Oak Ridge Dives Into Science at the Petascale

By Leo Williams (Science Writer, ORNL)

November 19, 2008

The petascale age is here. After years of predicting the scientific advancements they would be able to make with petaflop supercomputers capable of a thousand trillion calculations each second, researchers now have an opportunity to prove their point. Oak Ridge National Laboratory (ORNL) recently unveiled the first petascale system dedicated to scientific research, a Cray XT machine with a theoretical peak performance of 1.64 petaflops.

This behemoth — an upgrade to ORNL’s Jaguar system — comprises more than 45,000 quad-core AMD Opteron processors. It boasts an unprecedented 362 terabytes of memory, which is three times more than any other system, a 10-petabyte file system, 578 terabytes per second memory bandwidth, and input/output bandwidth of 284 gigabytes per second. We talked with Doug Kothe, director of science at ORNL’s National Center for Computational Sciences [NCCS], about the challenges of and potential breakthroughs in science now possible with this built-for-science petascale system.

HPCwire: ORNL’s upgraded Jaguar will be the first petascale supercomputer designed for and dedicated to open scientific research. What are the immediate plans for putting this system to use?

Doug Kothe: The current plan is for the system to be used during much of the coming year for specific high-impact projects of national importance. In addition, it will continue to support the INCITE program [Innovative and Novel Computational Impact on Theory and Experiment program, sponsored by the Office of Advanced Scientific Computing Research (ASCR) in the Department of Energy’s (DOE’s) Office of Science]. This first group is known as Transition to Operations, or T2O, projects. They are tackling science problems — both applied and fundamental — that cannot be solved without Jaguar’s speed, memory, and infrastructure. We have been working closely with ASCR and with members of the computational science community to identify projects that have application software that can effectively utilize a large fraction of the system.

We expect these projects to deliver important results. Since they will be led by the community’s most sophisticated users and prominent scientists, early simulations on Jaguar will also help us harden the system for a broader collection of projects later in the year.

The selection of science problems for early access to the petascale system is by no means finalized. Computational researchers who believe they can fully exploit this system to deliver far-reaching results should contact us via the Web. We have three principal goals during the system’s early phase: deliver important, high-impact science results and advancements; harden the system for production; and embrace a broad user community capable of and prepared for using the system.

HPCwire: Will specific science domains have precedence?

Kothe: We are looking at all research areas that are important to DOE’s mission, from energy assurance to climate-change science to more basic fundamental and applied science. The breadth and depth of critical science potentially solvable on this system are daunting, with domains including fusion, biology, atomic physics, chemistry, nuclear energy, materials and nanoscience, climate and geosciences, astrophysics, high-energy physics, turbulence, and combustion. And this is not an exhaustive list.

HPCwire: Can you give us some idea of the kind of results we can expect?

Kothe: Sure. Looking at climate studies, we hope to be able to say with increased confidence just how good global models will be at predicting regional climate change on the scale of decades. We should also be able to better predict the likelihood of abrupt climate change — change taking place over decades rather than centuries — and the potential for increasingly destructive storms around the world as the climate gets warmer.

As I mentioned before, energy assurance is extremely important to us as a DOE lab, and we will be looking at energy production, storage, and transmission from a variety of angles. We expect to see new insights into the physical properties of biomass that will help us overcome the technological impediments to mass cellulosic bioethanol production. We expect to make significant progress in understanding and controlling the core plasma turbulence that will exist in the ITER fusion reactor. And we expect to dramatically improve our understanding of what happens inside the core of a nuclear fission reactor by removing many of the simplifying assumptions and estimates that had previously been unavoidable in modeling neutron transport.

Other areas being investigated could ultimately affect how we as a society produce and use energy. We will be looking for significant new insights into electrical energy storage involving, for instance, the storage and flow of energy in carbon nanostructured supercapacitor systems. Advances in this area are important both to mobile devices and to the viability of renewable energy resources — such as solar and wind power — that must be stored and transported. We are working to embrace the energy storage community, and we currently have an exciting project committed to going after this challenge on the Jaguar petascale system.

