NEC User Group Highlights: the Right Climate for Petaflops

By Christopher Lazou

June 2, 2006

May 22-26, 2006, Toronto, Canada: Some sixty experts from four continents, but mostly from European countries and Japan, attended the NEC User Group, NUG-XVIII meeting, hosted by the University of Toronto, in Toronto, Canada.

Professor Dr. Michael Resch, director of the HPC center – Hoechstleistungsrechenzentrum (HLRS) in Stuttgart, and Chair of the NUG, grafted an excellent program of talks, which enabled experts in computing, meteorology, hydropower and other technical engineering fields, to share their latest research results, crystalize future hardware and software needs and collectively leverage NEC to take these onboard in their development plans for new systems. These needs are not only for faster more powerful systems for the scientific and technical market, but also in data management and storage handling, requiring petabytes-sized file systems. In short, users want an infrastructure that delivers a timely total solution to their application.

In the climate/environment field, short-term climate prediction requires 20 teraflops of performance and 10 terabytes of memory and this is currently satisfied by the Earth Simulator and the NEC SX-8. Long-term climate prediction requires around 200 teraflops and 100 terabytes, whilst forecast of local hazards requires around one petaflop and 500 terabytes. NEC described how it is putting in place a program to deliver this sustained performance on its journey toward petaflops computing, which is in line with the Japanese government program. Below are a few extracts from these talks to give a flavor of the event.

Because of global warming, the frequency of extreme events such as flash flooding are on the increase. So national weather centers are in the process of refining their local weather prediction models from coarser grids down to 1 kilometer to improve the accuracy of their prediction. To achieve this refinement in a timely fashion requires the most powerful and productive supercomputers available. At present informed opinion concurs, that the parallel vector processors in the NEC SX series, are most suited for this task. In the meteorology field, the Total Cost of Ownership (TCO) of the NEC SX system compares very favorably with PC clusters. Its productivity is not only unsurpassed, but it is streets ahead compared to any other system. In addition, its high integration and fast memory/communication bandwidth provides the capability for higher resolution global and regional numerical models, which at the same time enables the inclusion of new science and different assumptions in the models.

In Europe, several large facilities dedicated to weather and climate predictions run NEC SX systems. These include DKRZ in Germany, the UK Met Office and the Danish Meteorological Institute as well as other countries. They will soon to be joined by Meteo France, which as I understand, ordered a 21 teraflops NEC SX system.
 
In all these countries, climate modeling is an important element of their activities. They are actively involved in research for the International Panel on Climate Change (IPCC), validating scenario simulations for the IPCC AR4 conclusions expected in 2007. The results from these scenarios show that the world is going to be a much hotter place with higher sea levels by 2100, contributing to the destruction of several hundred million people due to the loss of arable land. This has serious strategic implications for long-term strategic economic planning, not only for individual countries, but also for the whole world.

Having listened to Al Gore on U.S. television, explaining to the American public the dire consequences of continuing to generate excessive greenhouse gases, one can't help but muse: “One tiny fluctuation on few voting forms in Florida may forever change the future course of the world….” (In the spirit of Lorenz, 1963).

Professor Richard Peltier, University of Toronto, gave the first keynote titled: “HPC Applications in Geophysical Fluid Dynamics.” He described the NCAR CCSM model using standard semi-spectral atmosphere and Cartesian structured grid ocean, the Jovian jet model using a triangular block-structured icosahedral grid, and a new technique of unstructured grid climate modeling using fully unstructured triangular grids. The reconstruction of past climates simulating 100 thousand years of North Atlantic climate variability was performed using the NCAR CCSM3 Model. This showed the Younger Dryas cold reversal and the Eemian last interglacial around 21 thousand years ago. The Greenland ice cores show that variability was stable until about 25,000 years ago. Between 25 and 10 thousand years ago, it became very unstable and deglaciation took place. The planet remembers circulation by the change in the ratio of isotopes. During a short period of warming, it reduced the ice by 40 percent and the water shot up north through the McKenzie river basin, stopping the “Global ocean conveyor” circulation and causing 10 degrees centigrade of cooling. The global circulation overturning the oceans is driven by the North Atlantic deepwater formation, originating in the Greenland, Iceland, and Norwegian (GIN) seas. These studies are used to verify climate models for simulation of future climate trends.

