HPC Wire

Subscribe to HPC Wire feed
Since 1987 - Covering the Fastest Computers in the World and the People Who Run Them
Updated: 5 hours 3 min ago

Activist Investor Ratchets up Pressure on Mellanox to Boost Returns

Tue, 01/09/2018 - 10:29

Activist investor Starboard Value has sent a letter to Mellanox CEO Eyal Waldman demanding dramatic operational changes to boost returns to shareholders. This is the latest missive in an ongoing struggle between Starboard and Mellanox that began back in November when Starboard raised its stake in the interconnect specialist to 10.7 percent. Starboard argues Mellanox is significantly undervalued and that its costs, notably R&D, are unreasonably high.

The letter, dated January 8 and under the signature of Peter Feld, is pointed as shown in this excerpt:

“As detailed in the accompanying slides, over the last twelve months Mellanox’s R&D expenditures as a percentage of revenue were 42%, compared to the peer median of 22%. On SG&A, Mellanox spent 24% of revenue versus the peer median of 17%. It is critical to appreciate that Mellanox is not just slightly worse than peers on these key metrics, it is completely out of line with the peer group.”

Mellanox issued 2018 guidance for “low-to-mid-teens” (percent) revenue growth. Starboard cites a ‘consensus’ estimate of $816.5 million in revenue for 2017 and $986.4 million (14.5 percent). At 70.6 percent, Mellanox has one of the highest gross margins among comparable companies, and one of the lowest operating margins at 13.8 percent, according to Starboard.

“We believe there is a tremendous opportunity at Mellanox, but it will require substantial change, well beyond just the Company’s recently announced 2018 targets,” wrote Feld.

Link to Starboard letter: http://www.starboardvalue.com/wp-content/uploads/Starboard_Value_LP_Letter_to_MLNX_01.08.2018.pdf

The post Activist Investor Ratchets up Pressure on Mellanox to Boost Returns appeared first on HPCwire.

ACM Names New Director of Global Policy and Public Affairs

Tue, 01/09/2018 - 10:24

NEW YORK, Jan. 9, 2018 — ACM, the Association for Computing Machinery, has named Adam Eisgrau as its new Director of Global Policy and Public Affairs, effective January 3, 2018. Eisgrau will coordinate and support ACM’s engagement with public technology policy issues involving information technology, globally and particularly in the US and Europe. ACM aims to educate and inform computing professionals, policymakers, and the public about information technology policy and its consequences, and to shape public technology policy through a deeper understanding of the information technology issues involved.

“ACM has long been committed to providing policy makers in the US and abroad with the most current, accurate, objective and non-partisan information about all things digital as they wrestle with issues that profoundly affect billions of people,” said ACM President Vicki L. Hanson. “We’re thrilled to add a communicator of Adam’s caliber to our team as the computing technologies pioneered, popularized and promulgated by ACM members become ever more integrated to the fabric of daily life.”

“Speaking tech to power clearly, apolitically and effectively has never been more important,” said Eisgrau. “The chance to do so for ACM in Washington, Brussels and beyond is a dream opportunity.”

A former communications attorney, Eisgrau began his policy career as Judiciary Committee Counsel to then-freshman US Senator Dianne Feinstein (D-CA). Since leaving Senator Feinstein’s office in 1995, he has represented both public- and private-sector interests in international forums and to Congress, federal agencies and the media on a host of technology-driven policy matters. These include: digital copyright, e-commerce competition, peer-to-peer software, cybersecurity, encryption, online financial services, warrantless surveillance and digital privacy.

Prior to joining ACM, Eisgrau directed the government relations office of the American Library Association. He is a graduate of Dartmouth College and Harvard Law School.

About ACM

ACM, the Association for Computing Machinery www.acm.org, is the world’s largest educational and scientific computing society, uniting computing educators, researchers and professionals to inspire dialogue, share resources and address the field’s challenges. ACM strengthens the computing profession’s collective voice through strong leadership, promotion of the highest standards, and recognition of technical excellence. ACM supports the professional growth of its members by providing opportunities for life-long learning, career development, and professional networking.

Source: ACM

The post ACM Names New Director of Global Policy and Public Affairs appeared first on HPCwire.

NOAA to Expand Compute Capacity by 50 Percent with Two New Dells

Tue, 01/09/2018 - 10:10

January 9, 2018 — NOAA’s combined weather and climate supercomputing system will be among the 30 fastest in the world, with the ability to process 8 quadrillion calculations per second, when two Dell systems are added to the IBMs and Crays at data centers in Reston, Virginia, and Orlando, Florida, later this month.

“NOAA’s supercomputers play a vital role in monitoring numerous weather events from blizzards to hurricanes,” said Secretary of Commerce Wilbur Ross. “These latest updates will further enhance NOAA’s abilities to predict and warn American communities of destructive weather.”

This upgrade completes phase three of a multi-year effort to build more powerful supercomputers that make complex calculations faster to improve weather, water and climate forecast models. It adds 2.8 petaflops of speed at both data centers combined, increasing NOAA’s total operational computing speed to 8.4 petaflops — or 4.2 petaflops per site.

Sixty percent more storage

The upgrade also adds 60 percent more storage capacity, allowing NOAA to collect and process more weather, water and climate observations used by all the models than ever before.

“NOAA’s supercomputers ingest and analyze billions of data points taken from satellites, weather balloons, airplanes, buoys and ground observing stations around the world each day,” said retired Navy Rear Adm. Timothy Gallaudet, Ph.D., acting NOAA administrator. “Having more computing speed and capacity positions us to collect and process even more data from our newest satellites — GOES-East, NOAA-20 and GOES-S — to meet the growing information and decision-support needs of our emergency management partners, the weather industry and the public.”

With this upgrade, U.S. weather supercomputing paves the way for NOAA’s National Weather Service to implement the next generation Global Forecast System, known as the “American Model,” next year. Already one of the leading global weather prediction models, the GFS delivers hourly forecasts every six hours. The new GFS will have significant upgrades in 2019, including increased resolution to allow NOAA to run the model at 9 kilometers and 128 levels out to 16 days, compared to the current run of 13 kilometers and 64 levels out to 10 days. The revamped GFS will run in research mode on the new supercomputers during this year’s hurricane season.

“As we look toward launching the next generation GFS in 2019, we’re taking a ‘community modeling approach’ and working with the best and brightest model developers in this country and abroad to ensure the new U.S. model is the most accurate and reliable in the world,” said National Weather Service Director Louis W. Uccellini, Ph.D.

Supporting a Weather-Ready Nation

The upgrade announced today – part of the agency’s commitment to support the Weather-Ready Nation initiative – will lead to more innovation, efficiency and accuracy across the entire weather enterprise. It opens the door for the National Weather Service to advance its seamless suite of weather, water and climate models over the next few years, allowing for more precise forecasts of extreme events a week in advance and beyond.

Improved hurricane forecasts and expanded flood information will enhance the agency’s ability to deliver critical support services to local communities. In addition, the new supercomputers will allow NOAA’s atmosphere and ocean models to run as one system, helping forecasters to more readily identify interaction between the two and reducing the number of operational models; as well as allow for development of a new seasonal forecast system to replace the Climate Forecast System in 2022, paving the way for improved seasonal forecasts as part of the Weather Research and Forecasting Innovation Act.

The added computing power will support upgrades to the National Blend of Models, which is being developed to provide a common starting point for all local forecasts; allow for more sophisticated ensemble forecasting, which is a method of improving the accuracy of forecasts by averaging results of various models; and provide quicker turnaround for atmosphere and ocean simulations, leading to earlier predictions of severe weather.

NOAA’s mission is to understand and predict changes in the Earth’s environment, from the depths of the ocean to the surface of the sun, and to conserve and manage our coastal and marine resources.

Source: NOAA

The post NOAA to Expand Compute Capacity by 50 Percent with Two New Dells appeared first on HPCwire.

Momentum Builds for US Exascale

Tue, 01/09/2018 - 10:08

2018 looks to be a great year for the U.S. exascale program. The last several months of 2017 revealed a number of important developments that help put the U.S. quest for exascale on a solid foundation. In my last article, I provided a description of the elements of the High Performance Computing (HPC) ecosystem and its importance for advancing and sustaining this strategically important technology. It is good to report that the U.S. exascale program seems to be hitting the full range of ecosystem elements.

As a reminder, the National Strategic Computing Initiative (NSCI) assigned the U.S. Department of Energy (DOE) Office of Science (SC) and the National Nuclear Security Administration (NNSA) to execute a joint program to deliver capable exascale computing that emphasizes sustained performance on relevant applications and analytic computing to support their missions. The overall DOE program is known as the Exascale Computing Initiative (ECI) and is funded by the SC Advanced Scientific Computing Research (ASCR) program and the NNSA Advanced Simulation and Computing (ASC) program. Elements of the ECI include the procurement of exascale class systems and the facility investments in site preparations and non-recurring engineering. Also, ECI includes the Exascale Computing Project (ECP) that will conduct the Research and Development (R&D) in the areas of middleware (software stack), applications, and hardware to ensure that exascale systems will be productively usable to address Office of Science and NNSA missions.

In the area of hardware – the last part of 2017 revealed a number of important developments. First and most visible, is the initial installation of the SC Summit system at Oak Ridge National Laboratory (ORNL) and the NNSA Sierra system at Lawrence Livermore National Laboratory (LLNL). Both systems are being built by IBM using Power9 processors with Nvidia GPU co-processors. The machines will have two Power9 CPUs per system board and will use a Mellenox InfinBand interconnection network.

Beyond that, the architecture of each machine is slightly different. The ORNL Summit machine will use six Nvidia Volta GPUs per two Power9 CPUs on a system board and will use NVLink to connect to 512 GB of memory. The Summit machine will use a combination of air and water cooling. The LLNL Sierra machine will use four Nvidia Voltas and 256 GB of memory connected with the two Power9 CPUs per board. The Sierra machine will use only air cooling. As was reported by HPCwire in November 2017, the peak performance of the Summit machine will be about 200 petaflops and the Sierra machine is expected to be about 125 petaflops.

Installation of both the Summit and Sierra systems is currently underway with about 279 racks (without system boards) and the interconnection network already installed at each lab. Now that IBM has formally released the Power9 processors, the racks will soon start being populated with the boards that contain the CPUs, GPUs and memory. Once that is completed, the labs will start their acceptance testing, which is expected to be finished later in 2018.