We will also be seeing first-principles studies of strongly correlated materials such as those often found in magnets and superconductors. If we can understand with confidence the effect of disorder on superconducting transition temperatures, we can revolutionize energy transmission, transportation, and a number of other areas. High-temperature superconducting cables, for instance, will be able to carry electricity indefinitely without any loss.

HPCwire: What other areas are being targeted during the early phases of the Jaguar petascale system?

Kothe: There are many other areas. In biology we hope to see the first accurate microscopic structural description of the dynamics of water. This will be indispensible as we move forward to atomic-scale biological simulations. And we will continue to play a major role in computational astrophysics research. For example, simulations of binary black holes and the gravitational radiation they emit will support both current and future projects aimed at detecting gravitational waves. And we will be looking at the first realistic model of the closest supernova in nearly 400 years — SN1987A. These simulations will make quantitative predictions of key observables associated with core-collapse supernovas, including element synthesis.

HPCwire: Does industry fit into your plans?

Kothe: Yes, very much so. The INCITE program has been very successful in attracting companies to perform large-scale simulation science on ASCR systems such as Jaguar. At ORNL, for example, I have worked closely with industry projects involving Boeing and General Motors. What I’ve seen is that these companies bring very talented researchers to the table with very challenging, compelling problems. Their problems are not easily simulated, and for the most part they demand scalable application tools just like DOE and academic projects do. To borrow from the Council on Competitiveness, industry must out-compute to out-compete. I firmly believe that statement is right; hence, our role with U.S. industry is to work with them in delivering science results that help them become more competitive. Given today’s economy, it is imperative that we focus all the more on helping these companies gain a stronger foothold.

HPCwire: The computational science community has been anticipating computers capable of a petaflop or greater for some time. How will the research performed on these systems differ from that done on earlier systems?

Kothe: A decade ago the game for people who wanted to do scientific research aided by computer simulation was simplify, simplify, simplify. We weren’t able to easily solve coupled nonlinear systems, so we would uncouple and linearize them to give us something we could solve. In those days you had to argue that these simplified models described reality, but more often than not they really didn’t at the level needed for predictive accuracy.

In contrast, the mindset today is very different; researchers no longer see the computer as a restraint. Young scientists don’t realize how great they have it. In fact, there is almost nothing out there that we can’t at least think about modeling, if not on current systems then one or two generations down the line.

HPCwire: So is this system going to be too difficult to use for scientists and engineers who have never been engaged in “big computing”?

Kothe: We don’t think so. We have already run at least a half-dozen simulation tools at scale on this system, and it’s still in its infant stage, just seven weeks after the last cabinet arrived. The performance of these applications, measured by raw sustained compute speed and parallel efficiency, is impressive.

This early evidence and our optimism are based on two simple facts. First, Jaguar’s hardware and software environments use the same programming model as before for users and developers. For them there are no drastic changes. The operating system is Linux-based, and the integrated development environment of compilers, debuggers, performance tools, and the like are unchanged. Existing scientific application software doesn’t have to be redesigned, refactored, or rewritten just to execute on the system. In retrospect, the seamlessness of the transition from that perspective was frankly surprising.

Second, Jaguar is a well-balanced system, designed for the targeted science applications and well matched to them. The AMD Opteron processor, for example, is a great chip for science: It is fast, has great memory and intersocket bandwidth, and is easy to program, since it uses the same X86 instruction set we have used for years. Similar examples exist in the interconnect and I/O infrastructure. The total memory on the system is incredible — more than three times any other system.

HPCwire: Why is memory so important?

Kothe: Without sufficient memory, scientists must oversimplify assumptions or run at resolutions so low they miss important characteristics. For example, global climate simulations do not produce hurricanes if the resolution is too low. More memory in systems such as Jaguar means more space for additional information about the simulation model, such as more model equations and more complicated model equations. Generally the ability of a simulation to match reality is directly correlated with that simulation having adequately complex models. And the list goes on. We’ve gone out of our way to ensure that Jaguar adequately addresses application requirements. In fact, we’ve documented our requirements collection process, data, and analysis in a number of recent reports, available here.