Professor Peltier summarized as follows: The UofT GFD group at Toronto University is involved in a wide range of CFD based analyses of important environmental processes. The current focus of much of the work is on the development of software, based upon unstructured grid technology. This work will be dramatically advanced with installation of a next-generation NEC SX parallel vector system.

Both NWP and climate research generate enormous amounts of data, presently over one petabyte per year, which requires archiving. Thus, weather/climate centers have heterogeneous computer facilities, with separate compute and file servers, the latter dedicated to handling the increasing data, using a unified file system.

Professor Eberhard Goede, University of Stuttgart, gave a talk on: “The beauty of engineering in hydropower.” He enthusiastically described a number of projects, where by using simulation on high productivity computers such as the NEC SX-8 they have in Stuttgart, solved engineering problems in the design of turbines, improving their running efficiency by 10 to 20 percent. He gave examples from the Itaipu, Rio Parana, Brazil, which until recently was the largest hydroelectric power station in the world. This plant is running 18 turbines using 8 meter blades, with a total production capacity of 13 GigaWatts. An even larger hydroelectric plant has been build by the Chinese, at the 3 Gorges. It uses 10 meter blades each weighing 430 tons, a total of 3,200 tons rotating at 750 revolutions per minute. It has the capacity of 10 nuclear power plants or 18 coal-fired ones. Using video sequences during his talk, he gave many real-time examples where instabilities due to turbulence were observed. Once identified, the problems were solved, reducing loss of energy and making the plant more efficient. Using mathematical simulation methods to optimize flow can often improve efficiency by 10 percent and the benefits from this improvement are enormous. He reminded the audience not to forget the power of simplification by quoting Albert Einstein: “All should be as simple as possible, but not simpler.”

Professor Satoshi Matsuoka, from the Tokyo Institute of Technology gave a talk titled: “The flight of Tsubame (swallow).” He briefly described the rationale and design decisions they made in the integration of the 100 teraflops cluster they have just installed at their Institute. This system was installed two months ago and is up and running, providing a service to 10,000 students at their campus.

Although the system main integrator was NEC, Sun Microsystems provided the Galaxy compute nodes, AMD, the Opteron CPU, ClearSpeed, the SIMD co-processor and NAGERI from Israel, the network. They used 10,480 cores in 655 nodes of Sun Galaxy (Opteron 8-way dual core), rated at 50.4 teraflops peak. This is augmented with 360 boards of low power ClearSpeed co-processors, adding another 35 teraflops peak. There are plans to upgrade this to 60 teraflops peak. They have 1.1 terabytes of memory and over one petabytes of storage from Sun and NEC. In addition they are planning to include an NEC SX-8 vector machine. The whole system uses a unified network using InfiniBand switches, currently at 2 gigabytes/sec per node, but this easily scales to 8 gigabytes/sec.

The design principles of Tsubame were based on capability and capacity. It uses a high performance low power x86 multi-core CPU with high versatility. It uses Fat Node architecture and, for vector acceleration, they opted for low power SIMD architecture, namely, ClearSpeed. For example, the 35 teraflops ClearSpeed co-processors fit in one rack and consumes 90 Kilowatts. Once the hybrid architecture was fixed, the installation took just three weeks. The system has been in operation for only two months, so it has not yet been fully stressed to ascertain whether it delivers the productivity envisaged in its design specification. The system was designed to last four years.

The Tsubame system has run HPL (Linpack) at 38.18 teraflops, making it the fastest system outside the USA. According to their tests it performs very well on HPCC and other benchmarks. The system delivers very uniform high performance and it is very scalable. For example, on the Gaussian test-397, it scaled up to 64 processors.

They are already looking to the future. Power constraint is a big issue. Improvements in vector acceleration and interconnect are other significant elements which need urgent attention. Nevertheless, after some analysis of what technologies are in the pipeline, they expect to have a one petaflop system in the 2010 time frame — at the same time as the 10 petaflops Blue Gene. Professor Satoshi Matsuoka went on to predict a desk-size one petascale system by 2016.

