insideHPC » HPTC

HPC gives scientific computing on your cell phone a boost

John West — Thu, 19 Aug 2010 14:23:48 +0000

Reader Jay Blair sent me a pointer to this story from TACC about an Android app that runs a reduced model locally on the cell phone based on results computed over a long series of runs on Ranger.

The team performed a series of expensive high-fidelity simulations on the Ranger supercomputer to generate a small “reduced model” which was transferred to a Google Android smart phone. They were then able to solve problems on the phone and visualize the results on the fly.

The project proved the potential for reduced order methods to perform real-time and reliable simulations for complicated problems on handheld devices.

This approach is already used operationally in a variety of civilian and defense scenarios to allow professionals ranging from bridge fatique assessment teams to rapid crisis response forces to tradeoff some accuracy for an answer right now. Typically these reduced models have run on laptops or larger portable computers, but today’s mobile devices are becoming quite powerful in their own right.

This is not the first time that model reduction algorithms have been used to ameliorate the complexities of large-scale physical simulations. The advantage of the system designed by Knezevic and his colleagues is its rigorous error bounds, which tell a user the range of possible solutions, and provide a metric of whether an answer is accurate or not. The error bounds are based on mathematical theory developed in Prof. Patera’s research group at MIT over a number of years.

“We have a bound on how much accuracy we’re losing with our reduced model, so we can say with rigor that we’re doing supercomputing on a phone,” Knezevic said.

The quantitative understanding of the error bounds is very important, and its a nice addition in this work.

PRACE awards 320 million hours to 10 projects

John West — Fri, 30 Jul 2010 14:55:14 +0000

This week PRACE announced a big award of computational time to 10 projects in Europe

Ten research projects, five from Germany, two from the UK one each from Italy, the Netherlands and Portugal, have been awarded access to the PRACE infrastructure. In total 321.4 Million compute core hours were granted. Sixty-eight applications requesting a total of 1870 Million compute hours were received in this call, which was the first opportunity for researchers to apply for PRACE resources.

The successful research projects are in the fields of astrophysics, earth sciences, engineering, and plasma and particle physics including collaborators from 31 Universities and research institutes in 12 countries. These projects will have access to JUGENE, IBM BlueGene/P, hosted by the Gauss-Centre for Supercomputing member site in Jülich, Germany, which is the first Petascale HPC system available to researchers through PRACE. It is the fastest computer in Europe available for public research.

These projects are all R&D, and all science (not industrial). The projects awarded time include

Simulation of electron transport in organic solar cell materials
Excess proton at water/hydrophobic interfaces: A Car-Parrinello MD study
Parallel space-time approach to turbulence: computation of unstable periodic orbits and the dynamical zeta function
QCD Thermodynamics with 2+1+1 improved dynamical flavors
Ab initio Simulations of Turbulence in Fusion Plasmas
Providing fundamental laws for weather and climate models
Plasmoid Dynamics in Magnetic Reconnection
A dislocation dynamics study of dislocation cell formation and interaction between a low angle grain boundary and an in-coming dislocation
Type Ia supernovae from Chandrasekhar-mass white dwarf explosions
QCD Simulations for Flavor Physics in the Standard Model and Beyond

PRACE announces call for allocation requests

John West — Wed, 16 Jun 2010 17:39:26 +0000

Continuing with our “PRACE day” theme, let’s pass along news that I got over the email transom today about PRACE’s first call [PDF] for allocation applications. This call covers JUGENE, the IBM Blue Gene/P installed at the Gauss Centre. Allocations are due in August and will run from Nov 1 2010 to Oct 31 2011.

The deadline for submission of proposals is August 15th at 16.00 CEST. All proposals must be submitted via the PRACE website at:
http://www.prace-project.eu/hpc-access

The First PRACE Regular call is intended for large-scale projects of high scientific quality and for which a significant impact at European and international level is anticipated. High scalability of the code (at least 8000 compute cores) must be demonstrated. Proposals for project access must be ready to run. The projects must demonstrate scientific excellence and should cover topics of major relevance for European research. They should also include elements of novelty, transformative aspects, have a recognised scientific impact and include a dissemination plan. Possible practical and timely applications resulting from the project are also desirable. The projects should also demonstrate the possibility of achieving results which will be publishable in journals of recognised scientific impact.

All proposals will undergo PRACE technical and scientific assessment. The assessment procedure will adhere to the PRACE principles of peer review. For this call only proposals from academia are eligible and the project leader must be homed in a European Union country or a PRACE AISBL country. The JUGENE at GCS at Jülich also has further restrictions due to US export rules. All applicants should expect to be notified of the outcome by the end of October 2010.

Blue Waters will not run AIX

John West — Tue, 23 Mar 2010 14:08:08 +0000

It isn’t too much of a surprise that Blue Waters, NCSA’s 10+ PFLOPS super expected to come online in 2011, will run Linux. According to the November 2009 list, 89% of the Top500 (or 446 systems) use some variant of that ubiquitous OS. But it is perhaps a little more surprising that the system, a POWER7-based IBM that grew out of DARPA’s HPCS program, will not run AIX.

Staff at NCSA and IBM evaluated both the open-source Linux and IBM’s AIX as potential operating systems for Blue Waters, extensively testing both on identical hardware at large scale. SARA Computing and Networking Services in the Netherlands and the U.S. National Center for Atmospheric Research (NCAR) shared their experiences and provided access to their large IBM POWER6 systems, both with more than 3,300 cores and both using Infiniband interconnects. Both ran IBM’s full HPC Software stack and used the GPFS file system. SARA’s system uses Linux, while NCAR uses AIX.

The team, led by HPC rock star Bill Kramer, looked at “more than 160 features” of both Linux and AIX that are expected over the five year operational life of the system before making a decision, and ran a bunch of application performance tests as well.

NCSA also conducted multiple application performance tests, including MILC, PARATEC, four of the NAS Parallel Benchmarks (Fast Fourier Transform, Conjugate Gradient, Lower-Upper symmetric Gauss-Seidel, and MultiGrid) for Class B and Class D problem sizes, HPC Challenge Benchmarks, and parallel 3D Fast Fourier Transform tests.

…”We determined that either OS could ably meet the requirements for Blue Waters and serve the science and engineering research community,” Kramer said. “Linux offered some additional advantages, such as being somewhat more familiar to the research community and offering the potential for NCSA to more directly participate in enhancing the OS.”

But which distribution, you ask? Well, I asked too, and was told that that discussion is still underway.

IBM POWER7 system at Rice offers free cycles for medical research

John West — Tue, 02 Mar 2010 15:30:43 +0000

Chron.com, the online presence of The Houston Chronicle, reported yesterday on a new super to be installed at Rice University. The system is one of the early POWER7-based systems shipped by IBM

IBM has given Rice one of its first supercomputers with the company’s new POWER7 microprocessors, a $7.6 million award, on the condition that Rice open up access to the computer to all in the Texas Medical Center. For free.

This effort builds on earlier collaborations between TMC and Rice, and peaks out at about 18 TFLOPS

When combined, Rice’s ‘Blue BioU’ system is accordingly 18 times more powerful than Deep Blue. It doubles Rice’s existing supercomputer capacity.

Just one planned use? Genomics

Baylor College of Medicine is participating in a federal project to ascertain the genetics of 20,000 tumor samples from 20 different cancers….But Baylor doesn’t want to just store the information, it wants to cross reference each of these 100 DNA sequences with the originally sequenced version of human DNA to see which gene mutations might contribute to ovarian cancer.

So they’ll try to run the analysis on Rice’s new machine. Worley expects it will take hours, when it now takes weeks.

Companies, researchers encouraged to apply for Blue Waters time

John West — Wed, 24 Feb 2010 13:38:28 +0000

This from NCSA this week

The National Science Foundation is currently soliciting proposals from researchers who are interested in tackling challenging science and engineering problems using Blue Waters, a sustained petascale supercomputer that will come online in 2011 at the National Center for Supercomputing Applications. Time on Blue Waters will be awarded through the National Science Foundation’s Petascale Computing Resource Allocation (PRAC) program.

The PRAC program is open to universities, government laboratories, and industry; proposals are due Mar 17, 2010. Access to Blue Waters will be free for non-proprietary research; those wishing to keep their results to themselves have to pay

Companies may also join NCSA’s Private Sector Program. The program offers opportunities to perform proprietary research on Blue Waters, as well as other work on Blue Waters and on other projects. Computing time through the Private Sector Program is available on a cost-recovery basis.

You’ll need to know what you are applying to use, so here’s a rundown

Blue Waters will have a peak performance of 10 petaflops, a performance that can be attained only with more than 300,000 compute cores, and a peak memory bandwidth of nearly 5 petabytes/second. Blue Waters will have more than 1 petabyte of memory, 10 petabytes of user disk storage, and 500 petabytes of archival storage. The interprocessor communications network will have a bisection bandwidth far greater than what is available today, greatly facilitating scaling to large numbers of compute cores. The high-performance I/O subsystem will enable the solution of the most challenging data-intensive problems.

FSU adds 3Leaf systems to HPC configuration

John West — Fri, 19 Feb 2010 15:55:53 +0000

3Leaf Systems, a relatively new company whose hardware can be used to create virtual SMPs, announced this week that Florida State University’s Department of Scientific Computing has signed up to install the 3Leaf Systems Dynamic Data Center Server into their HPC environment.

“As an organization that supports true multidisciplinary research, we provide a general access computing platform that must meet a diverse set of application requirements for our user community,” said Jim Wilgenbusch, Director of HPC, Research Associate in the Department of Scientific Computing of FSU’s shared-High Performance Computing facility. “3Leaf Systems is meeting our requirements for performance, flexibility and cost. With this platform we can add, remove, divide up, resize or reallocate resources across traditional hardware boundaries and seamlessly accommodate a variety of programming paradigms and application requirements. This provides us with a level of operational agility that we have not previously experienced.”

As far as I know, this is the company’s first (publicly-announced) HPC datacenter deployment. If you know different, drop in a comment.

Ames expects Obama budget will mean growth

John West — Mon, 08 Feb 2010 16:00:07 +0000

A story in the San Jose Mercury News last week cited officials confident that NASA’s Ames Research Center, located in Silicon Valley, will benefit from the proposed FY11 budget

“We’re highly confident that we’re going to get more money, and we do believe it’s going to enhance employment here, and in Silicon Valley,” said Lewis Braxton III, deputy director of Ames Research Center in Mountain View. He said it is too early to know how many additional local jobs would be created by NASA’s proposed new emphasis on unmanned space missions, climate change research, an extension of the International Space Station and plans to partner with private companies to launch astronauts into orbit.

The potential boon may extend to Ames’s HPC center as well

…Braxton said he could imagine more work for Ames’ supercomputer facility and additional partnerships, such as the center’s cooperation with Google, in which NASA has tapped Google’s expertise to help make its spacecraft more autonomous, and building on partnerships, such as one with Cisco Systems, to integrate the world’s climate change databases.

NASA goes further with its computational portal project

John West — Tue, 02 Feb 2010 16:04:25 +0000

Parabon announced yesterday that NASA has extended the small business grant it awarded to Parabon Computation last January. That project was for a portal to let NASA’s users run scientific codes from their browser. The new effort is a two year, $600,000 project

As part of the space agency’s highly competitive Small Business Innovation Research (SBIR) award program, Parabon will deliver a first-of-its-kind software service that enables scientists and engineers to interactively develop, execute and collaborate on modeling and simulation (M&S) applications from any standard web browser.

Built upon Parabon’s Frontier Grid Platform, which manages some of the largest computational grids, the new Modeling and Simulation as a Service (M&SaaS) solution will provide web-based Platform as a Service tools. These tools – such as a browser-based source code editor, online collaboration utilities, and virtualized build and runtime environment management interfaces – will allow developers to more efficiently create and modify a wide variety of high-performance computing (HPC) applications. In addition, the web-centric nature of the project will allow researchers around the world to work together seamlessly, removing barriers that have heretofore hampered scientific collaboration, dramatically increasing productivity for NASA and other organizations.

ORNL researcher proposes to use supers to identify health care fraud

John West — Mon, 11 Jan 2010 19:29:16 +0000

Government Computer News carried a story last week about a proposal from a computer scientist at Oak Ridge to put Jaguar to use processing — or at least analysing — healthcare claims looking for fraud

Combining and analyzing health care data in real time could save as much as $50 billion a year by eliminating waste and preventing fraud in government-run health care programs, and also could improve the quality of medical care, said Andrew Loebl, a senior researcher in the lab’s Computational Science and Engineering Division.

“We have never put all of this data together,” Loebl said. “My idea is to use the storage capacity of the supercomputers at Oak Ridge to analyze the data.”

$50B is a lot of money, and if he can get anyone in the fairly staid healthcare bureucracy to listen to him (no mean feat), he may get a taker. As a citizen I don’t like all that data being in one place, but then again if my taxes went down as a result it might be worth it.

Currently, the government uses five regional contractors to process claims for a variety of health care programs, including Medicare, Medicaid, and programs run by agencies including the Defense and Veterans Affairs departments, Indian Health Services, the Federal Employee Benefits Health Plan and others. The data is disaggregated for processing, and broken down geographically and to allow processing with limited capacity computers. Consequently, no one sees or understands all of the data

….That is the hurdle faced by the proposed program. Convincing agencies to combine traditionally siloed data into a single flow for processing on a single computer could be tough.

Let’s assume we can get around privacy concerns; despite Loebl’s assertion that real-time processing of all healthcare claims in the federal government wouldn’t interrupt science…

The processing would not interrupt the climate modeling or other advanced research being done on Jaguar, Loebl said.

…I’m not sold. The missions are fundamentally different — accounting vs. science — and you run a production enterprise accounting resource differently than you run a science resource. Those conflicts in operational status would inevitably cause choices to be made that would probably favor the dollar today versus the discovery tomorrow (I’ve been there, done that).And for my part I’m suspicious that the work is as low effort as Loebl claims.

Even if it is, then one set of answers will lead to a new set of questions, and the healthcare problem with its immediate payoff in actual dollars will eat up Jaguar with its longer-time science mission. It’s a bad idea to cross the streams.

Better to propose a new “Health Care Fraud Prevention Center” and house it in the acres of machine room space they have up there directly connected to the TVA than to mix the missions.

As Blue Waters gets closer, Daily Illini reporting some tension over costs

John West — Wed, 09 Dec 2009 16:27:19 +0000

Yesterday the Daily Illini, the student newspaper of the U of Illinois (home organization to NCSA and the Blue Waters project) posted an article about Blue Waters. I think the article does a reasonable job of reflecting the conversation that always goes on around large HPC projects with questions of costs and benefits

“Blue Waters will be used to do scientific research that relies on supercomputers,” she added. “It might be doing better weather prediction. It might be looking at how the universe evolved in the very early days, and it might be trying to find an answer to global warming. It will be, in 2011, the most powerful supercomputer in the world.”

And then the costs, which I don’t think I had seen brough altogether like this before

The computer itself will cost $208 million to build and will be funded through a grant from the National Science Foundation, or NSF. The cost of housing the supercomputer is $72 million; the University will pay $12 million and the state will pay $60 million, Barker said.

The U of I is getting a LOT of leverage on Blue Waters (set to come online next year), and it’s contribution to the university is certain to outweigh the $12M the university is carrying on capital expenditure for the deal. But the university is evidently having a budget crisis, and the $12M had to come from something

But with what administrators have called a “budget crisis” looming in the University’s near future, some have brought to question the morality of the University spending this much on a computer.

…“We need to question very seriously what the priorities of the administration are,” he said. “When they are telling us that they have to furlough staff and faculty, that they are freezing all hirings and that there is not enough money to efficiently fund instruction on campus, then we need to know the benefits of these high technology programs.”

Fair enough. Too bad Randy Kangas in the university’s planning arm isn’t helping the situation with answers like these

Kangas said the University will not be responsible for funding the operational cost of Blue Waters.

“There will be additional funds that come through for research and operations, at least we hope,” he said. “There’s also a consortium with a lot of other universities, and you would hope that there will be corporate interest in what a facility like this can do.”

“Not responsible…” followed by “…at least we hope.” Hope is not a strategy. I’m sure that the U of I has a funding strategy, and I don’t know why he didn’t just articulate it. Then there was this bit of helpfulness

“If you had to make the decision today, I don’t know, maybe the provost and the chancellor would make a different decision, but it’s a very important project for campus,” Kangas said. “You have to look at it from two points, not just the dollars, but what it will do academically for the campus.”

Which was made worse by the NSF, an organization that should be really good at dealing with this line of inquiry

“We (the NSF) are in the business of funding things that sometimes, at first blush, the practical implications for which are not immediately recognized,” Zgorski said. “It’s (Blue Waters) building for the future and providing the computational resources to attack the really complex, daunting problems we are facing in society.”

Sigh.

She’s right of course, but what a bad way to put it. She’s responding to an articulated concern about jobs and financial security, so why not focus on all the jobs the construction created, or the faculty and grad student research funding the new computer is expected to facilitate?

Cray launches Euro exascale research initiative

John West — Wed, 02 Dec 2009 15:40:32 +0000

Cray announced today from the Cray Executive Forum Europe in Frankfurt that they are launching a new exascale research effort

As the first company to design and build a supercomputer that achieved sustained application performance of more than one petaflops (quadrillion mathematical calculations per second), Cray is committed to the research and development of new technologies necessary to achieve exaflops computing. The challenges are significant and will require research and development into power and cooling infrastructure, system and application resiliency, lightweight communication, efficient processor and network architectures, and new programming models.

“We are very excited to be partnering closely with the European HPC community in kicking off this important initiative for the company,” said Peter Ungaro, president and CEO of Cray. “Reaching and surpassing the petaflops barrier was an extraordinary achievement and these systems are providing unsurpassed supercomputing resources for meeting significant scientific challenges. We know there are scientific breakthroughs in important areas such as new energy sources and global climate change that are waiting for exascale performance, and we are working hard on building next-generation supercomputers that will be capable of it.”

The effort is focused on Europe, with partners at big HPC customers EPCC (at The University of Edinburgh) and at the Swiss National Supercomputing Centre (CSCS).

On Dec. 3, The University of Edinburgh will announce the creation of its Exascale Technology Centre at EPCC. Peter Ungaro will speak at the official launch of this centre, which will be a joint investment of Cray and the University of Edinburgh and which forms part of Cray’s Exascale Research Initiative in Europe. The other initial partner within this Cray research initiative is CSCS. Cray will collaborate with CSCS in the framework of the HP2C project, which targets the development of future applications for very large scale simulations (see www.hp2c.org). Cray plans to launch new strategic alliances in the future to continually grow and expand this important research initiative.

Its interesting to me that this effort is focused in Europe, where I’ve remarked before on the high degree of pan-European planning and effort taking place in apex scientific computing.

Strathclyde launches new super in Scotland

John West — Fri, 27 Nov 2009 15:30:37 +0000

HPCwire pointed us to this bit of news that we missed from earlier this month about a new super on the west coast of Scotland.

The £500,000 High Performance Computer (HPC) will help the University’s Faculty of Engineering and Institute for Complex Systems tackle some of the most challenging engineering problems, from re-imagining aeroplane design for the best fuel efficiency to working out how to store hydrogen in nanoporous materials.

Project leader Professor Jason Reese of the University’s Department of Mechanical Engineering, said: “Our new HPC – and a newly appointed HPC Officer to support its users – will be central to a wide range of pioneering research projects, including the simulation of fluids at the nanoscale, the prediction of welding distortion, and the aerodynamics of future space re-entry vehicles.

…With 1,088 computing elements (cores) writing to a 100TB high performance disk storage area across a state-of-the art Quad Data Rate Infiniband network, the HPC is designed for a peak performance of almost 13 TeraFlops – or thirteen thousand billion mathematical operations per second.

Russian President Dmitry Medvedev attends supercomputer dedication

John West — Fri, 27 Nov 2009 14:43:32 +0000

T-Platforms announced this week that they’ve delivered a 420 TFLOPS super (named Lomonosov) to Moscow State University, and that the dedication was attended by Russia’s President Dmitry Medvedev.

Equipped with a new supercomputer, MSU has become one of the few universities worldwide to house and utilize a system of such scale. Choosing a site for the Russia’s largest computer was easy: for decades MSU leads scientific collaboration across the CIS on fundamental and applied research in energy, transportation, medicine, aerospace, nanotechnology and more. Although MSU already hosts a number of HPC systems, the ever growing volume and complexity of research quickly push their capabilities to the limit. Adding the superior simulation power to the route of exploration and discovery will move science at MSU to a new level, helping to tackle previously unattainable problems.

Lomonosov has propelled Russia into the leaders of the worldwide supercomputer industry. For the first time ever a made-in-Russia system has skyrocketed to the 12th position on the Top500 list of the world’s most powerful computers, and Russian-based T-Platforms has become one of the five developers of the largest systems on the list.

The press release makes a lot of hay about that position on the Top500, rightfully so. I love that the presso differentiates between the “peak” and “real” performance (realized LINPACK performance).

CSIRO lights up cluster, drops in GPUs

John West — Fri, 27 Nov 2009 14:00:16 +0000

This is being reported in a few places, but I found mine at NetworkWorld.com. We write about the Commonwealth Scientific and Industrial Research Organisation (CSIRO) in Canberra from time to time. This time the occasion is the launching of their new system, a 61,440 core system (this count includes GPU cores, which we haven’t exactly agreed should be counted in the core count, but whatever)

CSIRO Computational and Simulation Science leader, Dr John Taylor, said the Nvidia-based GPU technology gives the organisation the potential to deliver huge gains in computational efficiency, energy efficiency and scientific research.

…Under the hood:

128 Intel Dual Xeon E5462 compute nodes (a total of 1024 2.8GHz compute cores) with 16GB or 32GB of RAM, 500GB SATA storage and DDR InfiniBand interconnect

64 Tesla S1070 (200 GPUs with a total of 48,000 streaming processor cores)

128 port DDR InfiniBand Switch expandable to 144 ports

80 Terabyte Hitachi NAS file system

It looks like they are approaching the project with a good bit of realism though, which is good

A major barrier to be overcome, however, is moving the CSIRO scientists’ code over onto the new GPU cluster, and having the approach to coding changed.

“The key issue at the moment is you can’t take an application that’s running on a multi-CPU core system and run it on the GPU,” Taylor said. “You’ve got to change the code, which is very typical.”

Brown deploys new 14 TFLOPS IBM system

John West — Tue, 24 Nov 2009 14:50:51 +0000

Late last week Brown University and IBM announced a new super at Brown’s Center for Computation and Visualization

Examples of the wide areas of research that will take advantage of the increased computing horsepower include advances in genomics that could lead to drugs for treating specific diseases such as cancer; investigation of the mechanics of human and animal movement; exploration of the web of animal life and ocean ecosystems; and studies of the terrain of planetary bodies, such as Mars.

The presso refers to this as “a system” capable of 14 TFLOPS, which is a bit confusing given the description

The new supercomputer – with a total of 1,440 microprocessors – is based on three IBM iDataPlex systems, equal to the size of six refrigerators; an IBM Cluster 1350; and multiple IBM storage systems running General Parallel File System, supported by IBM Global Services.

AF building out its PS3 cluster with another 2200 consoles

John West — Tue, 24 Nov 2009 14:32:06 +0000

InformationWeek reported late last week that the Air Force is set to buy another 2,200 Sony PlayStation 3′s for its existing cluster

The PlayStation 3s will be used at the Air Force Research Laboratory’s information directorate in Rome, N.Y., where they will be added to an existing cluster of 336 PlayStation 3s being used to conduct supercomputing research.

The Air Force will use the system to “to determine the best fit for implementation of various applications,” including commercial and internally developed software specific to the PS3′s Cell Broadband Engine processor architecture. The research will help the Air Force decide where Cell Broadband Engine processor-derived hardware and software could be used in military systems.

The AF is using the cluster for a variety of things, including radar image processing. More in the article.

NSF funds Ga Tech to develop petascale tools for study of genomic evolution

John West — Tue, 17 Nov 2009 14:00:56 +0000

Georgia Tech announced today that the NSF, using money from the American Recovery and Reinvestment Act of 2009, has funded a four year, $1M effort to develop new petascale tools for genomic sequencing

Even on today’s fastest parallel computers, it could take centuries to analyze genome rearrangements for large, complex organisms. That is why the research team — which also includes Jijun Tang, an associate professor in the Department of Computer Science and Engineering at the University of South Carolina; and Stephen Schaeffer, an associate professor of biology at Penn State — is focusing on future generations of petascale machines, which will be able to process more than a thousand trillion, or 10^15, calculations per second. Today, most personal computers can only process a few hundred thousand calculations per second.

The researchers plan to develop new algorithms in an open-source software framework that will utilize the capabilities of parallel, petascale computing platforms to infer ancestral rearrangement events. The starting point for developing these new algorithms will be GRAPPA, an open-source code co-developed by Bader and initially released in 2000 that reconstructed the evolutionary relatedness among species.

SDSC gets $20M from NSF for new flash-based super

John West — Thu, 05 Nov 2009 15:40:40 +0000

SDSC is following up on its DASH experiment with a full-fledged system for data-intensive computing supported by flash drives

The San Diego Supercomputer Center (SDSC) at UC San Diego has been awarded a five-year, $20 million grant from the National Science Foundation (NSF) to build and operate a powerful supercomputer dedicated to solving critical science and societal problems now overwhelmed by the avalanche of data generated by the digital devices of our era.

The new system, named “Gordon” (get it?), will use flash memory for high data throughput and virtualization software from ScaleMP to create large shared-memory “supernodes.” The system is being built by Appro and will be installed in the mid-2011 timeframe. The completed system will have 245 peak TFLOPS, 64 TB of RAM, and 256 TB of flash memory.

“We are clearly excited about the potential for Gordon,” said SDSC Interim Director Michael Norman, who is also the project’s principal investigator. “This HPC system will allow researchers to tackle a growing list of critical ‘data-intensive’ problems. These include the analysis of individual genomes to tailor drugs to specific patients, the development of more accurate models to predict the impact of earthquakes on buildings and other structures, and simulations that offer greater insights into what’s happening to the planet’s climate.”

There is much more in the release, which is quite detailed.

China Daily reports 1 PF Intel/AMD hybrid super (but it’s not really)

John West — Mon, 02 Nov 2009 15:40:02 +0000

Late last week China Daily reported the launch of a new petaFLOPS super in China

The birth of the supercomputer, named “Tianhe”, which means “Milky Way”, makes China the second country, after the United States, to build a petaflop computer.

There is some confusion in the press about whether China is the second or third country to boast a PFLOPS. The China Top100 actually reveals that the machine is a peak PFLOPS machine, which makes it number 3, after Germany and JUGENE. The actual LINPACK score is 563,100 GFLOPS, which would have made it number 4 on the June list.

The new machine was announced concurrently with the annual update of the China Top100 (Google Translation to English of the Top100 site here). The machine is located at the National University of Defense Technology and is built out of Intel Xeon E5540s and AMD Radeon graphics cards for a GPU assist; the interconnect is IB. Tianhe is 5-ish more times more powerful that the previous champ, the Dawn 5000A, which was built out of AMD Barcelonas.

Supported by the national high-tech research and development program, also known as the 863 program, development of Tianhe has involved more than 200 computer experts who spent about two years in designing and producing the petaflop supercomputer, said Zhang Yulin, president of the defense university.

…Tianhe, made up of 103 refrigerator-sized silver gray cabinets, occupies an area of nearly 1,000 sq m and weighs 155 tons, containing 6,144 Intel CPUs and 5,120 AMD GPUs, with a storage capacity of 1 PB.

You can also find coverage at China’s official state-run news agency. A few weeks ago I pointed to the Asia Top500 (which actually has only 45 entries), as a regional entry alongside the country-specific Russia Top50. Someone less jaded than me would argue that the rise of the baby TopX lists is a reflection of the increasing recognition around the world of the importance of advanced computational capability in maintaining scientific and technical capability.