HPCwire: How have the challenges to using these systems grown?

Kothe: As I said, the programming architecture for the petascale Jaguar is very similar to earlier versions of the system. Current Jaguar users will have to optimize their codes for the new system, but they won’t necessarily have to redesign their software and algorithms.

That having been said, world-class supercomputers have always been a challenge to use, and new systems require far more parallelism from the codes running on them than we’ve ever seen before. I can remember when a simulation on 512 processors was considered massively parallel, but then again I’m not an “early-career” researcher. But now we’re working with systems that have hundreds of thousands of processing cores, and that number is only climbing.

Of course, we realize that not all of our users will be supercomputer experts. That’s why we have a comprehensive, that is, multileveled support system with a proven track record that others are emulating. Each major project at the NCCS has a scientific liaison assigned from our Scientific Computing Group. This is a group of mostly PhD-level computer scientists and domain scientists who are experts at taking important scientific questions and translating them into effective supercomputing applications. These folks also have productive research accomplishments and careers in their own right; in short, they are on top of their game, which makes them especially adept at being useful members of the NCCS project teams.

The challenge as these systems grow is to exploit the memory and processor hierarchy that we’re seeing in current and next-generation computing nodes. For the foreseeable future, what we see in computing nodes is a hybrid architecture. They’ll have two or three different types and levels of memory accessible in different ways. Heterogeneous architectures with floating-point acceleration, like the Los Alamos RoadRunner system, are also likely to stay. The challenge will be to have your application easily know that a particular processor or memory is different from another and respond accordingly.

We’re also going to have to build more robustness and fault tolerance into applications. The more processors you have, the more likely it is that one or more used by your application will go down during the course of a run. Currently, almost all applications need to halt and restart from the last saved state if a node or collection of nodes falls out. We need to program applications so that they are able to keep going.

It won’t be easy. We don’t have more fault tolerance now because it’s hard to program. It’s like having to change a flat tire while the vehicle is still moving.

HPCwire: Are you going to be able to handle all that data?

Kothe: This is a major consideration. We believe we’re well prepared for the input, output, processing such as analytics, knowledge discovery, and visualization, and transfer of data our scientific applications require. We expect to generate over 5 petabytes of new data just during this early science period, which is purported to be more than double the data embodied within all U.S. academic research libraries. That’s a lot, and it will be created over a period of just several months. Similar requirements are coming, for example, from the climate community in supporting their IPCC AR5 simulations [for the Fifth Assessment Report of the Intergovernmental Panel on Climate Change]. Standing up the I/O infrastructure to accommodate these data requirements is an incredible accomplishment, and we believe we have people with the talent and experience to actually pull this off. Without this data infrastructure, Jaguar and the scientific applications running on it would be effectively useless. Simulation-based science is data-intensive and data-driven.

HPCwire: What do you see for computational science in the longer-term future?

Kothe: I think we’re going to see large-scale computer simulation in areas that may seem strange today. We’ll see simulations of human behavior and social networks. We’ll see more sophisticated and more valuable simulations of biological systems; so instead of a chain of molecules, we’ll be able to simulate full cells, organs, and even individuals. We’ll see systems of systems; for example, instead of one nuclear reactor, we’ll see an entire nuclear fuel cycle. We’ll see first-principles-based simulations at larger and larger length scales and over longer and longer time. We’ll see such rapid turnaround on simulations that complex nonlinear optimizations will become commonplace. We’ll see materials and chemical catalysts by design. We’ll better understand the complex biogeochemical cycles that underpin global ecosystems and control the sustainability of life on Earth. We’ll see the deciphering and comprehending of the core laws governing the universe. Potentially, this will all happen in our lifetimes.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire