10:15-10:30 AM
Workshop B 10:30-12:00 PM

This session, organized by Texas Women in HPC (TXWHPC), will be a roundtable discussion of the challenges of growing and retaining a diverse talent pool in the oil & gas industry. Experts from oil & gas and computing industries, academia, and the national labs will discuss the issues of diversity, inclusion, and retention, share success stories, and offer suggestions for real solutions. Realistically increasing diversity in the workplace requires careful planning and support at all levels.
Speakers
TW

Monday March 2, 2020 8:15am - 12:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Workshop

8:15am PST

Workshop C: Best Practices in Supercomputing Systems Management
Breakfast 8:15-8:45 AM
Workshop C 8:45-10:15 AM
Break 10:15-10:30 AM
Workshop C 10:30-12:00 PM
Speakers
PA

Monday March 2, 2020 8:15am - 12:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Workshop

12:00pm PST

Registration & Networking
Monday March 2, 2020 12:00pm - 1:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Networking

1:00pm PST

Welcome
WATCH VIDEO
Speakers
LK

Lydia Kavraki

Director of the Ken Kennedy Institute, Rice University
JO

Jan Odegard

Executive Director of the Ken Kennedy Institute, Rice University

Monday March 2, 2020 1:00pm - 1:15pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030

1:15pm PST

The Deepwater Gulf of Mexico – A Success Story of High Performance Computing
The deepwater Gulf of Mexico is one of the most prolific and valuable oil production basins in the world. After over 25 years of production, investment levels remain high (and rising) and new frontiers in the basin continue to be pushed. Improvements in capital efficiency, operational safety and deepwater engineering have all been key to the growth story, but no single technology has had a bigger impact on the basin than Seismic, and Seismic processing and analysis is in turn driven by large scale compute. This talk will provide some insights from BP on how advances in computing have unlocked value in the basin and are expected to drive further improvements in the Gulf of Mexico for decades to come.
Speakers
EE

Emeka Emembolu

VP Reservoir Development, Gulf of Mexico & Canada, BP

Monday March 2, 2020 1:15pm - 2:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Keynote

2:00pm PST

Scaling TCO in a Post-Moore’s Law Era
WATCH VIDEO

While foundries bravely drive forward to overcome the technical and economic challenges posed by scaling to 5nm and beyond, Moore’s law alone can provide only a fraction of the performance / watt and performance / dollar gains needed to satisfy the demands of today’s high performance computing and artificial intelligence applications. To close the gap, multiple strategies are required. First, new levels of innovation and design efficiency will supplement technology gains to continue to deliver meaningful improvements in SoC performance. Second, heterogenous compute architectures will create x-factor increases of performance efficiency for the most critical applications. Finally, open software frameworks, APIs, and toolsets will enable broad ecosystems of application level innovation.
Speakers
BM

Bradley McCredie

Corporate VP, GPU Platforms, AMD

3. Final AMD McCredie Rice Oil Gas HPC Keynote pdf
Monday March 2, 2020 2:00pm - 2:30pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Plenary

2:30pm PST

Using Deep Learning to Accelerate Computational Fluid Dynamics
WATCH VIDEO

Numerical methods and high-performance computing have rapidly advanced in the past few decades, enabling accurate simulations of complex thermo-fluid systems in many engineering problems. However, faster and higher resolution simulations, particularly for design, optimization, and online control, are of increasing demand, requiring novel approaches to simulating multi-scale, multi-physics, thermo-fluid systems. In the past few years, a number of studies have explored how deep learning might accelerate computational fluid dynamics (CFD). In this talk, I will discuss some of the recent advances in this area, and in particular will present our group’s work on 1) using recurrent neural networks for data-driven integration of some of the computationally demanding equations in a CFD solver, and 2) using transfer learning to improve generalization (i.e., extrapolation) of a deep learning-based CFD solver. Examples from a multi-scale Lorenz system and plans for future applications to pipe flows and Rayleigh-Benard convection will be discussed.
Speakers
PH

Pedram Hassanzadeh

Assistant Professor of Mechanical Engineering and of Earth, Environmental and Planetary Sciences, Rice University

OnG Pedram pdf
Monday March 2, 2020 2:30pm - 3:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Plenary

3:00pm PST

Break
Monday March 2, 2020 3:00pm - 3:30pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Networking

3:30pm PST

Open Subsurface Data Universe (OSDU) HPC Project
WATCH VIDEO

The Open Group Open Subsurface Data Universe™ Forum is developing a standard data platform for the oil and gas industry, which will reduce silos and put data at the center of the subsurface community.
The OSDU data platform will: 
• Enable secure, reliable, global, and performant access to all subsurface and wells data 
• Reduce current data silos to enable transformational workflows 
• Accelerate the deployment of emerging digital solutions for better decision-making 
• Create an open, standards-based ecosystem that drives innovation 
This will revolutionize the industry’s ability to deliver new capabilities and reduce implementation and lifecycle costs across the subsurface community. 

The OSDU HPC project was kicked off in December 2019 by the OSDU EA subcommittee. 

The OSDU HPC project will focus on defining best practices and reference computing infrastructures optimized to support existing and emerging OSDU use cases in Azure GCP and AWS. The OSDU HPC project will focus on several main areas including: 
The definition of an optimized OSDU HPC environment to support the broad spectrum of applications and workloads of both existing and emerging uses cases 
• The existing use cases include seismic processing, basin modelling, seismic inversion, and reservoir simulation 
• The emerging use cases include AI (Machine Learning & Deep Learning – for both Training & Inference), intrinsically integrated HPC+AI, and large-scale data analytics (DA).
Speakers
JC
CR

OSDU HPC RICE pdf
Monday March 2, 2020 3:30pm - 3:50pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session A: HPC Standards & Applications & Cloud Usage

3:30pm PST

Solving Wave Equations and Seismic Inversion Using Deep Learning Platform
WATCH VIDEO

Accurate simulation of wave motion for the modeling and inversion of seismic wave propagation is a classical high-performance computing (HPC) application using the finite difference or finite element methods to solve the wave equations numerically. In this paper, instead of using traditional HPC software, we implement the solutions by means of recently developed tensor processing capabilities widely used in the deep learning software platform. The physical solutions are structured using a deep learning recurrent neural network (RNN) framework provided by the PyTorch deep learning package. Afterward, the team evaluates the RNN-structured physical solution's performance and accuracy on a GPU cluster. We also use the automatic differentiation functionality provided by PyTorch to solve the inverse problem of the wave equation.
Speakers
TC

Ted Clee

Prairie View A&M University
LH

Lei Huang

Prairie View A&M University

Lei Huang pdf
Monday March 2, 2020 3:30pm - 3:50pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session B: Data Science Technology & Applications

3:50pm PST

The Art of Serving HPC Products to the Business While They Are Still Hot
ExxonMobil has a rich history of proprietary HPC development.  Our parallel PDE solvers demonstrate record-breaking scalability and performance as featured in media and technical talks on multiple occasions.  Much less publicity is given to the streamlined and highly automated daily Development Operations (DevOps) that made such achievements possible.  

Every day the simulator development team releases multiple bugfixes and improvements, each passing the QC pipeline of hundreds of regression tests, and becoming available for downstream integration just 15 minutes after tagging.  We maintain several versions of the simulator, plus build and test executables for multiple HPC systems.  A large processing volume of operations like this cannot be sustained by a small team without investing time and effort into robust development and testing environment. 

What started a decade ago as a collection of helper scripts has blossomed into a comprehensive suite, which is a proprietary technology toolchain for automating QA/QC and deployment processes on HPC.  This technology takes the burden of routine tasks off the engineers and scientists shoulders so that they can focus on solving the problems that are simply too hard for machines: making high fidelity predictions about the behavior of complex subsurface systems.  The key success factors for this technology are conservative scope, focus on productivity and automation, and active support.  It has short list of dependencies and does not require elevated privileges.  

This talk will provide an overview of the challenges, basic principles, and core components of our DevOps technology, as well as provide a peek into the little features that make our daily work fun.
Speakers
VD

Vadim Dyadechko

ExxonMobil

Monday March 2, 2020 3:50pm - 4:10pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session A: HPC Standards & Applications & Cloud Usage

3:50pm PST

PHS: A Toolbox for Parallel Hyperparameter Search
We introduce an open source python framework named PHS - Parallel Hyperparameter Search to enable hyperparameter optimization on numerous compute instances of any arbitrary python function. This is achieved with minimal modifications inside the target function. Possible applications appear in expensive to evaluate numerical computations which strongly depend on hyperparameters such as machine learning. Bayesian optimization is chosen as a sample efficient method to propose the next query set of parameters.
Speakers
PH

Peter Habelitz

Fraunhofer
JK

Janis Keuper

Hochschule Offenburg

Keuper V2 pdf
Monday March 2, 2020 3:50pm - 4:10pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session B: Data Science Technology & Applications

4:10pm PST

Seismic Benchmark Results on Multiple Generations of CPUs
WATCH VIDEO

In this presentation we will compare the benchmarking results from the latest 2 generations of CPU hardware from AMD and Intel. The benchmarks were conducted using proprietary Seismic kernels as well as known open source benchmarks. The study focused on memory bandwidth, flops and overall performance and provides results of real-world production Seismic workloads.
Speakers
RC
AH
SR
DB

3. Shell Ron Cogswell Rice 2020 pdf
Monday March 2, 2020 4:10pm - 4:30pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session A: HPC Standards & Applications & Cloud Usage

4:10pm PST

An Attempt to Decode Reverse Time Migration with Machine Learning
WATCH VIDEO

In this presentation we will describe how to apply Machine Learning (ML) to speed up seismic imaging, essentially, we attempt to decode the PDE (partial differential equation) in our ML framework. First, we started with some randomly generated depth velocity models, and perform forward modeling to generate shot gathers. Next, we run reverse time migration (RTM) to generate depth images. Once the training data set is ready, we leverage the state-of-the-art of Machine Learning to extract features from shots and velocity model to generate seismic images. Once we have a satisfied ML model to function as a proxy for RTM, the future work could be focusing on using reinforcement learning to uplift velocity model building and seismic images.
Speakers
LZ

Licheng Zhang

University of Houston
CL

Chang-chun Lee

Rice University
YC
CZ

Cheng Zhan

University of Houston

An Attempt to Decode Reverse Time Migration Sponsor HIstory pdf
Monday March 2, 2020 4:10pm - 4:30pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session B: Data Science Technology & Applications

4:30pm PST

Running Novel FWI Algorithms Cost-Efficiently by Integrating Cloud-Native Architecture onto an Existing MPI-Based Production Codebase
WATCH VIDEO

Over the past year, S-Cube has been building a cloud-based FWI service to tackle challenging subsurface environments by employing a variety of novel algorithms. This includes adaptive loss function formulations and TV-based constraints which are needed to build salt structures directly from raw seismic data to enhance salt-affecting imaging and locate the most complex of targets. In this talk, we would like to share how we started with our existing legacy MPI-based production code designed to be run by a traditional on-prem job scheduler and gradually transformed it to natively take advantage of public cloud platform's main strengths in flexibility and on-demand scalability.

Our specific goals were (1) to utilize a heterogeneous set of compute hardware throughout the lifecycle of a production FWI run without having to provision them for the entire duration, (2) to take advantage of cost-efficient spare-capacity compute instances without uptime guarantees, and (3) to maintain a single codebase that can be run both on on-prem HPC systems and on the cloud. To achieve these goals meant transitioning the job-scheduling and "embarrassingly parallel" aspects of the communication code away from using MPI, and onto various cloud-based orchestration systems, as well as finding cloud-based solutions that worked and scaled well for the broadcast/reduction operation. Placing these systems behind a customized TCP-based stub for MPI library calls allowed us to run the code as-is in an on-prem HPC environment, while on the cloud we can asynchronously provision and suspend worker instances (potentially with very different hardware configurations) as needed without the burden of maintaining a static MPI world communicator. 

In salt-dominated fields where we employ a heavy TV regularization to the velocity model after gradient updates, this architecture decreased both our costs and runtime by 40% due to the ability to provision GPU instances for the regularization calculation and to suspend the main pool of workers used for wavefield simulations during this stage of the algorithm.
Speakers
TT
NS
ST

Srinivas Tadepalli

Amazon Web Services

S Cube Final pdf
Monday March 2, 2020 4:30pm - 4:50pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session A: HPC Standards & Applications & Cloud Usage

4:30pm PST

D-FLASH: A Distributed MPI Based LSH System for Large-Scale Similarity Search
WATCH VIDEO

We present D-FLASH, an MPI based distributed version of FLASH algorithm for similarity search with ultra-high dimensional and large datasets that do not fit any single machine. Our work extends the powerful Locality Sensitive Hashing (LSH) based FLASH system, which is the fastest approximate similarity search method on a single machine, to multiple nodes. D-FLASH. The distributed implementation allows FLASH to scale to massive distributed datasets that do not fit a single machine. We provide two novel workarounds to reduce the inter-machine communication overheads in FLASH, which leads to an order of magnitude improvements. First, we replace the histogram computations in FLASH with an approximate Topkapi subroutine, which reduces the inter-node communication while aggregating hash buckets. We then utilize a unique property of Topkapi that it is a mergeable sketch. In particular, this allows us to aggregate buckets in a tree structure. As a result, the query reduction process only requires MPI Allreduce instead of standard MPI Reduce. Due to MPI Allreduce the inter-node communication cost goes down exponentially. We provide a rigorous evaluation of every aspect of our system on the KDD 2012 click-through prediction competition, which contains around 150 million instances with 54 million feature dimensions. We provide a complete spectrum of performance evaluation on 18 computing nodes. The results show that our method leads to more than 10x reduction in the query time over a standard MPI implementation of FLASH.
Speakers
NM

Nicholas Meisburger

Rice University
AS

Anshumali Shrivastava

Rice University

4. D FLASH pdf
Monday March 2, 2020 4:30pm - 4:50pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session B: Data Science Technology & Applications

4:50pm PST

Large Scale Seismic Processing in Public Cloud
WATCH VIDEO

The complexity involved in using APIs and resource managers from multiple cloud providers, supercomputing centers, and partners highlights the need for having a single, unified way to launch and manage computation workloads on multiple target machines – on-premises or in the cloud. We were able to efficiently run a large scale 3D RTM benchmark with thousands of shots on an on-demand cloud-based GPU-cluster providing 5 PFlops (peak SP). For this run, we used a domain decomposition and process placement schema that achieves a sustained communication performance close to 160Gb/s on the border exchange used by the wave propagation finite difference method.
Speakers
PS
LF

Luiz Felipe

Petrobras
PA

Paulo Aragão

Amazon Web Services
LB

Lito Bejarano

Atrio Inc.
DT
AS
AA
FS

Fernando Sierra

Atrio Inc.
MA

Max Alt

Atrio Inc.

Large Scale Seismic Processing in Public Cloud Presentation pdf
Monday March 2, 2020 4:50pm - 5:10pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session A: HPC Standards & Applications & Cloud Usage

4:50pm PST

RALI & MIVisionX Optimizing Seismic Interpretation / Live Demo
What is MIVisonX? 
MIVisionX toolkit is a set of comprehensive computer vision and machine intelligence libraries, utilities, and applications bundled into a single toolkit. 

What is RALI? 
The AMD Radeon Augmentation Library (RALI) is designed to efficiently decode and process images and videos from a variety of storage formats and modify them through a processing graph programmable by the user. Rali currently provides C API.

RALI & MIVisionX optimizing seismic interpretation:

Seismic data processing will be manipulated before the end user can extract meaningful information regarding subsurface structures. This modification adds modeling and human biases but using seismic data directly is becoming possible with DL (deep learning). A DL uses analog velocity models and realistic raw seismic waveforms as input and produces subsurface velocity models as output. 

DL Technik requires a huge amount of seismic data and getting this amounts of seismic data sets and label them is not so easy but without sufficient data, DL algorithms will fail or overfit. 

RALI can help when insufficient data are used for training. The lack of quality data is solved by augmenting the original training data set with RALI. The new data can then be used by DL driven seismic tomography as input. 

The DL algorithm can predicts velocity models with higher accuracy after being trained with RALI velocity models. 

This unsupervised approach, Data Augmentation with RALI (using RALI to generate new labeled data) drive a way to mitigate this problem and opens geology, geophysics, and planetary sciences to more DL applications.
Speakers
MS

MIVisionX EVS TEMP Feb26 pdf
Monday March 2, 2020 4:50pm - 5:10pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session B: Data Science Technology & Applications

5:10pm PST

Networking Reception
Monday March 2, 2020 5:10pm - 7:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Networking
 
Tuesday, March 3
 

7:30am PST

Breakfast & Networking
Tuesday March 3, 2020 7:30am - 8:30am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Networking

8:30am PST

Welcome
WATCH VIDEO
Speakers
JO

Jan Odegard

Executive Director of the Ken Kennedy Institute, Rice University

Tuesday March 3, 2020 8:30am - 8:45am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030

8:45am PST

Citius, Altius, Fortius (Faster, Higher, Stronger): How Total’s HPC R&D Activity Supports the Group’s Ambition to Become the Responsible Energy Major
At the time of this presentation at the 2020 Rice OG-HPC Conference (and during a short window!), Total operates the largest, publicly-owned HPC in the world, located in southwest France.
The effort to provide computer performance in support of our business units is not new in our large, integrated, energy companies.

Details on Pangea III by the Project Manager who recently joined the Houston Total R&D Team will be provided in another presentation.

The talk will focus on the deep-seated change that Total initiated in 2016 to become the “Responsible Energy Major”.

The presentation will cover the evolution of Total’s R&D organization and portfolio, how digital and HPC are cornerstones of this transition, what lies ahead and how the Houston-based team supports this process.
Based on many years of experience in the areas of multicultural, collaborative work in competitive and uncertain environments, the author will also share lessons learned on how to go faster, higher, stronger .. and nimbly draw on the firepower and talent available in a large company.
Speakers
VS

Vincent Saubestre

CEO & President, Total EP Research & Technology

Vincent Saubestre PDF 2020 03 03 Total OGHPC Rice SaubestreWeb pdf
Tuesday March 3, 2020 8:45am - 9:30am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Keynote

9:30am PST

The Ion, A Catalyst for the Transformation and Growth of the Houston Technology Ecosystem
WATCH VIDEO

Collaboration is key to accelerating the growth of Houston’s technology economy. The Ion (4201 Main Street) will anchor the 16-acre South Main Innovation District and is destined to become an epicenter for Houston’s innovation ecosystem as an inclusive, dynamic, vibrant and highly dense hub focusing on quality collaborations between entrepreneurs, incubators, accelerators, corporations and the academic community when it opens in early 2021. In this talk we will outline how the Ion will play an important role, serving the entire ecosystem as a place where those collaborations occur. Houston’s incubators and accelerators, universities, corporations and venture capitalists will have a hub to leverage the innovation taking place all across the Houston region. The 270,000-square-foot building will accommodate multiple uses, including shared workspace, prototyping and maker resources, event space, classrooms, food and beverage offerings as well as indoor/outdoor communal areas with shared amenities.
Speakers
GG

Gabriella “Gaby” Rowe

Executive Director, The Ion

Gaby Rowe HPC FINAL Presentation 2020c pdf
Tuesday March 3, 2020 9:30am - 10:00am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Plenary

10:00am PST

Helping Innovative Technology Startups Succeed
Intel Capital has invested in technology startups for nearly 30 years, fostering an ecosystem of innovation. Since 1991, Intel Capital has invested US$12.6 billion in more than 1,560 companies worldwide, and 677 portfolio companies have gone public or participated in a merger. This talk will share insights on how Intel helps its portfolio companies succeed.
Speakers
TA

Tamiko A. Hutchinson

VP of Intel Capital and Senior Managing Director of Portfolio Development at Intel Corporation, Intel Capital

Tuesday March 3, 2020 10:00am - 10:30am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Plenary

10:30am PST

Break
Tuesday March 3, 2020 10:30am - 11:00am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Networking

11:00am PST

Pangea III - A Seismic Shift in Total's Computing Platform
WATCH VIDEO

Oil and Gas industry is continuously challenged to increase its hydrocarbon production in response to the growing demand for energy. Finding new oil and gas fields has become more challenging as resources are no longer easily accessible with an increase of geological complexity and environmental constraints. High-resolution and reliable subsurface imaging is getting more and more key but more and more internal customers are knocking at the door …
Speakers
DK

Tuesday March 3, 2020 11:00am - 11:20am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session C: HPC Systems & Facilities

11:00am PST

Serverless Seismic Imaging in the Cloud
WATCH VIDEO

This talk presents a serverless approach to seismic imaging in the cloud based on high-throughput containerized batch processing, event-driven computations and a domain-specific language compiler for solving the underlying wave equations. A 3D case study on Azure demonstrates that this approach allows reducing the operating cost of up to a factor of 6, making the cloud a viable alternative to on-premise HPC clusters for seismic imaging.
Speakers
PW

Philipp Witte

Georgia Institute of Technology
ML

Mathias Louboutin

Georgia Institute of Technology
CJ
FH

Felix Herrmann

Georgia Institute of Technology

Rice O&G pdf
Tuesday March 3, 2020 11:00am - 11:20am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session D: HPC Software Tools & Technologies

11:20am PST

Experience Installing a Cluster with Direct-to-Chip Water Cooling
WATCH VIDEO

At the end of 2019, BP installed our latest compute cluster of 720 x Cascade Lake-AP nodes using direct-to-chip water cooling. When we built our building in 2013 we anticipated that computer systems would eventually need water cooling to operate effectively. Due to power requirements of this new system and the forecast for future systems, we decided that now was the right time to implement a direct-to-chip water cooling solution. We will discuss some of the determining factors for this decision. This is the first step in a multi-year transition where all new clusters will be water cooled in one way or another. We will discuss the coordination required between building engineers, system architects, hardware vendors, installation personnel, and operations personnel. We will discuss how we will operate during a transition phase then what we can do once all our systems are water cooled. We will discuss what we learned during planning, installation, initial testing, tuning, and operation. We will discuss some primary benefits plus some secondary benefits that are advantageous to our users.
Speakers
KB
GK

20200303 OG HPC Presentation FINAL pdf
Tuesday March 3, 2020 11:20am - 11:40am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session C: HPC Systems & Facilities

11:20am PST

Large Scale-Data Reduction and Data State Management with SZ and VeloC
WATCH VIDEO

Data movement and storage are becoming main capability and capacity bottlenecks for many scientific applications at large-scale. These bottlenecks are affecting important system resources (memory, communication network, file system) and lead to significant performance degradations. In this talk we will present two main technologies mitigating these bottlenecks: (i) SZ: a multi-algorithm, adaptable lossy data compression framework and (ii) VeloC: a multi-level, asynchronous data state management framework. They can be used separately or in combination for different use-cases: a) running problems larger than what the memory can accommodate (without relying on out-of-core technique), b) storing more results than what the file system allows, c) streaming data at a higher intensity than what the network can sustain, d) accelerating drastically file accesses, e) reducing the overhead of fault tolerance to negligible level, f) accelerating DNN training. We will present at the conference results including for Oil and gas applications.
Speakers
FC

Franck Cappello

Argonne National Laboratory and University of Illinois at Urbana Champaign
SD

Sheng Di

Argonne National Laboratory
BN

Bogdan Nicolae

Argonne National Laboratory

Tuesday March 3, 2020 11:20am - 11:40am PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session D: HPC Software Tools & Technologies

11:40am PST

Cloud Capability Brings Global Flexibility and Increased Scalability to E&P HPC Workloads
Exploration and Production is a global business however the HPC loads used to support seismic imaging and reservoir simulation have often been restricted to a few on-Prem data center locations with the hardware capabilities to deliver the required compute power. The advent of cloud based HPC is starting to shift that paradigm. Chevron has partnered with Microsoft to pilot and deliver the architecture required to bring their proprietary seismic imaging and reservoir simulation. The pilots focused on delivering these workflows from the Microsoft South Central region where they were supported by the HB clusters running on AMD EPYC 7551 (Naples) series technology. Workloads performance scaled effectively to the cloud providing Chevron the ability to scale beyond the capacity of its on-prem data centers in times of high demand. As these technologies get deployed to more Azure regions this will enable Chevron’s global business units to run their HPC workloads without having to coordinate through a central data center.

Partnering with Microsoft also frees Chevron from a static lifecycle of capitalized HPC hardware. Chevron is now testing these workloads on the HBv2 based on AMD EPYC 7742 (Rome) processors accelerating Chevron’s access to this technology ahead of when it or other new technology would be able to be brought into the On-Prem data center.

These new cloud based capabilities are bringing the digital oil field to life by greater access to HPC tools for the engineers and geophysicists making the decisions required to provide the world with energy it needs for today and tomorrow.
Speakers
JN
JP

Joe Patry

Chevron
HS

Hussein Shel

Microsoft
AM

Alex Morris

Microsoft

Tuesday March 3, 2020 11:40am - 12:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session C: HPC Systems & Facilities

11:40am PST

Automatic Code Generation for GPUs using Devito
WATCH VIDEO

This presentation focuses on adding GPU support to the finite difference domain
specific language Devito. Devito is already capable of generating highly
optimized finite difference code for CPUs, including the Intel KNL, ARM and
Power architectures, parallelized using OpenMP and MPI. It is typically used
as a wave propagator and to calculate gradients using the adjoint-state
method in RTM and FWI. We consider a range of GPU programming models for
automatic code generation with respect to simplicity, performance and
portability to GPUs from different chip manufacturers.

We present early performance results on GPUs using automatically generated
code with OpenMP 5 offloading. We also illustrate how to use the software
graceful degradation feature in Devito to enable HPC programming specialists
to augment the generated code and hand-tune performance. Successful (e.g.
reduced time to solution) code augmentation experiments can then be analyzed
to identify strategies that are then incorporated to the compiler.
Speakers
FL

Fabio Luporini

Imperial College London
GG

Gerard Gorman

Imperial College London

rice2020 devito gpu pdf
Tuesday March 3, 2020 11:40am - 12:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session D: HPC Software Tools & Technologies

12:00pm PST

Moving Massive Amounts of Data across Any Distance Efficiently
WATCH VIDEO

The objective of this talk is to present two on-going projects aiming at improving and ensuring highly efficient bulk transferring or streaming of massive amounts of data over digital connections across any distance. It examines the current state of the art, a few very common misconceptions, the differences among the three major type of data movement solutions, a current initiative attempting to improve the data movement efficiency from the ground up, and another multi-stage project that shows how to conduct long distance large scale data movement at speed and scale internationally. Both projects have real world motivations, e.g. the ambitious data transfer requirements of Linac Coherent Light Source II (LCLS-II) [1], a premier preparation project of the U.S. DOE Exascale Computing Initiative (ECI) [2]. Their immediate goals are described and explained, together with the solution used for each. Findings and early results are reported. Possible future works are outlined.
Speakers
CF

Chin Fang

Zettar Inc.

zettar 2020 pdf
Tuesday March 3, 2020 12:00pm - 12:20pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session C: HPC Systems & Facilities

12:00pm PST

Hybrid-Parallel Multigrid Preconditioners in a Pore-Scale Flow Simulation
WATCH VIDEO

The simulation of multi-phase flow at the pore scale has important applications in hydrocarbon recovery, including the estimation of relative permeabilities and capillary pressure. Accurate simulations with advanced numerical methods, such as discontinuous Galerkin (DG) methods, are computationally demanding, and the feasibility of such simulations often depends on the availability of scalable iterative linear solvers. In this presentation, we show how multigrid methods can accelerate the solution of linear systems in DG discretizations of two-phase flow in complex pore geometries. We discuss the hybrid-parallel implementation of a combined p-multigrid and algebraic multigrid method and evaluate its computational performance in numerical experiments with pore-scale flow simulations in rock samples. We demonstrate speedups of up to 15.4x compared to a Jacobi-preconditioned conjugate gradient solver, resulting in speedups of up to 2.2x for the overall simulation.
Speakers
CT

Christopher Thiele

Rice University
BR

Beatrice Riviere

Rice University

rice oghpc20 pdf
Tuesday March 3, 2020 12:00pm - 12:20pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session D: HPC Software Tools & Technologies

12:20pm PST

Timing-Agnostic SVE Analysis of Seismic Kernels
The diversity of HPC architectures available~(heterogeneity, high core counts or depth of the memory hierarchy) leads to increasing concerns regarding real applicative performances. Therefore, it is admitted that co-design approaches will play a major role to ensure that oil and gas applications will be in best position to get the most of future hardware designs. In this paper, after a review of recent contributions for the optimization of geophysical stencils, we discuss key feature from Arm architectures~(e.g. SVE) that may influence classical implementations of seismic kernels.
Speakers
FD
CH

Tuesday March 3, 2020 12:20pm - 12:40pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session C: HPC Systems & Facilities

12:20pm PST

Discontinuous Galerkin Method for Seismic Applications
WATCH VIDEO

In geophysical problems, the goal is to get information from the subsurface using imaging (RTM) and inverse problems (FWI). Both those methods requires to solve wave equations in complex media. In some cases where finite differences has difficulties, an alternative is to use Discontinuous Galerkin (DG) methods.

There are multiple strategies that all contribute to the computational gain of DG methods for complex seismic applications. In this work we present the development of different DG strategies in order to accelerate the computations given an unstructured computational mesh. Variable spatial order regulates the spatial distribution of the degrees of freedom. Multi-rate time stepping allows us to take different timesteps in different regions of the model. This allows to take a smaller timestep where it is required for stability, due to the computational mesh. Using the Bernstein-Bezier formulation, we can obtain sparse operators. Higher order model representation can be obtained using the Weight Adjusted DG to increase the accuracy of the solution. This is important when we increase the spatial order of the wavefield, since the mesh can contain larger elements. We present the latest development for porting these codes to GPUs, using openACC. Finally we demonstrate some full waveform inversion results using this solver.
Speakers
AA

Andreas Atle

Total EPRT USA
AC

Aurelien Carron

Total EPRT USA
PJ

Pierre Jacquet

INRIA Magique3D

Rice O&G Andreas Atle pdf
Tuesday March 3, 2020 12:20pm - 12:40pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Parallel Session D: HPC Software Tools & Technologies

12:40pm PST

Lunch & Networking
Tuesday March 3, 2020 12:40pm - 1:45pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Networking

1:45pm PST

The Cray Shasta Architecture: Designed for the Exascale Era
WATCH VIDEO

With the announcement of multiple exascale systems, we’re now entering the Exascale Era, marked by several important trends. CMOS is nearing the end of its roadmap, leading to hotter and more diverse processors as architects chase performance through specialization. Organizations are dealing with ever larger volumes of data, stressing storage systems and interconnects, and are increasingly augmenting their simulation and modeling with analytics and AI to gain insight from this data. And users and administrators are demanding flexible, cloud-like software environments that let them flexibly manage their systems, and develop and run code anywhere. While these issues are most acute in extreme scale HPC systems, they are becoming increasingly relevant across the broader enterprise. This talk provides an overview of the Cray Shasta system architecture, which was motivated by these trends, and designed for this new heterogeneous, data-driven world.
Speakers
SS

Steve Scott

SVP, Senior Fellow, & CTO, Hewlett Packard Enterprise

Cray Shasta Steve Scott Rice Oil Gas 3 3 20 2 pdf
Tuesday March 3, 2020 1:45pm - 2:30pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Keynote

2:30pm PST

Stable High Order Methods for Time-Domain Wave Propagation in Complex Geometries and Heterogeneous Media
WATCH VIDEO

Discontinuous Galerkin (DG) methods enable high order accurate time-explicit simulations on domains with complex geometric and interior interfaces. However, DG methods face several challenges: they can lose stability or high order accuracy in the presence of heterogeneous media or complex physics, and they become significantly more expensive as the order of approximation increases. In this talk, we review developments which enable provably stable, high order accurate, and efficient time-explicit simulations for acoustic, acoustic-elastic, and poro-elastic wave propagation in complex geometries and heterogeneous media.

Speakers
JC

Jesse Chan

Rice University

Chan OGHPC 2020 pdf
Tuesday March 3, 2020 2:30pm - 3:00pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Plenary

3:00pm PST

Performance Tools for GPU-Accelerated Computing
WATCH VIDEO
Speakers
JM

John Mellor-Crummey

Professor of Computer Science and of Electrical and Computer Engineering, Rice University

7. John Mellor Crummey pdf
Tuesday March 3, 2020 3:00pm - 3:30pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Plenary

3:30pm PST

Poster Session & Networking Reception
3D Immersed Boundary Generation From Topography Point Clouds: an Implementation in Devito
Edward Caunt, Rhodri Nelson, Fabio Luporini and Gerard Gorman


3D Printed Tubulanes as Light-weight Hypervelocity Impact Resistant Structures
Seyed Mohammad Sajadi, Cristiano F. Woellner, Douglas S. Galvão, Chandra Sekhar Tiwary and Pulickel M Ajayan


3D Printing Under Visible Light Using Cadmium Selenide Semiconducting Quantum Dots as Photoinitiators
Yifan Zhu and Eilaf Egap


A Discontinuous Galerkin Method for the Two-phase Flow in Deformable Porous Media
Boqian Shen and Beatrice Riviere


A Framework for Monetary Loss Assessment of Petroleum Infrastructure in Storm Surge Events
Kendall Capshaw, Majid Ebad Sichani, Katherine Anarde and Jamie Padgett


A Graphical Tool for Latent Feature Representation
Ye Emma Zohner and Jeffrey Morris


A Python Framework for Rapidly Migrating, Building and Optimizing a High-Performance Parallel Multi-Stencil System
Charly Bechara and Santiago Cortijo


A Spline Theory of Deep Networks
Randall Balestriero and Richard Baraniuk


A Tool for Top-Down Performance Analysis of GPU-Accelerated Applications
Keren Zhou, Mark Krentel and John Mellor-Crummey


aiRock(TM): Simultaneous Chemistry and Physics Simulations Using High-Performance Computing in the Cloud and Machine Learning
Babak Shafei


Accelerating Real-World Stencil Computations using Temporal Blocking: Handling Sparse Sources and Receivers
George Bisbas, Fabio Luporini, Mathias Louboutin, Gerard Gorman and Paul H.J. Kelly


Break GPU Memory Wall Using Distributed Deep Learning for Large Scale Image Segmentation
Sergio Botelho, Vinay Rao and Santi Adavani


Columns with Multiple Phase Divisions: Simulation and Application in Crude Oil Distillation
Lilian Biasi, Matthias Heinkenschloss, Fabio Batista, Roger Zemp, Ana Romano and Antonio Meirelles


Data-Driven Forward and Spot Price Manipulation Detection: Gazprom Case
Aminam Talipova


Data Weighted Full-Waveform Inversion for Near-Surface Seismic Refraction Data
Ao Cai and Colin Zelt


Exact Blind Community Detection from Signals on Multiple Graphs
T. Mitchell Roddenberry, Santiago Segarra, Michael T. Schaub and Hoi-To Wai


Experiments and Modeling of Porous Media Flow Using Laser-Diagnostics and Direct Numerical Simulation
Stephen King, Mustafa Alper Yildiz, Robert Muyshondt, Thien Nguyen, Yassin Hassan and Victor Ugaz


Fast Generation of Virtual 3D Concrete Microstructures Coupled with Experimentally Validated Creep Prediction Model
Christa Torrence, Aishwarya Baranikumar, Edward Garboczi and Zachary Grasley


Fast Inversion of Directional Resistivity Logging-While-Drilling Data Using Supervised Descent Method
Yanyan Hu, Rui Guo, Yuchen Jin, Xuqing Wu, Maokun Li, Aria Abubakar and Jiefu Chen


GPU Accelerated Modeling of Triaxial Induction Resistivity Logging and Electromagnetic Telemetry in Multilayered Formation
Shubin Zeng, Han Lu, Xin Fu, Jiefu Chen and Yueqin Haung


High Performance Seismic Imaging Using Low-Rank Matrix Computations
Kadir Akbudak, David Keyes and Hatem Ltaief


Investigating CEO Individual Differences and Firm Outcomes Using Deep Learning
Ying Fung Yiu and Chi Hon Li


Make Multicast Great Again
Afsaneh Rahbar, Sushovan Das, Xinyu Wu, Weitao Wang, Dingming Wu, Ang Chen and Eugene Ng


Modeling the Effects of Stress, Defects and Surface Reaction on Microstructure Evolution in Battery Electrodes
Kaiqi Yang, Liang Hong and Ming Tang


Newton-Based Methods for the Numerical Solution of Risk-Averse PDE-Constrained Optimization Problems
Mae Markowski and Matthias Heinkenschloss


Observation and Elucidation of Stress-Induced Non-Uniform Reaction Behavior in LiFePO4 Secondary Particles
Fan Wang, Kaiqi Yang, Mingyuan Ge, Wah-Keat Lee and Ming Tang


Optimising Finite Difference Schemes through Exploiting Subdomains in Devito
Vitor Hugo Mickus Rodrigues, Rhodri Nelson, Gerard Gorman and Samuel Xavier-De-Souza


Parareal-Based Preconditioners for Linear-Quadratic Optimal Control Problems
Shengchao Lin and Matthias Heinkenschloss


Phase-field Modeling of Micellar Morphological Transitions
Shaoxun Fan and Ming Tang


PHS: A Toolbox for Parellel Hyperparameter Search
Peter Habelitz and Janis Keuper


Porosity Prediction Using the Spectral Decomposition Attributes and Machine Learning
Lian Jiang, Leon Thomsen and John Castagna


Preconditioner for Estimation of Multipole Sources via Full Waveform Inversion
Mario Bencomo and William Symes


Predictor of Best Catalytic Performance for Fischer Tropsch Activity: A DFT Study
Sumegha Godara, Daniela Mainardi and Suraj Gyawali


Recurrent Neural Network for Predicting Polycrystalline Microstructure Evolution
Yifan Cao, Kaiqi Yang, Ming Tang and Fei Zhou


Repeated and Merged Bloom Filter (RAMBO) for Sequence Search: Indexing 170TB data in 14 hours
Gaurav Gupta, Minghao Yan, Benjamin Coleman, R. A. Leo Elworth, Todd Treangen and Anshumali Shrivastava


Scalability Studies for Nonlinear Contact Problems
Nidish Narayanaa Balaji and Matthew Rw Brake


Self-Evolved Hierarchical A* Search
Ying Fung Yiu and Rabi Mahapatra


Shared Memory Communication Optimizations to Adaptive MPI
Sam White and Laxmikant Kale


Simplifying Heterogenous Computing
Sujata Tibrewala and Gergana Slavova


Simulation Study of CO2-Foam Injection for Diversion of a Hydrocarbon Miscible Solvent in a Fractured Carbonate Reservoir
Reza Amirmoshiri, Manmath Panda, Raul Valdez, Sibani Lisa Biswal and George Hiraski


Stabilizing Deep Convolutional Neural Networks for Image Segmentation
Jonas Actor, Beatrice Riviere and David Fuentes


StreamQL: A Query Language for Efficient Data Stream Processing
Lingkun Kong and Konstantinos Mamouras


Stress Effects on Phase Morphology and Compositional Non-Uniformity in Intercalation Compounds
Youtian Zhang and Ming Tang


Two-Phase Partially Miscible Flow Simulation in Porous Media
Lu Lin and Beatrice Riviere


Wettability Alteration Implications on Pore-scale Multiphase Flow in Porous Media
Mohamed Nemer, Parthib Rao and Laura Schaefer


Tuesday March 3, 2020 3:30pm - 5:30pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Poster Session
 
Wednesday, March 4
 

8:15am PST

Workshop D: From Zero-to-Devito
The Devito workshop will consist of a morning session where participants will learn how to implement finite difference and inverse solvers using Devito. The workshop will be in Jupyter notebook which can be either installed on the participants laptops or run directly on a BinderHub instance provided for the workshop. The afternoon session will be run hackathon style, giving experienced HPC developers the opportunity to develop the Devito JIT-escape hatch (ie directly customizing the Devito generated source code), where the target problem is performance optimization of Devito on GPU processors running on Azure Cloud.
Speakers
DG

Dr. Gerard Gorman

Imperial College London
DF

Dr. Fabio Luporini

Imperial College London
DR

Dr. Rhodri Nelson

Imperial College London
MN

Mr. Navjot Kukreja

Imperial College London

Wednesday March 4, 2020 8:15am - 3:45pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Workshop

8:15am PST

Workshop E: HPC in the Cloud Tutorial
Abstract: Cloud computing technologies have rapidly matured and nearly all HPC production workloads can now be efficiently accommodated in a cloud environment. Cloud computing presents a wide array of services and the ability to accommodate the most demanding computational workloads at scale. However, the complexity and scale that comes with such an environment also can make the first experience a daunting proposition.

During this tutorial, attendees will be introduced to compute, storage, and network technologies offered by AWS and demonstrate how to best utilize them in HPC workflows. This will be followed by a deep dive on the computational capabilities offered by cloud computing and, in particular, autoscaling and serverless computing. An overview of storage and network will be provided with a discussion on optimizing for object storage or parallel file systems.

This tutorial will be composed of presentations and hands-on sessions where attendees will have the opportunity to put into practice their learnings on the AWS cloud. A laptop with a wifi connection is required for the hands-on portions of this tutorial. No prior cloud experience is required, but participants will need minimal comfort with the Linux command line interface. AWS accounts will be provided.

Note: Attendees will need to come with a laptop or tablet with a keyboard with Wifi. The hands-on can be done through the web browser and accounts will be provided for the day with no setup time required.
Speakers
PY

Pierre Yves-Aquilanti

Amazon Web Services
MK

Matt Koop

Amazon Web Services
SS

Stephen Sachs

Amazon Web Services

Wednesday March 4, 2020 8:15am - 3:45pm PST
BioScience Research Collaborative (BRC) Building Rice University, 6500 Main Street, Houston, TX 77030
  Workshop
 
Filter sessions
Apply filters to sessions.
Monday, March 2
Tuesday, March 3
Wednesday, March 4
BioScience Research Collaborative (BRC) Building
  • Keynote
  • Networking
  • Parallel Session A: HPC Standards & Applications & Cloud Usage
  • Parallel Session B: Data Science Technology & Applications
  • Parallel Session C: HPC Systems & Facilities
  • Parallel Session D: HPC Software Tools & Technologies
  • Plenary
  • Poster Session
  • Workshop
  • Popular