Past Winners

• Alenia Aermacchi (Italy). Engineers at Alenia Aermacchi utilized technical computing and scientific computing tools in the “Clean Sky” project to design a new generation of environmentally friendly aircraft. While seeking the most promising configuration, the engineers shaped two different wings to fulfill separately the two goals of enhancing aerodynamic performance and reducing the wing weight. This was accomplished through the use of CFD and other technical computing tools. The two promising configurations enhanced cruise efficiency by 2.5% and reduced the wing weigh by 4%. Lead: Enrica Marentino
• High Performance GeoComputing Laboratory at the University of California, San Diego (U.S.) The High Performance GeoComputing Laboratory (HPGeoC) at the San Diego Supercomputer Center, an Organized Research Unit of the University of California, San Diego, has developed a highly scalable and efficient GPU-based finite difference code based on AWP-ODC, a community code developed and supported by the Southern California Earthquake Center (SCEC) for large-scale earthquake simulations. This GPU code, re-structured to enable maximized throughput and reduced time-to-solution, achieved near-perfect weak-scaling parallel efficiency and sustained 2.33 petaflop/s on ORNL’s Titan supercomputer. Moreover, a 110-fold speedup has been demonstrated over CPU in key strain tensor calculations critical to probabilistic seismic hazard analysis. This achievement makes a California state-wide hazard model a goal that’s now reachable with existing supercomputers. The performance of the code is expected to take physics-based seismic hazard analysis to a new level using petascale, heterogeneous computing resources, with the potential of saving hundreds of million core-hours as required by building engineering design. Lead: Yifeng Cui
• DOD High Performance Computing Modernization Program (U.S.). This HPCMP-supported project, within the armor/anti-armor portfolio, provided direct support to the Warfighter program. The fundamental goals for using modeling and simulation in support of the armor/anti-armor development programs were to reduce time, resources, and risk while improving the quality of information available to designers, users, and decision makers. HPCMP capabilities enabled the development and testing of new armor/anti-armor capabilities in the following areas: propulsion, interior ballistics, external trajectory determination, terminal ballistics, warhead analysis, and sensors. The total upper bound ROI amounted to $935 million. Lead: Deborah Schwartz
• DOD High Performance Computing Modernization Program (U.S.). This HPC innovation comprised a suite of cutting-edge computational efficiency enhancement methodologies applied to complex missile-related, aero-propulsive problems with combustion. These methodologies were integrated into CRAFT CFD® and CRUNCH CFD®, two CFD codes in widespread use by DoD to support missile design and evaluation, and encompassed gas-phase/multi-phase combustion as well as laminar/turbulent chemistry, including tabulated/neural network approaches, reduced/adaptive chemistry, turbulent scalar fluctuation model (SFM) and GPU acceleration. By leveraging state-of-the-art HPC resources provided by DoD HPCMP, these innovative methodologies delivered higher-fidelity predictive capabilities to the analysis of missile systems/components, thereby enabling CFD to serve as a cost-effective design tool and virtual testbed for missile evaluation. Lead: John West
• ESTECO and Airworks Engineering (Italy).  Airworks, a multidisciplinary company for mechanical engineering, was facing the challenge of improving efficiency in converting wind energy into electrical power by optimizing the whole assembly of a wind power unit rotor. Experts in the firm’s wind power unit, specialists in CAD (computer aided design) and CFD (computational fluid dynamics) professionals from different organizations were involved in a complex design scenario and needed to collaborate effectively. Engineers from the University of Trieste prepared the parametric CAD model and set up CFD simulations, while Airworks developed aerodynamic performance calculations of the wind rotor blade and subsequently performed the optimization analysis. The technical computing-driven solution enabled the seamless execution of inter-organizational simulation workflows. With the set up of the optimization workflow, Airworks professionals were able to explore and evaluate new parametric geometry, leading to innovative designs. The end result was a wind turbine design with an outstanding power coefficient and an annual energy production increase as high as 1.26%. Lead: Paolo Vercesi
• University College London and NAG HECTOR dCSE (UK). HPC experts from NAG, working under NAG’s Computational Science and Engineering (CSE) support service for HECToR, the UK’s national academic supercomputing facility, have optimized a Quantum Monte-Carlo application for multicore architectures, resulting in a performance increase of a factor of four. The objectives of this dCSE project were to enable the CASINO Quantum Monte Carlo code to effectively use the multicore processors of HECToR’s Cray XT supercomputer and thus model more complex physical systems with greater efficiency. Shared memory techniques were introduced to allow larger models to be computed with greater efficiency by enabling multiple MPI processes on a single node to share a common data set, thus reducing the number of nodes needed for a given simulation. Further work including hierarchical parallelism with OpenMP and I/O optimizations improved the scalability of the code, enabling CASINO to run 60-80% faster for simulations using more than 10,000 cores. Following NAG’s work, the scientists were able to run on 40,000 cores of the Jaguar petascale supercomputer at Oak Ridge National Laboratory. It is estimated that this dCSE work saved 12 million AUs (allocation units) for a one year research project on HECToR, equivalent to savings of as much as £760k and the potential for future savings of up to several million pounds. Lead: HECToR dCSE
• University of Warwick and NAG HECTOR dCSE (UK). HPC experts from NAG, working under NAG’s Computational Science and Engineering (CSE) support service for HECToR, the UK’s national academic supercomputing facility, have improved the scalability and performance of DL_POLY_3, a widely used software package for studying molecular dynamics. The 20-fold improvement in performance achieved by this project enabled a study of egg-shell formation that was infeasible with previous performance. DL_POLY_3 is a general-purpose package for classical molecular dynamics (MD) simulations from STFC Daresbury Laboratory. University of Warwick researchers Mark Rodger and David Quigley, in collaboration with colleagues at the University of Sheffield, used DL_POLY_3 and the HECToR supercomputers to study the role of a protein called ovocleidin-17 (OC-17) in chicken eggshell formation. Significant performance improvements were needed to make the modeling possible in feasible timescales using the HECToR supercomputers, especially in terms of parallel I/O. Lead: HECToR dCSE
• Bottero S.p.A. (Italy). Bottero has recently introduced the EMOC, a completely new family of Mold Opening and Closing (MOC) mechanisms for the hollow glass industry. This complex project combined the innovation, in terms of mold movement (speed and precision), cooling system, maintenance facility (“top mounted”), and maintaining compatibility with previous standards. This multidimensional aspect led to hard space constraints, requiring complex 3D kinematic schema, and required advanced CFD simulation and design tools. A high performance level was required for this mechanism, particularly regarding clamping forces in closed positions, force available at the beginning of the molds stroke, reduced closing time, absence of vibrations during movement, and robustness with respect to irregularities in air supply. CFD tools and simulation techniques were utilized to improve the design in EMOC. Lead: Alberto Marino
• Polestar Racing (Sweden). The Polestar Racing vehicle model had to be modified without considering the chassis design parameters, which were previously the core of the optimization analysis. Design simulation acquired an even greater importance as the best combination of the front-to-rear weight, aerodynamics, and brake distribution had to be executed in three days. Polestar Racing utilized a suite of simulation tools from MSC Software, the Lap Time Simulation (LTS) in-house code, and modeFRONTIER to devise a technical computing-driven approach to improve performance. The resulting improvements in weight, aerodynamics, and brake distribution led to lap time reductions ranging from 0.19 to 0.50 seconds. Lead: Per Blomberg
• RENCI (U.S.). Scientists and researchers at RENCI developed a comprehensive informatics framework called NCGENES, enabling medical decision support by systemizing genomic analysis and high performance computing to mine genomic data for clinical and research use. The framework tackles one of the biggest challenges in genomic medicine – the need for automated sorting of the millions of variants generated by genome-scale sequencing to identify the very few with actual clinical relevance. The innovative framework also provides a proof of principle for how genomic medicine can be carried out in a practical manner. NCGENES breaks new ground by guiding and defining best practices for the use of whole genome sequencing as a diagnostic tool to aid patients and clinicians in making sense of medical informatics. Lead: Phil Owen
• RENCI – ADCIRC Surge Guidance System (U.S.). This system uses the coupled coastal circulation, storm surge, and wind wave model ADCIRC+SWAN to produce high-resolution forecasts of storm surge, near shore waves, and water inundation when tropical or extra-tropical storms threaten the U.S. Atlantic or Gulf Coasts. The complete system, the ADCIRC Surge Guidance System (ASGS), includes an advanced web-based display (NC-Coastal Emergency and Risk Assessment). During an active storm, ASGS is run 2-4 times each day on a 150-node Dell PowerEdge M610/cluster (2 x 2.8Ghz Intel Nehalem-EP 5560, quad core) at the Renaissance Computing Institute (RENCI). The outputs from these runs are incorporated into guidance and forecasting efforts by the National Weather Service, the National Hurricane Center, and agencies such as the U.S. Coast Guard, the U.S. Army Corps of Engineers, FEMA, and local and regional emergency management personnel. The resulting forecasts are used for evacuation decisions, to position supplies and response personnel, for search and rescue, and for other event-based decision support as needed. Lead: Rick Luettich, Brian Blanton

• GE Global Research (U.S.) Using a 40 million CPU hour Department of Energy award, GE Global Research has modeled the freezing behavior of water droplets on six different engineered surfaces under six operating conditions on the hybrid CPU/GPU Titan at Oak Ridge National Lab (ORNL). Through recent advances in the field, including a joint simulation enhancement effort with Oak Ridge National Lab to fully leverage hardware infrastructures, GE Global Research has been able to accelerate simulations by approximately 200-fold compared to even just two years ago. Lead: Masako Yamada
• The Procter & Gamble Company (U.S.) P&G researchers and collaborators at Temple University developed models at the molecular and mesoscale level to understand complex molecular interactions of full formula consumer products such as shampoos, conditioners, facial creams, laundry detergents, etc. The HPC-driven research helped shed light on the performance of the complex formula interactions versus inferring performance based on isolated calculations. Results of the HPC -driven research led to a better understanding of interfacial phenomena, phase behavior, and the performance of several P&G products. Lead: Kelly L. Anderson
• National Institute of Supercomputing and Networking, Korea Institute of Science and Technology Information (Korea) The EDISON (EDucation and research Integration through Simulation On the Net) Project, funded by the Ministry of Science, ICT and Future Planning, Korea, established an infrastructure on the Web where users across the country could easily access and utilize various engineering/science simulation tools for educational and research purposes. The EDISON project is accelerating research in five key areas: Computational Fluid Dynamics, Computational Chemistry, Nano Physics, Computational Structural Dynamics, and Multi-disciplinary Optimization. The Project utilizes a novel partnership model between the project and the respective domains to develop area-specific simulation tools that make HPC accessible to domain specialists. Lead: Kumwon Cho
• GE Global Research (U.S.) GE Global Research’s work on Large Eddy Simulations (LES) leveraged petascale computing to break barriers in accurately characterizing the key flow physics of multi-scale turbulent mixing in boundary layer and shear flows. Findings from this research will significantly improve the prediction and design capabilities for next-generation aircraft engines and wind turbines, both from demonstrating the viability of LES as a characterization tool and as a source of physics guidance. Lead: Umesh Paliath
• Spectraseis Inc (U.S.) and CADMOS, University of Lausanne (Switzerland) Researchers doubled both acoustic and elastic solver throughput, at the same time improving code size and maintainability, harnessing the massive parallel computing capabilities of Fermi and Kepler GPUs. With improved efficacies obtained by code redesign for GPU the time to solution was reduced from hours to seconds. The improved capability allowed Spectraseis to move from 2D to 3D and, in several cases, obtain more than 100x speed-up. Lead: Igor Podladtchikov and Yury Podladchikov
• Intelligent Light (U.S.) Intelligent Light addressed the challenge of high volumes of CFD data using FieldView 14 data management and process automation tools. Intelligent Light contributed results from approximately 100 cases with more than 10,000 time steps each to deliver a complete response to the workshop objectives. A Cray XE6 was used to generate the CFD solutions and perform much of the post-processing. This project successfully demonstrated the value and practicality of using innovative workflow engineering with automation and data management for complex CFD studies. Lead: Dr. Earl P.N. Duque
• Facebook (U.S.) Facebook manages a social graph that is composed of people, their friendships, subscriptions, and other connections. Facebook modified Apache Giraph to allow loading vertex data and edges from separate sources (GIRAPH-155). Facebook was able to run an iteration of page rank on an actual one trillion edge social graph formed by various user interactions in fewer than four minutes with the appropriate garbage collection and performance tuning. Facebook can now cluster a monthly active user data set of one billion input vectors with 100 features into 10,000 centroids with k-means in less than 10 minutes per iteration.Lead: Avery Ching / Apache Giraph
• HydrOcean/Ecole Centrale Nantes (France) SPH-flow is an innovative fluid dynamic solver based on a meshless, compressible, and time-explicit approach. SPH-flow solver has been used in several industrial projects, including: impact forces of aircraft and helicopter ditching; free surface simulations of ship wake and wave fields; multiphase emulsion simulations; extreme wave impacts on structures; simulation of hydroplaning of tires; water film around car bodies; and underwater explosions. This project is lead by Dr. Erwan Jacqin, CEO of HydrOcean, a spinoff from Ecole Centrale fluid dynamic lab, and Prof. David Le Touze, in charge of the SPH-flow research team at Ecole Centrale Nantes.
• Imperial College London and NAG (UK) HPC experts from NAG and Imperial College London have implemented scientifically valuable new functionality and substantial performance improvements in the Incompact3D application. After the improvements, the simulations can now scale to 8000 cores efficiently, with a run time of around 3.75 days (wall-clock time), which is over 6x faster. Furthermore, meshes for new high resolution turbulence mixing and flow control simulations, which use up to 4096*4096*4096 grid points, can now utilize as many as 16384 cores. Lead: NAG HECToR CSE Team
• Queen Mary University of London and NAG (UK) NAG and Queen Mary University of London made significant improvements to CABARET (Compact Accurate Boundary Adjusting high Resolution Technique) code so that it may be used to solve the compressible Navier-Stokes equations and, in the context of this project, for the investigation of aircraft noise. The newly developed code was validated and tested against the serial code and a parallel efficiency of 72% was observed when using 250 cores of the XT4 part of HECToR with the quad core architecture. Lead: NAG HECToR CSE Team
• Southern California Earthquake Center (U.S.) SCEC has built a special simulation platform, CyberShake, which uses the time-reversal physics of seismic reciprocity to reduce the computational cost by 1000x. Additionally, the production time for a complete regional CyberShake model at seismic frequencies up to 0.5 Hz has been reduced by 10x, and four new hazard models have been run on NCSA Blue Waters and TACC Stampede. SCEC researchers have developed highly parallel, highly efficient CUDA-optimized wave propagation code, called AWP-ODC-GPU, that achieved a sustained performance of 2.8 Petaflops on ORNL Titan. LEAD: Southern California Earthquake Center Community Modeling Environment Collaboration
• Princeton University/Princeton Plasma Physics Laboratory (U.S.) Using high-end supercomputing resources, advanced simulations of confinement physics for large-scale MFE plasmas have been carried out for the first time with very high phase-space resolution and long temporal duration to deliver important new scientific insights. This research was enabled by the new GTC-P code, developed to use multi-petascale capabilities on world-class systems such as the IBM BG-Q “Mira” @ ALCF and “Sequoia” @ LLNL. Leads: William Tang, Bei Wang, and Stephane Ethier
• Oak Ridge Leadership Computing Facility, Oak Ridge National Laboratory (U.S.) Researchers at ORNL have used the Titan supercomputer to perform the first simulations of organic solar cell active layers at scales commensurate with actual devices. By modifying the LAMMPS molecular dynamics software to use GPU acceleration on Titan, the researchers were able to perform simulations to study how different polymer blends can be used to alter the device morphology. The new insights will aid in the rational design of cheap solar cells with higher efficiency. Results are published in the journal Physical Chemistry Chemical Physics. Lead: W. Michael Brown and Jack C. Wells
• Ford Werke GmbH (Germany) Researchers at Ford Werke GmbH have developed and deployed a new CAE process, which enables the optimization of the airflow through the cooling package of a vehicle using complex 3D CFD analysis. The Ford team also demonstrated it could run these complex simulations fast enough to enable their use within the time constraints of a vehicle development project. The team’s work on Jaguar at Oak Ridge National Lab will help Ford maximize the effectiveness and fuel efficiency of engine bay designs throughout the company. Lead: Dr. Burkhard Hupertz and Alex Akkerman

• Argonne National Laboratory (US). Researchers in the Computing, Environment and Life Sciences Directorate (CELS) at Argonne National Laboratory developed ACCOLADES, a scalable workflow management tool that enables automotive design engineers to exploit task parallelism using large-scale computing (e.g., GPGPUs, multicore architectures, or the cloud). By effectively harnessing such large-scale computing capabilities, engine design engineers can concurrently simulate the drive cycle of thousands of vehicles in the wall-time it normally takes to complete a single dyno test. According to experts from a leading automotive manufacturer, ACCOLADES in conjunction with dyno tests can greatly accelerate the test procedure yielding an overall saving of $500,000 to $1 million during the R&D phase of an engine design/development. Leads: Shashi Aithal and Stefan Wild
• Korea Institute of Science and Technology Information (Republic of Korea) The Department of Supercomputing Modeling and Simulation at the Korea Institute of Science and Technology Information (KISTI) runs a modeling and simulation program that offers Korean SMEs the opportunity to develop high-quality products using the supercomputers at KISTI. Through open calls, the project selects about 40 engineering projects of SMEs every year and provides technical assistance, access to the supercomputers, and the appropriate modeling and simulation software technology such as CFD, FEA etc. From 2004 to date, the project has assisted about 420 SMEs. The project recently assisted the development of a slow juicer made by NUC Corp by improving the juice extraction rate from 75% to 82.5% through numerical shape optimization of a screw by using the “Tachyon II” supercomputer and fluid/structural analysis. This helped sales dramatically increase from about $1.9 million in 2010 to $96 million 2014, and the company has hired 150 new employees through building the additional manufacturing lines. Lead: Jaesung Kim and Sang Min Lee
• Sardina Systems (Tallinn, Estonia): This new company developed a technology that automates HPC operations in large-scale cloud data centers, such as collecting utilization metrics, driving scalable aggregation and consolidation of data, and optimizing resource demand to resource availability. The product, FishDirector, incorporates high performance parallel data aggregation and consolidation, coupled with high performance solvers that continuously solve for optimal layout of virtual machines (VMs) across an entire compute facility, taking into account costs such as VM movement/migration costs and constraints around placement of certain VMs, to drive higher overall server utilization and lower energy consumption The firm states that it has demonstrated raising utilization from 20 percent to over 60 percent at one government facility by optimizing the performance of over 150,000 VMs. Lead: Kenneth Tan

The Centre for Computational Medicine, University of Toronto: From genomics to medical imaging, almost every discipline in health care is dealing with a “Data Deluge”. Translating this into something that will ultimately benefit patients requires massive amounts of computation and storage in an environment that is fast, secure, and run with optimal efficiency. The University of Toronto’s SickKids Centre for Computational Medicine uses a supercomputer operating at 107 trillion calculations per second to predict the minute differences between individual children to identify the most precise treatment possible for each child under their care.

Disney Animation Studios,  Frozen: Software engineers used advanced mathematics and physics, with assistance from mathematics researchers at the University of California Los Angeles (and 4,000 computers), to design breathtaking, believable scenes.  Tangled: This film employed a unique artistic style by blending features of computer-generated imagery (CGI) and traditional animation, while using non-photorealistic rendering to create the impression of a painting. Disney developed unique techniques and tools to makes the computer “as pliable as the pencil” to create the ultimate (and most expensive) 3D movie of all time. Big Hero 6: Walt Disney Animation Studios created new software, called Denizen, to create over 700 distinctive characters. Another, called Bonzai, was used to create the city’s 250,000 trees, and a new rendering tool, called Hyperion, offered new illumination possibilities. Disney had to assemble a new supercomputing cluster just to handle Hyperion’s intense processing demands, which consisted of over 2,300 Linux workstations in four data centers, backed by a central storage system with capacity of five petabytes.

DreamWorks Animation How to Train Your Dragon: Over the five years before the film’s release, DreamWorks Animation overhauled its production workflow and animation software. How to Train Your Dragon 2 was the first DreamWorks Animation film that used “scalable multi-core processing”, developed together with Hewlett-Packard. This “next revolution in filmmaking” enabled artists for the first time to work on rich complex images in real time, instead of waiting eight hours to see the results the next day. Programmes named Premo and Torch allowed unique subtlety, improving facial animation and enabling “the sensation of skin moving over muscle instead of masses moving together”.. Kung Fu Panda: The computer animation used in this film was more complex than anything DreamWorks had applied before. They found help through the Department of Energy’s Innovative and Novel Computational Impact on Theory and Experiment (INCITE) programme – the company was awarded a grant to refine and test its redesigned software on the leadership-class supercomputer at the Oak Ridge National Laboratory. The prototype software was successfully tested and immediately put to use. Knowledge gained from the INCITE grant informed an entirely new rendering architecture and has become essential in creating all of DreamWorks’ animated films. But the biggest win was the raw speed at which those responsible for the film’s lighting could get back frames. Iterations were sped up by an order of magnitude – a tenfold savings in total processing time. An iteration that once took hours was now accomplished in mere seconds. Monsters vs. Aliens: It took approximately 45.6 million computing hours to make this film, more than eight times as many as the original Shrek. Several hundred Hewlett-Packard workstations were used, along with a “render farm” of HP ProLiant blade servers with over 9,000 server processor cores to process the animation sequence. Animators used 120 terabytes (TB) of data to complete the film. They used 6 TB for an explosion scene. Since Monsters vs. Aliens, all feature films released by DreamWorks Animation are produced in a stereoscopic 3D format, using Intel’s InTru3D technology.

Fortissimo/Ergolines: Ergolines, an Italian small and medium enterprise (SME), is a world leader in supplying control systems to the steel industry. The focus is mainly on continuous casting, where liquid steel becomes a solid mechanical structure. Simulation of the casting process enables the design of the control systems, which allow steel-casting plans to operate at optimal levels. Without Cloud-based-HPC support, Ergolines would not have the wherewithal to perform the simulations that have led to better control systems with remarkable benefits to customers, workers, and the marketplace, such as lower costs, greater productivity, and improved safety. This was the company’s first experience with HPC and its benefits, but early results show that a total savings of up to 670,000 euro per year, per medium-size steel plant, is very achievable.

KINETICA/United States Postal Service: How does a company that makes daily deliveries to more than 154 million addresses, using several hundred thousand vehicles and employees, improve efficiencies using visualizations and analytics of real-time data? It’s called Kinetica, an in-memory GPU accelerated database, and it’s how the United States Postal Service (USPS) is optimizing its operations. Faced with the daunting task of managing the nation’s postal service, which covers a larger geographical area than any other, with increasingly limited resources, USPS needs to continuously improve safety, efficiency, and services without overspending. The complexities and dynamics of USPS’ logistics have reached all-time highs, while consumers have greater demands and more alternative options than ever before; they require sophisticated services like the just-in-time supplies, tracking, and delivery updates, and dynamic shipment routing. Improving end-to-end business process performance while concurrently reducing costs requires the ability to make fast business decisions based on live data. The USPS has accomplished this with Kinetica, optimizing operational efficiencies to save time and money.

Novartis/Amazon Web Services (AWS)/Cycle Computing: Novartis ran a project that involved virtually screening ten million compounds against a common cancer target in less than a week in 2013. They calculated it would take 50,000 cores and close to a $40 million investment if they wanted to run the experiment internally. Partnering with Cycle Computing and AWS, Novartis built a platform leveraging Amazon Simple Storage Service (Amazon S3), Amazon Elastic Block Store (Amazon EBS), and four Availability Zones. The project ran across 10,600 Spot Instances (approximately 87,000 compute cores) and allowed Novartis to conduct 39 years of computational chemistry in nine hours, all for a cost of $4,232. Out of the 10 million compounds screened, three were successfully identified.

University of Rochester Medical Center: By combining genetics, neurobiology, and supercomputing, researchers at the University of Rochester Medical Center, New York, identified a genetic mutation responsible for a potentially deadly seizure disorder found in infants and young children. Young people with the condition who survive beyond infancy often struggle for the rest of their lives with developmental disabilities, autism, and uncontrollable seizures. The researchers used a supercomputer cluster to quickly obtain a full genetic profile – more than 20,000 genes – for each study subject and to compare the results with data from other families. These findings opened up what was a “black box”, enabling researchers to more fully understand the biological pathways associated with these disorders and why some patients do not respond to treatment.