Mr. Jun Inasaka, NEC, gave a talk titled: “Critical technologies for future vector Architecture.” The talk was structured around three main topics: advanced CMOS LSI technologies for high speed and low electrical power, ultra high-speed transmission and ultra high-density (packaging) technologies.

Achieving low power consumption in devices is one of the most pressing problems facing HPC. Current technology is using a 65nm process, clocking at 3-4GHz and the next generation processors will be using 45nm, clocking at 4-5GHz. Chip power density is approaching that of a nuclear reactor and this trend is obviously unsustainable. Power density has become the biggest constraint for increasing performance, so several new innovations were incorporated in the design. For example, strained silicon transistors enhance performance without leakage penalty. Using multi-vt techniques allows threshold voltage optimization, minimizing leakage. The use of low-k interlayer dialectric decreases power requirements and by using high-k oxide it reduces leakage through gate dialectric.

Intra-node optical transmission is needed to deliver bandwidth for petaflops systems. One approach for implementing the intra-note interface for HPC is to use multi-tap pre-emphasis equalizer for optical asynchronous transmission. The requirement is for 1,000 channels with very high transmission. The challenges are manifold. For example, the reduction of attenuation and waveform distortion to achieve low jitter during transmission and reception at high speeds, that is, at speeds above 10 Gbps. Signal integrity requires high fidelity transmission cable to control impedance characteristics, minimizing transmission noise. To ensure power integrity supply noise, which may cause some deterioration of I/F link, speed must be suppressed. Noise increases non-linearly, by a factor of 10 for high speed CPUs, so the challenges are non-trivial. Nevertheless, the optical interface is expected to deliver up to 30 Gbps per channel by 2010. This optical interconnect is necessary for future systems.

Packaging is an integral and essential part of customizing the new technologies into the next generation devices, which are expected to deliver one teraflop on a chip. These chips are not likely to be air-cooled. New liquid cooling technology is developed with micro-channels for heat-sink and other enabling innovations needed to handle devices for petaflops computing.

The details in Mr. Inasaka's talk were very impressive and convinced this writer that NEC is well on its way towards petaflops systems with vector parallel architecture playing a pivotal part. Technology strategy overviews from other NEC executives were also given explaining how their deep technologies as microprocessor and telecommunications vendors is leveraged, to develop HPC future products. In addition, there were many interesting user talks, illustrating fascinating results from supercomputing centers. All in all, the meeting in Toronto, like its predecessor meetings, was very fruitful. The user community not only shared their own research and operational experiences, but also had the opportunity to leverage technology from all of NEC's businesses for their own benefit. For more detailed descriptions of the presentations, visit www.nec-ug.org.

Thomas Sterling, as always, gave a thought provoking keynote and lamented at the failure of qualitatively improving supercomputer systems, saying that in recent years, despite the hype about exceeding Moore's Law, we are failing to take advantage of the two to three orders of magnitude more performance available with current technologies. He claims this can be achieved by introducing not even radical modifications to current computer architectures. He went on to describe the current state of supercomputer hardware, the difficulties of increasing clock frequency, the problems of cooling multi-cores and the inherent challenges of increasing programming complexity. He stated that NEC leads in high end computing but suggested that the SX-8 can be improved with a co-processor to do process management. A more comprehensive exposition of his views and vision of supercomputing will be contained in my interview of Thomas Sterling, soon to be published, so watch this space.

—–

Copyright (c) Christopher Lazou, HiPerCom Consultants, Ltd., UK. Brands and names are the property of their respective owners.

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!

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…

Nvidia Appoints Andy Grant as EMEA Director of Supercomputing, Higher Education, and AI

March 22, 2024

Nvidia recently appointed Andy Grant as Director, Supercomputing, Higher Education, and AI for Europe, the Middle East, and Africa (EMEA). With over 25 years of high-performance computing (HPC) experience, Grant brings a 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…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the fi 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