Another important piece of news about the DOE exascale program is the clarification of the status of the Argonne National Laboratory (ANL) Aurora machine. This system was part of the collaborative CORAL procurement that also selected the Sierra and Summit machines. The Aurora system is being manufactured by Intel with Cray Inc. acting as the system integrator. The machine was originally scheduled to be an approximately 180 peak petaflops system using the Knights Hill third generation Phi processors. However, during SC17, we learned that Intel is removing the Knights Hill chip from its roadmap. This explains the reason why during the September ASCR Advisory Committee (ASCAC) meeting, Barb Helland, the Associate Director of the ASCR office, announced that the Aurora system would be delayed to 2021 and upgraded to 1,000 petaflops (aka 1 exaflops).

The full details of the revised Aurora system are still under wraps. We have learned that it is going to use “novel” processor technologies, but exactly what that means is unclear. The ASCR program subjected the new Aurora design to an independent outside review. It found, “The hardware choices/design within the node is extremely well thought through. Early projections suggest that the system will support a broad workload.” The review committee even suggested that, “The system as presented is exciting with many novel technology choices that can change the way computing is done.” The Aurora system is in the process of being “re-baselined” by the DOE. Hopefully, once that is complete, we will get a better understanding of the meaning of “novel” technologies. If things go as expected, the changes to Aurora will allow the U.S. to achieve exascale by 2021.

An important, but sometimes overlooked, aspect of the U.S. exascale program is the number of computing systems that are being procured, tested and optimized by the ASCR and ASC programs as part of the buildup to exascale. Other computing systems involved with “pre-exascale” systems include the 8.6 petaflops Mira computer at ANL and the 14 petaflops Cori system at Lawrence Berkeley National Lab (LBNL). The NNSA also has the 14.1 petaflops Trinity system at Los Alamos National Lab (LANL). Up to 20 percent of these precursor machines will serve as testbeds to enable computing science R&D needed to ensure that the U.S. exascale systems will be able to productively address important national security and discovery science objectives.

The last, but certainly not least, bit of hardware news is that the ASCR and ASC programs are expected to start their next computer system procurement processes in early 2018. During her presentation to the U.S. Consortium for the Advancement of Supercomputing (USCAS), Barb Helland told the group that she expects that the Request for Proposals (RFP) will soon be released for the follow-ons to the Summit and Sierra systems. These systems, to be delivered in the 2021-2023 timeframe, are expected to be provide in excess of exaFLOP/s performance. The procurement process to be used will be similar to the CORAL procurement and will be a collaboration between the DOE-SC ASCR and NNSA ASC programs. The ORNL exascale system will be called Frontier and the LLNL system will be known as El Capitan.

2017 also saw significant developments for the people element of the U.S HPC ecosystem. As was previously reported, at last September’s ASCAC meeting, Paul Messina announced that he would be stepping down as the ECP Director on October 1st. Doug Kothe, who was previously the applications development lead, was announced as the new ECP Director. Upon taking the Director job, Kothe with his deputy, Stephen Lee of LANL, instituted a process to review the organization and management of the ECP. At the December ASCAC conference call, Doug reported that the review had been completed and resulted in a number of changes. This included paring down ECP from five to four components (applications development, software technology, hardware and integration, and project management). He also reported that ECP has implemented a more structured management approach that includes a revised work breakdown structure (WBS) and additional milestones, new key performance parameters and risk management approaches. Finally, the new ECP Director reported that they had established an Extended Leadership Team with a number of new faces.

Another important, element of the HPC ecosystem are the people doing the R&D and other work need to keep the ecosystem going. The DOE ECI involves a huge number of people. Last year, there were about 500 researchers who attended the ECP Principle Investigator meeting and there are many more involved in other DOE/NNSA programs and from industry. The ASCR and ASC programs are involved with a number of programs to educate and train future members of the HPC ecosystem. Such programs are the ASCR and ASC co-funded Computational Science Graduate Fellowship (CSGF) and the Early Career Research Program. The NNSA offers similar opportunities. Both the ASCR and ASC programs continue to coordinate with National Science Foundation educational programs to ensure that America’s top computational science talent continues to flow into the ecosystem.

Finally, in addition to people and hardware, the U.S. program continues to develop the software stack (aka middleware) to develop end users’ applications to ensure that exascale will be used productively. Doug Kothe reported that ECP has adopted standard Software Development Kits. These SDKs are designed to support the goal of building a comprehensive, coherent software stack that enables application developers to productively write highly parallel applications that effectively target diverse exascale architectures. Kothe also reported that ECP is making good progress in developing applications software. This includes the implementation of innovative approaches that include Machine Learning to utilize the GPUs that are part of the future exascale computers.

All in all – the last several months of 2017 have set the stage for a very exciting 2018 for the U.S. exascale program. It has been about 5 years since the ORNL Titan supercomputer came onto the stage at #1 on the TOP500 list. Over that time, other more powerful DOE computers have come online (Trinity, Cori, etc.) but they were overshadowed by Chinese and European systems. It remains unclear whether or not the upcoming exascale systems will put the U.S. back on the top of the supercomputing world. However, the recent developments help to reassure the country is not going to give up its computing leadership position without a fight. That is great news because for more than 60 years, the U.S. has sought leadership in high performance computing for the strategic value it provides in the areas of national security, discovery science, energy security, and economic competitiveness.

About the Author

Alex Larzelere is a senior fellow at the U.S. Council on Competitiveness, the president of Larzelere & Associates Consulting and HPCwire’s policy editor. He is currently a technologist, speaker and author on a number of disruptive technologies that include: advanced modeling and simulation; high performance computing; artificial intelligence; the Internet of Things; and additive manufacturing. Alex’s career has included time in federal service (working closely with DOE national labs), private industry, and as founder of a small business. Throughout that time, he led programs that implemented the use of cutting edge advanced computing technologies to enable high resolution, multi-physics simulations of complex physical systems. Alex is the author of “Delivering Insight: The History of the Accelerated Strategic Computing Initiative (ASCI).”

The post Momentum Builds for US Exascale appeared first on HPCwire.

Stampede1 Helps Researchers Examine a Greener Carbon Fiber Alternative

Tue, 01/09/2018 - 07:48

Jan. 9, 2018 — From cars and bicycles to airplanes and space shuttles, manufacturers around the world are trying to make these vehicles lighter, which helps lower fuel use and lessen the environmental footprint.

One way that cars, bicycles, airplanes and other modes of transportation have become lighter over the last several decades is by using carbon fiber composites. Carbon fiber is five-times stronger than steel, twice as stiff, and substantially lighter, making it the ideal manufacturing material for many parts. But with the industry relying on petroleum products to make carbon fiber today, could we instead use renewable sources?

In the December 2017 issue of Science, Gregg Beckham, a group leader at the National Renewable Energy Laboratory (NREL), and an interdisciplinary team reported the results of experimental and computational investigations on the conversion of lignocellulosic biomass into a bio-based chemical called acrylonitrile, the key precursor to manufacturing carbon fiber.

The catalytic reactor shown here is for converting chemical intermediates into acrylonitrile. The work is part of the Renewable Carbon fiber Consortium. Photo by Dennis Schroeder/NREL

Acrlyonitrile is a large commodity chemical, and it’s made today through a complex petroleum-based process at the industrial scale. Propylene, which is derived from oil or natural gas, is mixed with ammonia, oxygen, and a complex catalyst. The reaction generates high amounts of heat and hydrogen cyanide, a toxic by-product. The catalyst used to make acrylonitrile today is also quite complex and expensive, and researchers still do not fully understand its mechanism.

“That’s where our study comes in,” Beckham said. “Acrylonitrile prices have witnessed large fluctuations in the past, which has in turn led to lower adoption rates for carbon fibers for making cars and planes lighter weight. If you can stabilize the acrylonitrile price by providing a new feedstock from which to make acrylonitrile, in this case renewably-sourced sugars from lignocellulosic biomass, we might be able to make carbon fiber cheaper and more widely adopted for everyday transportation applications.”

To develop new ideas to make acrylonitrile manufacturing from renewable feedstocks, the Department of Energy (DOE) solicited a proposal several years ago that asked: Is it possible to make acrylonitrile from plant waste material? These materials include corn stover, wheat straw, rice straw, wood chips, etc. They’re basically the inedible part of the plant that can be broken down into sugars, which can then be converted to a large array of bio-based products for everyday use, such as fuels like ethanol or other chemicals.

“If we could do this in an economically viable way, it could potentially decouple the acrylonitrile price from petroleum and offer a green carbon fiber alternative to using fossil fuels,” Beckham said.

Beckham and the team moved forward to develop a different process. The NREL process takes sugars derived from waste plant materials and converts those to an intermediate called 3-hydroxypropionic acid (3-HP). The team then used a simple catalyst and new chemistry, dubbed nitrilation, to convert 3-HP to acrylonitrile at high yields. The catalyst used for the nitrilation chemistry is about three times less expensive than the catalyst used in the petroleum-based process and it’s a simpler process. The chemistry is endothermic so it doesn’t produce excess heat, and unlike the petroleum-based process, it doesn’t produce the toxic byproduct hydrogen cyanide. Rather, the bio-based process only produces water and alcohol as its byproducts.

From a green chemistry perspective, the bio-based acrylonitrile production process has multiple advantages over the petroleum-based process that is being used today. “That’s the crux of the study,” Beckham said.

XSEDE’s Role in the Chemistry

Beckham is no stranger to XSEDE, the eXtreme Science and Engineering Discovery Environment that’s funded by the National Science Foundation. He’s been using XSEDE resources, including Stampede1, Bridges, Comet and now Stampede2, for about nine years as a principal investigator. Stampede1 and Stampede2 (currently #12 on the Top500 list list) are deployed and maintained by the Texas Advanced Computing Center.

Most of the biological and chemistry research conducted for this project was experimental, but the mechanism of the nitrilation chemistry was only at first hypothesized by the team. A postdoctoral researcher in the team, Vassili Vorotnikov of NREL, was recruited to run periodic density functional theory calculations on Stampede1 as well as the machines at NREL to elucidate the mechanism of this new chemistry.

Over about two months and several millions of CPU-hours used on Stampede1, the researchers were able to shed light on the chemistry of this new catalytic process. “The experiments and computations lined up nicely,” Vorotnikov said.

Because they had an allocation on Stampede1, they were able to rapidly turn around a complete mechanistic picture of how this chemistry works. “This will help us and other Top500 institutions to develop this chemistry further and design catalysts and processes more rationally,” Vorotnikov said. “XSEDE and the predictions of Stampede1 are pointing the way forward on how to improve nitrilation chemistry, how we can apply it to other molecules, and how we can make other renewable products for industry.”

“After the initial experimental discovery, we wanted to get this work out quickly,” Beckham continued. “Stampede1 afforded a great deal of bandwidth for doing these expensive, computationally intensive density functional theory calculations. It was fast and readily available and just a great machine to do these kind of calculations on, allowing us to turn around the mechanistic work in only a matter of months.”

Next Steps

There’s a large community of chemists, biologists and chemical engineers who are developing ways to make everyday chemicals and materials from plant waste materials instead of petroleum. Researchers have tried to do this before with acrylonitrile. But no one has been as successful in the context of developing high yielding processes with possible commercial potential for this particular product. With their new discovery, the team hopes this work makes the transition into industry sooner rather than later.

The immediate next step is scaling the process up to produce 50 kilograms of acrylonitrile. The researchers are working with several companies including a catalyst company to produce the necessary catalyst for pilot-scale operation; an agriculture company to help scale up the biology to produce 3-HP from sugars; a research institute to scale the separations and catalytic process; a carbon fiber company to produce carbon fibers from the bio-based acrylonitrile; and a car manufacturer to test the mechanical properties of the resulting composites.

“We’ll be doing more fundamental research as well,” Beckham said. “Beyond scaling acrylonitrile production, we are also excited about is using this powerful, robust chemistry to make other everyday materials that people can use from bio-based resources. There are lots of applications for nitriles out there — applications we’ve not yet discovered.”

Source: Faith Singer-Villalobos, TACC

The post Stampede1 Helps Researchers Examine a Greener Carbon Fiber Alternative appeared first on HPCwire.

Mixed-Signal Neural Net Leverages Memristive Technology

Mon, 01/08/2018 - 13:11

Memristive technology has long been attractive for potential use in neuromorphic computing. Among other things it would permit building artificial neural network (ANN) circuits that are processed in parallel and more directly emulate how neuronal circuits in the brain work. Recent work led by researchers at Oak Ridge National Laboratory and the University of Tennessee proposes a mixed signal approach that leverages memristive technology to build better ANNs.

“[Our] mixed-signal approach implements neural networks with spiking events in a synchronous way. Moreover, the use of nano-scale memristive devices saves both area and power in the system… The proposed [system] includes synchronous digital long term plasticity (DLTP), an online learning methodology that helps the system train the neural networks during the operation phase and improves the efficiency in learning considering the power consumption and area overhead,” writes Catherine Schuman, a Liane Russell Early Career Fellow in Computational Data Analytics at Oak Ridge National Laboratory, and colleagues[i].

Their paper, Memristive Mixed-Signal Neuromorphic Systems: Energy-Efficient Learning at the Circuit-Level, was published in the IEEE Journal on Emerging and Selected Topics in Circuits and Systems.

The researchers point out that digital and analog approaches to building ANNs each have drawbacks. While digital implementations have precision, robustness, noise resilience and scalability, they are area intensive. Conversely, analog counterparts are efficient in terms of silicon area and processing speed, but “rely on representing synaptic weights as volatile voltages on capacitors or in resistors, which do not lend themselves to energy and area efficient learning.”

Instead, they propose a mixed-signal system where communication and control is digital while the core multiply-and-accumulate functionality is analog. Researchers used a hafnium-oxide memristor design based on earlier work (“A practical hafnium-oxide memristor model suitable for circuit design and simulation,” in Proceedings of IEEE International Symposium on Circuits and Systems).

Their design (figure two, shown below) consists of m x n memristive neuromorphic cores. “Each core has several memristive synapses and one mixed-signal neuron (analog in, digital out) to implement a spiking neural network. This arrangement helps maintain similar capacitance at the synaptic outputs and corresponding neurons. The similar distance between synapse and inputs also results in negligible difference in charge accumulation,” write the authors.

Also exciting is the researchers’ approach to implementing learning. Most ANNs require offline learning. For a network to learn online, Long Term Plasticity plays an important role in training the circuit with continuous updates of synaptic weights based on the timing of pre- and post-neuron fires.

“Instead of carefully crafting analog tails to provide variation in the voltage across the synapses, we utilize digital pre- and post-neuron firing signals and apply pulse modulation to implement a digital LTP (DLTP) technique…Basically the online learning process implemented here is one clock cycle tracking version of Spike time Dependent Plasticity… A more thorough STDP learning implementation would need to track several clock cycles before and after the post-neuron fire leading to more circuitry and hence increased power and area. Our DLTP approach acts similarly but ensures lower area and power,” write the authors.

Link to paper: http://ieeexplore.ieee.org/document/8119503/

Feature image source: ORNL

[i] Gangotree Chakma, Student Member, IEEE, Md Musabbir Adnan, Student Member, IEEE, Austin R. Wyer, Student Member, IEEE, Ryan Weiss, Student Member, IEEE, Catherine D. Schuman, Member, IEEE, and Garrett S. Rose, Member, IEEEAustin R. Wyer, Student Member, IEEE, Ryan Weiss, Student Member, IEEE, Catherine D. Schuman, Member, IEEE, and Garrett S. Rose, Member, IEEE

The post Mixed-Signal Neural Net Leverages Memristive Technology appeared first on HPCwire.

Curie Supercomputer Uses HPC to Help Improve Agricultural Production

Mon, 01/08/2018 - 11:47

Jan. 8, 2018 — Agriculture is the principal means of livelihood in many regions of the developing world, and the future of our world depends on a sustainable agriculture at planetary level. High Performance Computing is becoming critical in agricultural activity, plague control, pesticides design and pesticides effects. Climate data are used to understand the impacts on water and agriculture in many regions of the world, help local authorities in the management of water and agricultural resources, and assist vulnerable communities in the region through improved drought management and response.

Image courtesy of the European Commission.

The demand for agricultural products has increased globally and meeting this growing demand would have a negative effect on the environment.  Increased agricultural production needs the use of 70% of the world’s water resources and a rise in greenhouse gas emissions.

To be able to reduce the negative impact to the ecosystem, seed companies are on the lookout for new plant varieties that yield more produce. Companies normally find such new varieties through field trials.  These field trials are a simple observation method but they cost a lot of money and are time consuming taking years to find the best ones.

Using High Performance Computing (HPC), the Curie supercomputer is able to provide the most efficient solution to this problem.  HPC enables numerical simulations of plant growth that help seed companies to achieve superior varieties instead of doing field trials which are more expensive and harmful for the environment.

For example, if a farmer wants to know what the conditions are for a plant to grow best in ( its genetic parameter), they would have to test its growth rate under various conditions to select the best parameter corresponding to the specific environment of the region. With the help of HPC, the estimation of these parameters is made more accurate and simpler by simulating plant growth. The simulation models take into account, the plant’s interaction with the environment.  It reduces the number of field trials by a large percent, for example, instead of 100, 10 field trials would be enough to  estimate the best genetic parameter.

Cybele Tech, the French company has used High Performance Computing to enable farmers to produce more with less and know what exactly their plants need to get a better yield.

They’ve been awarded with 4 million core hours on Curie hosted by GENCI at CEA, France.

Source: European Commission

The post Curie Supercomputer Uses HPC to Help Improve Agricultural Production appeared first on HPCwire.

Ellexus Publishes White Paper Advising HPCers on Meltdown, Spectre

Mon, 01/08/2018 - 10:10

Jan 8 — Can you afford to lose a third of your compute real estate? If not, you need to pre-empt the impact of Meltdown and Spectre.

Meltdown and Spectre are quickly becoming household names and not just in the HPC space. The severe design flaws in Intel microprocessors that could allow sensitive data to be stolen and the fixes are likely to be bad news for any I/O intensive applications such as those often used in HPC.

Ellexus Ltd, the I/O profiling company, has released a white paper: How the Meltdown and Spectre bugs work and what you can do to prevent a performance plummet.

Why is the Meltdown fix worse for HPC applications?

The changes that are being imposed on the Linux kernel (called the KAISER patch) to more securely separate user and kernel space are causing additional overhead to context switches. This is having a measurable impact on the performance of shared file systems and I/O intensive applications, which is particularly noticeable in I/O heavy workloads. A performance penalty could reach 10-30%.

Systems that were previously just about coping with I/O heavy workloads could now be in real trouble. It’s very easy for applications sharing datasets to overload the file system and prevent other applications from working, but bad I/O can also affect each program in isolation, even before the patches for the attacks make that worse.

Profile application I/O to rescue lost performance

You don’t have to put up with poor performance in order to improve security, however. The most obvious way to mitigate performance losses is to profile I/O and identify ways to optimise applications’ I/O performance.

By using the tool suites from Ellexus, Breeze and Mistral, to analyse workflows it is possible to identify changes that will help to eliminate bad I/O and regain the performance lost to these security patches.

Ellexus’ tools locate bottlenecks and applications with bad I/O on large distributed systems, cloud infrastructure and super computer clusters. Once applications with bad I/O patterns have been located, our tools will indicate the potential performance increases as well as pointers on how to achieve them. Often the optimisation is as simple as changing an environment variable, changing a single line in a script or changing a simple I/O call to read more than one byte at a time.

In some cases, the candidates for optimisation will be obvious – a workflow that clearly stresses the file system every time it is run, for example, or one that runs for significantly longer than a typical task.

In others it may be necessary to perform an initial high-level analysis of each job. Follow three steps to optimise application I/O and mitigate the impact of the KAISER patch:

1.       Profile all your applications with Mistral to look for the worst I/O patterns

Mistral, our I/O profiling tool, is lightweight enough to run at scale. In this case Mistral would be set up to record relatively detailed information on the type of I/O that workflows are performing over time. It would look for factors such as how many meta data operations are being performed, the number of small I/O and so on.

2.       Deal with the worst applications, delving into detail with Breeze

Once the candidate workflows have been identified they can be analysed in detail with Breeze. As a first step, the Breeze trace can be run through our Healthcheck tool that identifies common issues such as an application that has a high ratio of file opens to writes or a badly configured $PATH causing the file system to be trawled every time a workflow uses “grep”.

3.       Put in place longer-term I/O quality assurance

Implement the Ellexus tools across your systems to get the most from the compute and storage and to prevent problems reoccurring.

By following these simple steps and our best practices guidance it is easy to find and fix the biggest issues quickly and give you more time to optimise for the best performance possible.

Source: Ellexus Ltd

The post Ellexus Publishes White Paper Advising HPCers on Meltdown, Spectre appeared first on HPCwire.

AMD Previews Processor and Graphics Products for HPC at CES 2018

Mon, 01/08/2018 - 07:33

LAS VEGAS, Jan. 8, 2018 — AMD has detailed its forthcoming roll-out plan for its new and next generation of high-performance computing and graphics products during an event in Las Vegas just prior to the opening of CES 2018. Alongside announcing the first desktop Ryzen processors with built-in Radeon Vega Graphics, AMD also detailed the full line up of Ryzen mobile APUs including the new Ryzen PRO and Ryzen 3 models, and provided a first look at the performance of its upcoming 12nm 2nd generation Ryzen desktop CPU expected to launch in April. In graphics, AMD announced the expansion of the “Vega” family with Radeon Vega Mobile and that its first 7nm product is planned to be a Radeon “Vega” GPU specifically built for machine learning applications.

“We successfully accomplished the ambitious goals we set for ourselves in 2017, reestablishing AMD as a high-performance computing leader with the introduction and ramp of 10 different product families,” said AMD President and CEO Dr. Lisa Su. “We are building on this momentum in 2018 as we make our strongest product portfolio of the last decade even stronger with new CPUs and GPUs that bring more features and more performance to a broad set of markets.”

Technology Updates

AMD CTO and SVP Mark Papermaster shared updates on AMD’s process technology roadmaps for both x86 processors and graphics architectures.

  • x86 Processors
    • The “Zen” core, currently shipping in Ryzen desktop and mobile processors, is in production at both 14nm and 12nm, with 12nm samples now shipping.
    • The “Zen 2” design is complete and will improve on the award-winning “Zen” design in multiple dimensions.
    • AMD is on track to deliver performance and performance-per-watt improvements through 2020.
  • Graphics Processors
    • Expanding  the “Vega” product family in 2018 with the Radeon Vega Mobile GPU for ultrathin notebooks.
    • The first 7nm AMD product,  a Radeon “Vega” based GPU built specifically for machine learning applications.
    • A production-level machine learning software environment with AMD’s MIOpen libraries supporting common machine learning frameworks like TensorFlow and Caffe on the ROCm Open eCosystem platform. The industry’s first fully open heterogeneous software environment, which is making it easier to program using AMD GPUs for high performance compute and deep learning environments.

Client Compute Updates

AMD SVP and General Manager, Computing and Graphics Business Group Jim Anderson detailed upcoming AMD client compute processors including:

  • The Ryzen desktop processor with Radeon graphics
    • Desktop Ryzen APUs combine the latest “Zen” core and AMD Radeon graphics engine based on the advanced “Vega” architecture, bringing:
      • The highest performance graphics engine in a desktop processor
      • Advanced quad core performance with up to 8 processing threads
      • 1080p HD+ gaming performance without a discrete graphics card
      • Beautiful display features with Radeon FreeSync technology
      • Full benefit of Radeon software driver features including Radeon Chill, Enhanced Sync and Radeon ReLive
    • Planned to be available starting February 12, 2018.

 

About AMD

For more than 45 years, AMD has driven innovation in high-performance computing, graphics, and visualization technologies ― the building blocks for gaming, immersive platforms, and the datacenter. Hundreds of millions of consumers, leading Fortune 500 businesses, and cutting-edge scientific research facilities around the world rely on AMD technology daily to improve how they live, work, and play. AMD employees around the world are focused on building great products that push the boundaries of what is possible.

Source: AMD

The post AMD Previews Processor and Graphics Products for HPC at CES 2018 appeared first on HPCwire.

ANL’s Rick Stevens on CANDLE, ARM, Quantum, and More

Mon, 01/08/2018 - 07:30

Late last year HPCwire caught up with Rick Stevens, associate laboratory director for computing, environment and life Sciences at Argonne National Laboratory, for an update on the CANDLE (CANcer Distributed Learning Environment) project on which he is a PI. CANDLE is an effort to develop a “a broad deep learning infrastructure able to run on leadership class computers and other substantial machines” for use in cancer research. While most of the conversation covered CANDLE’s deep learning efforts, Stevens also offered thoughts on ARM technology’s prospects in HPC and challenges facing quantum and neuromorphic computing.

For background on CANDLE see HPCwire article, Deep Learning Thrives in Cancer Moonshot. There are many elements to the CANDLE program; Stevens is the PI for a pilot program that is screening drugs against cancer cell lines and xenograft tumor tissue and using the data to build models able to predict how effective the drugs will be against various cancers. Progress has been remarkably quick and Stevens presented a paper at SC17 – Predicting Tumor Cell Line Response to Drug Pairs with Deep Learning – showcasing his group’s efforts so far.

“There will be drugs, I predict, in clinical trials based on the results that we achieve this year,” Stevens told HPCwire.

Presented here are portions of the interview with Stevens.

HPCwire: Let’s start with CANDLE. Can you give us an update? As I understand it there’s a release on GitHub.

Rick Stevens, Argonne National Laboratory

Rick Stevens: We did the first release of the CANDLE environment this summer. It’s running on the Theta (Cray) machine at Argonne, on Cori (Cray) at NERSC, on SummitDev (IBM) at Oak Ridge, and will soon be running on Summit (IBM). It’s also running on an Nvidia DGX-1 and at the NIH campus on Biowulf. Those are the main platforms. We’ve been using CANDLE both as the production engine but also using it to search for better model parameters and better hyper-parameters for the cancer models in the drug responder problem.

HPCwire: This is the pilot to screen drugs against cancer cell lines and xenografts and to use results to develop predictive models of how effective the drugs are?

Rick Stevens: Yes. We actually have a new model and it’s achieving much better performance than anything anybody else has in terms of predictive drug pair response. Now this is using experimental data from cell lines, it’s not yet using clinical data. One of the key problems is that although it seems like we have a lot of data, we actually have very little data of the type we need, which is high-quality labeled clinical data that we can link back to high-quality molecular data.

HPCwire: What makes the new model so much better?

Rick Stevens: We’ve been experimenting with convolutional networks, which are widely used in computer vision, and we thought that was giving us better performance than networks we’d been trying before which were simpler. We did a bunch of experiments which showed that in fact they were training faster, but the accuracy we achieved with convolutions wasn’t better than the accuracy we achieved without convolutions – it just trained about ten times faster.

So we went back and started trying different network types. The one that we are currently using is based on residual networks. Basically it uses what is called a tower architecture[i] and it essentially is borrowing a different kind of idea developed for computer vision. Residual networks are where each layer of the network is both computing [i.e., learning] a new function, but it is also taking input from the previous layer. In other words, it allows the network to decide as it’s learning whether to use a transform feature it computes or whether to use the residual of the difference between that transform feature and the original version.

It comes up with its own weights during training and is doing that across thousands of connections, literally tens of thousands of connections. That architecture just works better. We have some theoretical understanding as to why it works better, but one notion of why it works better is that it gives the network a slightly simpler thing to learn each time.

That’s currently our best performing model across cell lines and it is being used in both single drugs and drug pairs. The drug pairs problem is the really hard one and we can [already] predict with about 93 percent accuracy the growth inhibition [or not] of the tumor when given these two drugs. That’s used to prioritize drugs for further testing. We’re using it right now to [design] follow-on experiments (network diagram below).

Neural network architecture. The orange square boxes, from bottom to top, represent input features, encoded features, and output growth values. Feature models are denoted by round shaded boxes: green for molecular features and blue for drug features. There are multiple types of molecular features that are fed into submodels for gene expression, proteome, and microRNA. The descriptors for the two drugs share the same descriptor model. All encoded features are then concatenated to form input for the top fully connected layers. Most connecting layers are linked by optional residual skip connections if their dimensions match. Source: Fig. 2 from the cited paper, Predicting Tumor Cell Line Response to Drug Pairs with Deep Learning.


HPCwire:
These models are correlation models, built on the results you see. Are you also working with mechanistic models?

Stevens: Although it is not a done deal yet, we are talking to a company that has built a mechanistic model for cancer drug response prediction that couples the machine learning models with the mechanistic models. The mechanistic models use mutational data and signaling pathways. This [collaboration] will help us fill in the holes where those [mechanistic] models fall down. We have a collaboration that is spinning up in a few months and maybe we’ll have some progress to show with this hybridization [approach].

HPCwire: Will the hybrid model outperform either of the models individually?

Rick Stevens: That’s what we are shooting for. These mechanistic models in very narrow cancer types are about 80 percent predictive. What we are hoping is that by combining these things we can push the combined engine up 96-97 percent. At that point you are probably in the noise in data at which things have been misclassified. So we have also been testing a lot of classification data. This is all tumor data from the large archives, NCI Genomic Data Commons, and we are building classifiers that can recognize between normal and tumor data and can also identify the cancer type and the site of origin based on just expression data. We can get these predictions to be about 98 percent accurate.

HPCwire: Poor data quality seems to be a constant problem in both cancer research and deep learning. Until recently much of the descriptive data in cancer came from pathologists looking at tissue under a microscope. Interpretations varied.

Rick Stevens: You know the worst thing for training is to have bad data so you want to clean the data throughout the outliers and have the best possible representation of the distribution you are trying to learn. The idea is building these kinds of quality control front ends. We’re doing it for cancer but it turns out that the autonomous vehicle people are doing exactly the same thing and so we are sharing architectural ideas about how to do that. Going into this kind of production, large-scale use of AI, everybody’s got the same infrastructure needs and that’s what CANDLE is. We’re debugging it around cancer but we have already started using it for drug design that’s a different problem.

[For example,] one thing CANDLE can do is these large searches. One of the problems for doing drug design is you need to generate libraries of lead-like structures (structures likely to have pharmacological activity). That’s a huge search problem and you need to be able to manage that search problem in a principled way. We built into CANDLE a set of optimizers that are not optimizing the internal parameters of the model but they are optimizing the search. The model is optimizing its own internal parameters but the CANDLE search supervisor, we call it, is using an optimization algorithm to decide which part of this search space to try next based on how well you’ve been doing.

HPCwire: Down to the structure of the molecule?

Rick Stevens: Exactly. We can use CANDLE to optimize the search space for these drugs. You are just trying to generate these molecules. Another interesting factoid is we started incorporating some software from Uber. Uber is moving aggressively on self-driving cars and they collaborated with Nvidia earlier this year to produce a piece of software call Horovod. It comes from a Russian dance which is this kind of funky folk dance that implements a very efficient ring-based sort of communication. They made it open source in a way that is generic so we have incorporated that into CANDLE.

We are going to borrow any piece technology we can get so we just plugged that right in. It turns out that everybody is trying to solve the same problem. If you take away the application, there’s deep learning and I have got a bunch of data and models and you’ve got to try to find the optimal models against my data and I’ve got data that’s dirty and data that’s not balanced and so forth; so the generic technology behind AI is all of the same stuff whether you are working on robotics or on computer driven cars or cancer or choosing ads in Facebook. It’s all the same underlying problem you are trying to solve from a data management and optimization [perspective].

HPCwire: Who’s actually using CANDLE at this point?

Rick Stevens: The first beta release of the whole system was in July and we’ve done some tutorials. It’s installed at NIH and we’ve got probably 20 users there and they are all trying different things and all in the early stages of debugging their machine learning approach. CANDLE is really aimed at groups that kind of know what they are doing. [You don’t] want to burn millions of node hours trying to optimize a model if you have no idea if your model is any good. For people that are just tinkering, CANDLE is not the place to start because you can easily burn up all of your allocation quickly.

The other thing we are doing there is stepping up the work on portable model representation. It turns out there’s three different standards emerging for taking neural network models and making them portable between systems. We were hoping it would be one but it turns out there’s three. There’s two that are coming from the community and Nvidia is doing a third one. NIH has taken the lead on that. Ultimately we want to build deliverables from these projects, models that other people can use. This is on two levels. One is the code for those models. But the other is the model itself, an executable of the model that can be put it some pipeline. We are creating a database of models that are independent of the language used to describe them.

HPCwire: How else is the CANDLE infrastructure being used now?

Rick Stevens: We are also using it to produce very large scale predictions right now. NCI has a high throughput experimental lab where they can do thousands of experiments a day and we want to apply optimal experimental design strategies to those. To do that we not only have to build models but also we have to optimize the models to run in inference mode and then use them to make literally millions of predictions against tumor samples that NCI has that they can do experiments on.

The part of that that is really interesting is we have experimental data for drug combinations. Just doubles. Pairs. They took 100 of the top small FDA compounds and paired them out, so that’s 5,000. It would be 10,000 but we only have to do half the pairs, and you have to do it at like ten doses and in a large number of cell lines. But it is only 100 compounds. We’ve got a database of million compounds that we want to test but we can’t afford to test a million times a million. Nobody is ever going to do that. So the idea is we can train the models up on all the data we have and we run them on these pairs or triplets against 1,000 cell lines and 1,000 xenografts we have – it’s literally billions of predictions that we are making.

HPCwire: Let’s change gears for a second. You’ve said in the past that for ARM to gain a bigger foothold in HPC, it needed to have a clearer accelerator strategy. Do you still think that? What’s your take on ARM’s prospects?

Rick Stevens: ARM is fine. The chips that are out are showing many benchmarks that are comparable to server class Xeon. The 64-bit ARM core probably does not have exactly the same thread performance as the state of the art Xeon but they are not that far behind. The memory architecture is still evolving and the compilers for the server class machines have to get a little bit better. But for everything we’re (CANDLE) doing and that science is doing, not everything but lots of it, you need another order of magnitude of power efficiency and you are not going to get that without adding accelerators.

If you look at where the leadership-class machines are, we are not fielding any machines that are not accelerated in some way. These ARM server class nodes are not manycore in the same way that say Xeon Phi is and they are not GPUs even though they could be paired with GPUs. The few that are out there right now are not NVLink supporting, so it would be a PCIe offloaded model for accelerated stuff.

If the goal for ARM is essentially to be an alternative to Xeon in the computer center, it doesn’t necessarily give you a reason to move to it because 99 percent of your workload is going to be on the accelerator and the host processor is not particularly interesting. Look at the Summit and Sierra machines, the total amount of capability that’s in the accelerator versus the host is [close to] 98 percent in the accelerator. If you are just running on the host you are only using 2 percent of the silicon you have access to. That’s not a particularly good place to be from a price/performance application.

I think it’s important to have this really innovative ecosystem and getting ARM in there is good, because it causes everybody to think harder about where to go. It also gets players that haven’t been in the HPC business before. On the other hand, you’ve got to be able to field a machine that can win bids and so the guys making machines with ARM must have an accelerator strategy so they can win bids. If you are trying to compete in the kind of HPC simulation space or the deep learning space I think it would be very hard to win bids without an accelerator strategy.

HPCwire: Deep learning is often associated with neuromorphic technology and the idea that closely mimicking actual brain neuronal functioning will dramatically cut power and boost performance. Has CANDLE looked at neuromorphic technology?

Rick Stevens: We’re doing some exploration there. We are obviously interested in what Intel (Nervana) is doing and what IBM (True North) is doing. There’s some early results that are mostly from inside of the labs where they are still doing things in emulation or simulation that are pretty encouraging from the standpoint of being able to very efficiently solve problems, from a power and number of neurons used perspective. But there hasn’t been somebody taking a production deep neural network, pick your favorite, and running that on neuromorphic hardware. There’s no proof of principle that we can do that yet.

The principal problem here is that for the deep neural networks we’re using back propagation and stochastic gradient descent or some derivative to train these things, and while you can use back propagation to train neuromorphic hardware, there’s a penalty; it kind of defeats the whole purpose. We’ve got to have a way to train these networks that takes advantage of the kind of synaptic plasticity that’s built in to the designs that are actually trainable. The IBM early (neuromorphic) chips were not trainable. You did all the training off line and then moved them onto the network. The newer chips will be online trainable but how well that will work is not clear. This whole idea of how to train neuromorphic hardware on things that are not model problems is still TBD.

HPCwire: How about quantum computing? There’s more buzz daily. What’s your take on the reality?

An IBM cryostat wired for a prototype 50 qubit system. (PRNewsfoto/IBM)

Rick Stevens: So our interest, it is like the same thing with quantum, you have to track it and the best way to track is to get your hands dirty trying to do it. For what we are doing, quantum computers, as they exist today, are really not appropriate for moving large amounts of data. Quantum computers require you to store a bunch of superpositions and trying to do this is something like machine learning; you have to essentially preload the data and it takes exponential [time]. If you have a lot of ‘n’ different states it is going to take you that many cycles to load the data so that’s kind of the opposite of big data machines, they are like tiny data machine, or no data machines. The best algorithms are ones where there is no data at all. The functions that you are trying to compute, you kind of generate on the fly. Because we don’t have quantum storage, we don’t have quantum communication. It’s very painful to get data inside a quantum computer today.

Now there are ideas people have on how to deal with this. So you use a classical algorithm, non quantum, to train something and then calculate a reduced state of that, in other words a mathematical algorithm function that kind of approximates the function and then try to form an analytical version of that and then you use that to load…I mean there’s all these tricks people are thinking about. But none of it is practical. I think quantum computing is very important but I can’t draw a line now where I say in 2028 we will stop using our 100 exaflops machines, or whatever we are using at that time, and start using quantum machines for this problem. Until we can solve data, they are going to be good for things like quantum simulation where I am using the quantum computer to simulate a quantum system, a quantum chemistry problem for example. The reason you can do that is there is no data. You have a pure algorithmic formulation.

The other question is how big do the qubits have to be. You may need more physical qubits to get one logical qubit because of error management. The problem is when you start going into these larger collections of them, there is also the notion of what’s the topology of the qubits. So for it to be a universal quantum computer, things have to be entangled. That means in some sense each pair of bits has to be able to talk to each other somehow and it’s really hard to do that if you make a linear array. The distance between the edges is quite far so people are thinking of doing these 2D arrays or 2-and-a-half D arrays or one-and-a-half D arrays to try to make it possible for the qubits to entangle each other without having to move states across very large distances because that’s really hard to do. You want these things to be compact. Yet you want all the bits to see each other in some way or see each other in some minimum number of hops so they can entangle each other.

Plus these superconducting devices are, you know, physics experiments. I read the press release on the latest IBM machine. It stays coherent for 90 microseconds or something.

HPCwire: How about D-Wave’s quantum annealing approach?

Rick Stevens: There have been some experiments to show that maybe there’s some quantum speedup there but quantum annealing is a very special case and it’s not clear how many problems we can map into that, number one, and it’s not clear by the time you do that and you end up having to probabilistically solve this thing, that you are getting speedup. So there’s some controversy over that. I’m not saying yes or no there. Just there’s enough controversy that you have to question whether or not it makes any sense. It’s essentially a special purpose machine. Well I’ve got lots of other special purpose machine ideas that we could target but I mean as a physics experiment it needs to keep going.

I think of quantum computing…you know the hype cycle right. The hype cycle has these humps. In quantum computing we are still in this first part. We haven’t fallen into the valley of disillusionment. I think we will fall into there when people realize, ok IBM and Google will knock each other out for awhile, and they will do some mock problems that shows quantum supremacy and they will say, ok, now what. As people start to get more and more understanding of it they’ll say ok there is a class of problems that we can make hardware solve but it’s not nearly as broad as the popular press has made it sound like. At the same time there could be revolutionary advances. The Chicago Quantum Exchange is working on defect based qubits. That’s using an off the shelf technology.

Brief Stevens Bio

Rick Stevens is Argonne’s Associate Laboratory Director for Computing, Environment and Life Sciences. Stevens has been at Argonne since 1982, and has served as director of the Mathematics and Computer Science Division and also as Acting Associate Laboratory Director for Physical, Biological and Computing Sciences. He is currently leader of Argonne’s Petascale Computing Initiative, Professor of Computer Science and Senior Fellow of the Computation Institute at the University of Chicago, and Professor at the University’s Physical Sciences Collegiate Division. From 2000-2004, Stevens served as Director of the National Science Foundation’s TeraGrid Project and from 1997-2001 as Chief Architect for the National Computational Science Alliance.

[i] Generalization Tower Network: A Novel Deep Neural Network Architecture for Multi-Task Learning, https://arxiv.org/abs/1710.10036

The post ANL’s Rick Stevens on CANDLE, ARM, Quantum, and More appeared first on HPCwire.

Supercomputer Simulations Allow Researchers to Understand Characteristics of Diamonds

Fri, 01/05/2018 - 10:10

Jan. 5, 2018 — For centuries diamonds have been revered for their strength, beauty, value and utility. Now a team of researchers from Argonne National Laboratory, running molecular dynamics calculations at the Argonne Leadership Computing Facility (ALCF) and Berkeley Lab’s National Energy Research Scientific Computer Center (NERSC), are finding additional reasons to celebrate this complex material—and it has nothing to do with color, cut or clarity.

In a series of papers published in in ScienceNature and Nature Communications, experimentalists and computational scientists from Argonne’s Center for Nanoscale Materials (CNM) shared several “firsts” in their ongoing efforts to uncover new characteristics in diamond and diamond-like carbons that make these materials even more attractive, particularly for industrial applications.

For example, the Nature study, published in August 2016, highlights their discovery of a revolutionary diamond-like film that is generated by the heat and pressure of an automotive engine. This ultra-durable, self-lubricating tribofilm (a film that forms between moving surfaces) could have profound implications for the efficiency and durability of future engines and other moving metal parts that can be made to develop self-healing, diamond-like carbon tribofilms.

The phenomenon was first discovered several years ago through experiments conducted by researchers in the Tribology and Thermal-Mechanics Department in Argonne’s Center for Transportation Research. But it took theoretical insight using supercomputing resources to fully understand what was happening at the molecular level in the experiments. Argonne nanoscientist Subramanian Sankaranarayanan and postdoctoral researcher Badri Narayanan ran molecular dynamics simulations on Argonne’s Mira system and NERSC’s Edison system to understand what was happening at the atomic level. These calculations helped them determine that the catalyst metals in the nanocomposite coatings were stripping hydrogen atoms from the hydrocarbon chains of the lubricating oil, then breaking the chains down into smaller segments. The smaller chains then joined together under pressure to create the highly durable DLC tribofilm.

“This is an example of catalysis under extreme conditions created by friction. It is opening up a new field where you are merging catalysis and tribology, which has never been done before,” said Sankaranarayanan. “This new field of tribocatalysis has the potential to change the way we look at lubrication.”

In the Nature Communications study, published in July 2016, a team of Argonne and University of California-Riverside researchers once again used a combination of experiments and molecular dynamics simulations to demonstrate how diamond—in this case ultrananocrystalline diamond that serves as a substrate—can be used to grow graphene that contains relatively few impurities and costs less to make, in shorter time and at lower temperatures compared to the process widely used to make graphene today. Current graphene fabrication protocols introduce impurities during the etching process itself, which involves adding acid and extra polymers, and when they are transferred to a different substrate for use in electronics. These impurities negatively affect the electronic properties of the graphene, the researchers noted.

The simulations—which were developed by Sankaranarayanan and his post-docs, Badri Narayanan and Sanket Deshmukh, and utilized 300,000 to 500,000 node hours at NERSC in addition to computing time at Argonne—helped the team understand the molecular-level processes underlying graphene growth. They ran three different sets of calculations on NERSC’s Edison supercomputer to tease out the sequence of events leading to graphene nucleation on nickel and to determine what kind of graphene structures can grow on different crystal orientations.

“NERSC is a very good resource to have because it allows the flexibility to do intermediate, production-run calculations,” Sankaranarayanan said. “In this example, you have a lot of things happening mechanistically, and the experimentalists have an end point and the time scales involved are quite fast. But they have not yet reached a stage where in situ experiments can be performed on these kinds of rapidly evolving interfaces, and they want to understand the dynamics of what is happening at the nanosecond and microsecond time scales. It is this dynamical evolution that the experimentalists want us to simulate.”

In an earlier, related study published in Science, the Argonne team described how a series of molecular dynamics simulations paved the way for the design of a near-frictionless hybrid material. The research team again used a combination of experiments and simulations to demonstrate that superlubricity can be realized at engineering scale when graphene is used in combination with nanodiamond particles and diamond-like carbon. Considering that nearly one-third of every fuel tank is spent overcoming friction in automobiles, a material that can achieve superlubricity would greatly benefit industry and consumers alike.

“The beauty of this particular discovery is that we were able to see sustained superlubricity at the macroscale for the first time, proving this mechanism can be used at engineering scales for real-world applications,” Sankaranarayanan said. “It was really a big breakthrough that purely came out of calculations that we did initially at NERSC and then at ACLF.”

About NERSC and Berkeley Lab

The National Energy Research Scientific Computing Center (NERSC) is a U.S. Department of Energy Office of Science User Facility that serves as the primary high-performance computing center for scientific research sponsored by the Office of Science. Located at Lawrence Berkeley National Laboratory, the NERSC Center serves more than 6,000 scientists at national laboratories and universities researching a wide range of problems in combustion, climate modeling, fusion energy, materials science, physics, chemistry, computational biology, and other disciplines. Berkeley Lab is a DOE national laboratory located in Berkeley, California. It conducts unclassified scientific research and is managed by the University of California for the U.S. DOE Office of Science. »Learn more about computing sciences at Berkeley Lab.

Source: Kathy Kincade, NERSC and Berkeley Lab

The post Supercomputer Simulations Allow Researchers to Understand Characteristics of Diamonds appeared first on HPCwire.

Chip Flaws Meltdown and Spectre Loom Large

Thu, 01/04/2018 - 17:23

The HPC and wider tech community have been abuzz this week over the discovery of critical design flaws that impact virtually all contemporary microprocessors. The bugs leave memory content open to malicious theft. Worse yet the fixes for these flaws are either unclear at this point or likely to incur significant slowdowns.

As the story evolved, many reports centered on the “Intel chip flaw” but the problem is much bigger than that and impacts AMD and ARM CPUs as well. The New York Times has done a great job of putting all the moving pieces together.

There are two major flaws, the Times reports. The first, dubbed Meltdown, has currently only been shown to impact Intel microprocessors (due to a type of speculative execution that Intel chips allow, covered comprehensively by Ars Technica). The Linux patch, called KPTI (formerly KAISER), has been shown to slow performance speeds of processors by as much as 30 percent, depending on the application.

The second issue, called Spectre, is conceivably even more problematic as it affects virtually all chip lines on the market, leaving potentially billions of devices, including phones, vulnerable to exploits. “Researchers believe this flaw is more difficult to exploit. There is no known fix for it and it is not clear what chip makers like Intel will do to address the problem,” wrote the Times.

Intel released a statement yesterday downplaying the ramifications and emphasizing that competing chips are also affected.

“Intel and other technology companies have been made aware of new security research describing software analysis methods that, when used for malicious purposes, have the potential to improperly gather sensitive data from computing devices that are operating as designed. Intel believes these exploits do not have the potential to corrupt, modify or delete data,” the company asserted.

“Recent reports that these exploits are caused by a ‘bug’ or a ‘flaw’ and are unique to Intel products are incorrect. Based on the analysis to date, many types of computing devices — with many different vendors’ processors and operating systems — are susceptible to these exploits.”

Intel went on to say that for the “average computer user,” performance impacts “should not be significant and will be mitigated over time.”

This prompted one contributor to a popular HPC mailing list to respond: “We, ‘non-average computer users,’ are still [verb of your choice here].”

As this issue was still coming to light, the US government issued a dire statement (on Jan. 3), implying the problematic CPUs were essentially unsalvageable. “The underlying vulnerability is primarily caused by CPU architecture design choices. Fully removing the vulnerability requires replacing vulnerable CPU hardware,” wrote US-CERT, the computer safety division of Homeland Security.

A revised version of the notice offers less extreme, but vague, guidance. Affected parties are now advised that “operating system and some application updates mitigate these attacks.”

There is still a lot of uncertainty about the full ramifications of these major flaws. AMD and ARM have also released statements:

AMD: https://www.amd.com/en/corporate/speculative-execution

ARM: https://developer.arm.com/support/security-update

The impacted tech companies have known about the flaws for months and have been working to solve the issues before a public disclosure. This is common practice to stay ahead of hackers, but the timing is bringing attention to a major stock sale made late last year by Intel CEO Brian Krzanich. On November 29, Krzanich sold off $24 million worth of company stock and options, reducing his share down to the bare minimum required by his contract with Intel. The scope of the transactions were within permissible bounds, but could draw additional scrutiny now with regard to the timing of the sell-off and the potential for hardware vulnerabilities to impact stock prices. A spokesperson for Intel said Krzanich’s sale was “unrelated.”

Computing professionals have taken to mailing lists, social media forums and message boards to vent frustrations and discuss strategies for balancing security interests with performance mandates. There is already talk of seeking compensation for lost performance. This is especially relevant when it comes to HPC systems, which not only comprise thousands of nodes, but have workloads and usage patterns that make them targets for the higher-range penalties.

Additional reading:

https://spectreattack.com/

https://meltdownattack.com/

Ground zero post:

http://pythonsweetness.tumblr.com/post/169166980422/the-mysterious-case-of-the-linux-page-table

Meltdown and Spectre logos were designed by  Natascha Eibl and used above via Creative Commons license. 

The post Chip Flaws Meltdown and Spectre Loom Large appeared first on HPCwire.

Mellanox Ships BlueField System-on-Chip Platforms and SmartNIC Adapters to OEMs and Hyperscale Customers

Thu, 01/04/2018 - 11:46

SUNNYVALE, Calif. & YOKNEAM, Israel, Jan. 4, 2018 — Mellanox Technologies, Ltd. (NASDAQ: MLNX), a leading supplier of high-performance, end-to-end smart interconnect solutions for data center servers and storage systems, today announced the first shipments of its BlueField system-on-chip (SoC) platforms and SmartNIC adapters to major data center, hyperscale and OEM customers. Mellanox BlueField dual-port 100Gb/s SoC is ideal for cloud, Web 2.0, Big Data, storage, enterprise, high-performance computing, and Network Functions Virtualization (NFV) applications.

BlueField sets new NVMe-over-Fabrics performance records, demonstrating seven and a half million IOPS during initial testing, with zero CPU utilization. Furthermore, BlueField delivers under three microseconds of NVMe latency to enable less than five microseconds of additional latency for end to end access to remote NVMe device over a local NVMe device. BlueField’s NVMe over Fabrics advanced hardware acceleration offload guarantees maximum performance with no CPU utilization, thereby improving system total cost of ownership (TCO). In addition, BlueField delivers up to a smashing close to 400Gb/s of RDMA bidirectional traffic bandwidth over dual 100Gb/s ports.

“We are excited to ship BlueField systems and SmartNIC adapters to our major customers and partners, enabling them to build the next generation of storage, cloud, security and other platforms and to gain a competitive advantage,” said Yael Shenhav, vice president of products at Mellanox Technologies. “BlueField products achieve new performance records, delivering industry-leading NVMe over Fabrics and networking throughput. We are proud of our world-class team for delivering these innovative products, designed to meet the ever growing needs of current and future data centers.”

The BlueField family of products is a highly integrated system-on-a-chip optimized for NVMe storage systems, Network Functions Virtualization (NFV), security systems, and embedded appliances. BlueField dual port 100Gb/s SoC solutions combine Mellanox’s leading ConnectX-5 network acceleration technology with an array of high-performance 64-bit Arm A72 processor cores and a PCIe Gen3 and Gen4 switch.

About Mellanox

Mellanox Technologies (NASDAQ: MLNX) is a leading supplier of end-to-end InfiniBand and Ethernet smart interconnect solutions and services for servers and storage. Mellanox interconnect solutions increase data center efficiency by providing the highest throughput and lowest latency, delivering data faster to applications and unlocking system performance capability. Mellanox offers a choice of fast interconnect products: adapters, switches, software and silicon that accelerate application runtime and maximize business results for a wide range of markets including high performance computing, enterprise data centers, Web 2.0, cloud, storage and financial services. More information is available at: www.mellanox.com.

Source: Mellanox

The post Mellanox Ships BlueField System-on-Chip Platforms and SmartNIC Adapters to OEMs and Hyperscale Customers appeared first on HPCwire.

The @hpcnotes Predictions for HPC in 2018

Thu, 01/04/2018 - 08:48

I’m not averse to making predictions about the world of High Performance Computing (and Supercomputing, Cloud, etc.) in person at conferences, meetings, causal conversations, etc.; however, it turns out to be a while since I have stuck my neck out and widely published my predictions for the year ahead in HPC. Of course, such predictions tend to be evenly split between inspired foresight and misguided idiocy. At least some of the predictions will have readers spluttering coffee in indignation at how wrong I am. But, where would the fun in HPC be if we all played safe? So, here goes for the @hpcnotes predictions for HPC in 2018 …

Intel

After spending much of 2017 being called out for ambitiously high pricing of Skylake for HPC customers, and following that with the months of Xeon Phi confusion – eventually publicly admitting at SC17 that Knights Hill has been cancelled, still not clear about the future of Phi overall – Intel seems to have continued into 2018 in the worst way, with news of kernel memory hardware bugs flooding the IT news and social media space. [NB: these bugs have now been confirmed to affect CPUs from AMD, ARM and other vendors too.] 2018 will also see widespread availability of AMD EPYC, Cavium ThunderX2, and IBM Power9 processors and so it seems Intel has a tough year ahead. The hardware bug is especially painful here as it negates the “Intel is the safe option” thinking. To be clear, HPC community consensus so far (including NAG’s impartial benchmarking work with customer codes) says Skylake is a very capable and performance leading processor. However, Skylake has three possible let downs: (1) price substantially higher, relative to the benefits gained, than customers are comfortable with; (2) reduced cache per core compared with other CPUs; (3) dependence on a code’s saturation of the vector units to extract the maximum performance. In some early benchmarks, EPYC and TX2 are winning on both price and performance. My prediction is that Intel will meaningfully drop the Skylake price early in 2018 to pull back into a competitive position on price/performance.

AI and ML

Sorry, the media and marketing hype for AI/ML taking over HPC shows no sign of going away. Yes, there are many real use cases for AI and ML (e.g., follow Paige Bailey and colleagues for real examples); however, the aggressive insertion of AI and ML labels into every HPC-related conference agenda (taking over from the mandatory mentions of Big Data) doesn’t add a lot of value, I think. I’m not suggesting that the HPC community (users or providers) ignore AI/ML – indeed, I would firmly advocate that you add these to your portfolio. But, HPC is an exceptionally powerful and widely applicable tool in its own right – it doesn’t need AI/ML to justify itself. My prediction is that AI/ML will continue to hog a share of the HPC marketing noise unrelated to the scale of actual use in the HPC arena.

New processors

As noted above, 2018 sees credible HPC processors from AMD (EPYC), Cavium (ThunderX2) and other ARM chips, and IBM (Power9) surge into general availability. In my view, these are not (yet) competing with Intel Xeon; they are competing with each other to be the best of the rest. Depending on how Intel behaves (NB: this is not just about technology) and how well AMD/ARM/IBM and their system partners actually execute on promises, one of these might close out 2018 being a serious competitor to Intel’s dominance of the HPC processor space. Either way, I predict we will see at least one meaningful (i.e., competitively won, large scale, for production use) HPC deployment of each of these processors in 2018. I’m also going to add a second prediction to this section: a MIPS based processor option will start to gain headlines as a real HPC processor candidate in 2018 (not just in China).

Cloud

In most cases, HPC is still cheaper and more capable through traditional in-house systems than via cloud deployments. No amount of marketing changes that. Time might change it, but not by the end of 2018. However, cloud as an option for HPC is not going away. It does present a real option for many HPC workloads, and not just trivial workloads. I am hopeful we are at the end of the era where the cloud providers hoped to succeed by trying to convince everyone that “HPC in-house” advocates were just dinosaurs. The cloud companies all show signs of adjusting their offerings to the actual needs of HPC users (technical, commercial and political needs). This means that an impartial understanding of the pros and cons of cloud for your specific HPC situation is going to be even more critical in 2018. I am certainly being asked to help address the question of HPC in the cloud by my consulting customers with increasing frequency. Azure has been ramping up efforts in HPC (and AI) aggressively over the last few months through acquisitions (e.g., Cycle Computing) and recruitments (e.g., Developer Advocate teams), and I’d expect AWS and Google to do likewise. My prediction is that all three of the major cloud providers (AWS, Azure, Google) will deliver substantially more HPC-relevant solutions in 2018, and at least one will secure a major (and possibly surprising) real HPC customer win.

GPUs

Nvidia also got an unwelcome start to 2018 as they tried to ban (via retrospective changes to license conditions) the use of their cheaper GPUs in datacenter (e.g., HPC, AI, …) applications. Of course, it is no surprise that Nvidia would prefer customers to buy the much more expensive high-end GPUs for datacenter applications. However, it doesn’t say much for the supposedly compelling business case or sales success of the high-end GPUs if they have to force people off the cheaper products first. We (NAG) have done enough benchmarking across enough different customer codes to know that GPUs are flat-out the fastest widely available processor option for codes that can take effective advantage of highly parallel architectures. However, when price of the high-end GPUs is taken into account, plus the performance left on the floor for the non-accelerated codes, then the CPUs often look a better overall choice. Ultimately, adapting many codes to use GPUs (not just a selected few codes to show easy wins) is a big effort. So is adapting workflows to the cloud. With limited resources available, I think users will decide that investing effort in cloud porting is a better long-term return than GPUs. Yes – oddly, I think cloud, not CPUs, will be the pressure that limits the success of GPUs! My prediction is that Nvidia’s unfortunate licensing assertions, coupled with marginal gains in performance relative to total cost of ownership (TCO), plus scarcity of software engineering resources, is that fewer newly deployed on-site HPC systems will be based around GPUs. On the other hand, I think use of GPUs in the cloud, for HPC, will grow substantially in 2018.

Zettascale

Yes, really. After all, exascale is within grasping distance now. We will see multiple systems at >0.1 EF in 2018. Exascale is being talked about in terms of when and which site first, rather than how and which country first. As exascale now seems likely to happen without all those disruptive changes that voices across the community foretold would be critical, computer science researchers and supercomputer center managers will need to start using the zettascale label to drive the next round of funding bids for novel technologies. There have already been a few small gatherings on zettascale, at least as far back as 2004 (!), but I predict 2018 will see the first mainstream meeting with a session focused on zettascale – perhaps at SC18?

Cybersecurity

The consumer world was wracked in 2017 by a range of large scale cybersecurity breaches. The government community has been hit badly in previous years too. Sadly, I see cybersecurity moving up the agenda in the HPC world. Not sad that it is happening, but sad that I think it will be forced to happen by one or more incidents. In general, HPC systems are fairly well protected, largely because they are expensive, capable assets and, in some cases, have regulatory criteria to meet. However, performance and ease-of-use for a predominantly research-led userbase have been the traditional strong drivers of requirements, often meaning the risk management decisions have been tilted towards a minimally compliant security configuration. (Security is arguably one area where HPC-in-the-cloud wins.) My prediction for 2018 is twofold: (1) there will be a major security incident on a high profile HPC system; (2) cybersecurity for HPC will move from a niche topic to a mainstream agenda item for some of the larger HPC conferences.

Finally, Growth

I saw HPC and related things such as AI, cloud, etc., gain lots of momentum in 2017. This included several technologies heralded in confidence finally coming to fruition, new HPC deployments across public and private sectors customers, a notable uptick in our HPC consulting work, interesting personnel moves, and an overall excitement and enthusiasm in the HPC community that had been dulled recently. My final prediction is that 2018 will see this growth and energy in the HPC community gather pace. I look forward to new HPC sites emerging, to significant new HPC systems being announced, and to the growing attention on the broader aspects of HPC beyond FLOPS – people, business aspects, impact stories, and more.

I hope you enjoyed my HPC predictions for 2018. Please do engage with me via Twitter (@hpcnotes) or LinkedIn (www.linkedin.com/in/andrewjones) if you want to comment on my inspired foresight or misguided idiocy. I’ll be back with a follow-up article in a week or two on how you can exploit these predictions to your advantage.

The post The @hpcnotes Predictions for HPC in 2018 appeared first on HPCwire.

PASC18 Outlines Opportunities for Student Participation, Issues Reminder for its Call for Proposals

Thu, 01/04/2018 - 08:12

Jan. 4, 2018 — The PASC18 Organizing Team has begun the New Year with an announcement about opportunities for student participation, as well as with a reminder that submission deadlines are rapidly approaching.

Opportunities for student participation

TRAVEL GRANTS

PASC18 is announcing that this year for the first time, it will offer travel grants to enable two undergraduate or postgraduate students to attend the conference. The travel grants are generously provided by SIGHPC, with PASC18 covering the corresponding registration fees.

Applications are due by February 14, 2018, and further information on the application process is available at: pasc18.pasc-conference.org/about/student-travel-grants/

STUDENT VOLUNTEER PROGRAM

Submissions for the Student Volunteer Program are now open.

PASC18 is looking for enthusiastic students who are interested in helping us with the administration of the event. Selected students will be granted a complimentary registration for the conference.

Further information on this opportunity is available at: pasc18.pasc-conference.org/about/student-volunteer-program/

Call for submissions reminder

CONFERENCE

PASC18, co-sponsored by the Association for Computing Machinery (ACM), is the fifth edition of the PASC Conference series, an international platform for the exchange of competences in scientific computing and computational science, with a strong focus on methods, tools, algorithms, application challenges, and novel techniques and usage of high performance computing.

PASC18 welcomes submissions in the form of minisymposiapapers and posters. Contributions should demonstrate innovative research in scientific computing related to the following domains:

  • Chemistry and Materials
  • Life Sciences
  • Physics
  • Climate and Weather
  • Solid Earth Dynamics
  • Engineering
  • Computer Science and Applied Mathematics
  • Emerging Applications Domains (e.g. Social Sciences, Finance, …)

Submissions that are interdisciplinary in nature are strongly encouraged.

Full submission guidelines are available at: pasc18.pasc-conference.org/submission/submissions-portal/

Submissions are received through the online submission portal and the rapidly approaching submission deadlines are listed below: 

  • Minisymposia: January 7, 2018
  • Papers: January 19, 2018
  • Posters: February 4, 2018 

CONFERENCE CHAIRS

  • Florina Ciorba (University of Basel, Switzerland)
  • Erik Lindahl (Stockholm University, Sweden)

MINYSIMPOSIA AND POSTERS PROGRAM CHAIRS

  • Florina Ciorba (University of Basel, Switzerland)
  • Erik Lindahl (Stockholm University, Sweden)
  • Sabine Roller (University of Siegen, Germany)
  • Jack Wells (Oak Ridge National Laboratory, US)

PAPERS PROGRAM CHAIRS

  • Sabine Roller (University of Siegen, Germany)
  • Jack Wells (Oak Ridge National Laboratory, US)

PASC18 Scientific Committee: pasc18.pasc-conference.org/about/organization

Further information on the conference is available at: pasc18.pasc-conference.org/

Source: PASC18

The post PASC18 Outlines Opportunities for Student Participation, Issues Reminder for its Call for Proposals appeared first on HPCwire.

Microsoft Extends Hybrid Cloud Push with Avere Deal

Wed, 01/03/2018 - 23:00

Microsoft continued its foray into the high-end cloud storage sector with a deal this week to acquire hybrid cloud data storage and management vendor Avere Systems.

The deal announced on Wednesday (Jan. 3) follows Microsoft’s acquisition last August of Cycle Computing to bolster its “big compute” initiatives on the Azure cloud. Microsoft said the Cycle and Avere deals are part of its strategy to bring high-end computing to hybrid cloud deployments. (Note that Cycle and Avere had entered into a technology partnership a year and a half ago, so there was some history here already).

Terms of the deal for Pittsburgh-based Avere Systems were not disclosed.

Avere’s scalable cloud storage platform dubbed FXT Edge Filers targets enterprises trying to integrate applications requiring file systems into the cloud. Along with data storage access, the platform helps scale computing and storage depending on application requirements.

Jason Zander, vice president of Microsoft Azure, said the acquisition gives the public cloud vendor a combination of file system and caching technologies. Avere works with animation studios that run computing intensive workloads, and the deal is expected to give Microsoft entrée into those media and entertainment sectors.

“By bringing together Avere’s storage expertise with the power of Microsoft’s cloud, customers will benefit from industry-leading innovations that enable the largest, most complex high-performance workloads to run in Microsoft Azure,” Zander asserted in a blog post announcing the deal.

Avere Systems CEO Ron Bianchini added that the acquisition would expand the reach of its data storage technology from datacenters and public clouds to hybrid cloud storage and “cloud-bursting environments.”

Along with the entertainment sector, Avere Systems’ customers include the Library of Congress, Johns Hopkins University and automated test equipment manufacturer Teradyne. Its customer base also includes financial services and oil and gas customers along with the education, healthcare and manufacturing sectors.

Last year, Avere Systems announced an investment round that included Microsoft cloud rival Google. It also disclosed partnerships with private cloud storage vendors, including support for Dell EMC’s Elastic Cloud Storage platform. The partners said the software-defined object storage approach would allow private cloud customers to consolidate content archives and file storage systems in a central repository.

Meanwhile, Microsoft’s August 2017 acquisition of Cycle Computing combined the startup’s orchestration technology for managing Linux and Windows computing and data workloads with Microsoft’s Azure cloud computing infrastructure.

Observers praised Microsoft’s acquisition of Avere Systems, noting that its storage technology could help Microsoft boost its Azure revenues by allowing customers to use the public cloud while keeping some data on-premises.

The post Microsoft Extends Hybrid Cloud Push with Avere Deal appeared first on HPCwire.

Hyperion Sets Agenda for HPC User Forum in Europe

Wed, 01/03/2018 - 12:46

Regional HPC strategies, including the perennial jostling for sway among Europe, the U.S., and Japan, will highlight the HPC User Forum, Europe, scheduled for March 6-7 in France. Hyperion Research, organizer of the event and administrator of the HPC User Forum, just released the preliminary agenda. Besides global competition, AI, industrial HPC use, and a dinner talk on quantum computing are all on the docket.

“Because of China’s rapid rise in the Top500 rankings and the Chinese government’s well-funded plans to deploy the first exascale system, many HPC observers see future supercomputing leadership as a two-horse race between relative newcomer China and long-time leader the United States. In reality, it’s at least a four-horse race that includes two other serious contenders, Europe and Japan,” says Steve Conway, SVP research, Hyperion.

Steve Conway, Hyperion SVP

“That’s one of the things we’ll highlight at the March HPC User Forum meeting: the global nature of the push to advance the state-of-the-art in supercomputing. We’ll have senior officials from Europe, Japan and the U.S.  We haven’t secured a speaker from China, but we’d welcome one. Once you abandon the idea that leadership is limited to Linpack performance, it becomes clear that each of these four contenders is likely to be the future supercomputing leader in some important respects and to advance the state-of-the-art in supercomputing. I’m talking about the 2023-2024 era, when we’re likely to see productive exascale supercomputers from multiple parts of the world, rather than just smaller-scale prototypes and early machines.”

Here are a few agenda highlights:

  • European HPC Strategy, Thomas Skordas or Leonardo Flores, European Commission
  • Japan’s Flagship 2020 Project, Shig Okaya, Flagship 2020/RIKEN
  • View from America, Dimitri Kuznesov and Barbara Helland, DOE
  • HPC-based AI/Deep Learning in the Commercial World, Arno Kolster, Providentia Worldwide
  • Worldwide Study on HPC Centers and Industrial Users, Steve Conway, Hyperion
  • Exascale Computing Project Update, Doug Kothe

According to Conway, the U.S. has a hefty lead in processors and accelerators, but he expects gains for ARM-based designs and indigenous Chinese processors. “When it comes to highly scalable system and application software, the U.S. and Europe stand out, with each excelling in different scientific domains. Europe is also very strong in scalable software for industrial applications. The U.S., Japan and Europe have the largest, most experienced HPC user communities but China is gaining ground quickly. Historically, Japan has shown the ability to mount herculean efforts and jump to the head of the pack,” he says.

So, the race is on. Interestingly, there’s still a fair amount of international collaboration. Rougly a year ago, France’s CEA and Japan’s RIKEN announced they would join forces to advance the ARM ecosystem. CEA and Teratec are co-hosting the March User Forum meeting. There are also initiatives such as the International Exascale Software Project. The exascale race promises to be more multi-dimensional than just a Linpack bake-off says Conway.

This meeting, held at the Très Grand Centre de Calcul du CEA (TGCC) on the Teratec campus in Bruyères-le-Châtel, France (close to Paris), will be the first HPC User Forum run by Hyperion since becoming an independent company in December. Registration for the conference is free. For more information about registering: http://www.hpcuserforum.com

The post Hyperion Sets Agenda for HPC User Forum in Europe appeared first on HPCwire.

TACC Supercomputers Help Researchers Design Patient-Specific Cancer Models

Wed, 01/03/2018 - 11:20

Jan. 3, 2018 — Attempts to eradicate cancer are often compared to a “moonshot” — the successful effort that sent the first astronauts to the moon.

But imagine if, instead of Newton’s second law of motion, which describes the relationship between an object’s mass and the amount of force needed to accelerate it, we only had reams of data related to throwing various objects into the air.

This, says Thomas Yankeelov, approximates the current state of cancer research: data-rich, but lacking governing laws and models.

The solution, he believes, is not to mine large quantities of patient data, as some insist, but to mathematize cancer: to uncover the fundamental formulas that represent how cancer, in its many varied forms, behaves.

Model of tumor growth in a rat brain before radiation treatment (left) and after one session of radiotherapy (right). The different colors represent tumor cell concentration, with red being the highest. The treatment reduced the tumor mass substantially (Lima et. al. 2017, Hormuth et. al. 2015).

“We’re trying to build models that describe how tumors grow and respond to therapy,” said Yankeelov, director of the Center for Computational Oncology at The University of Texas at Austin (UT Austin) and director of Cancer Imaging Research in the LIVESTRONG Cancer Institutes of the Dell Medical School. “The models have parameters in them that are agnostic, and we try to make them very specific by populating them with measurements from individual patients.”

The Center for Computational Oncology (part of the broader Institute for Computational Engineering and Sciences, or ICES) is developing complex computer models and analytic tools to predict how cancer will progress in a specific individual, based on their unique biological characteristics.

In December 2017, writing in Computer Methods in Applied Mechanics and Engineering, Yankeelov and collaborators at UT Austin and Technical University of Munich, showed that they can predict how brain tumors (gliomas) will grow and respond to X-ray radiation therapy with much greater accuracy than previous models. They did so by including factors like the mechanical forces acting on the cells and the tumor’s cellular heterogeneity. The paper continues research first described in the Journal of The Royal Society Interface in April 2017.

“We’re at the phase now where we’re trying to recapitulate experimental data so we have confidence that our model is capturing the key factors,” he said.

To develop and implement their mathematically complex models, the group uses the advanced computing resources at the Texas Advanced Computing Center (TACC). TACC’s supercomputers enable researchers to solve bigger problems than they otherwise could and reach solutions far faster than with a single computer or campus cluster.

According to ICES Director J. Tinsley Oden, mathematical models of the invasion and growth of tumors in living tissue have been “smoldering in the literature for a decade,” and in the last few years, significant advances have been made.

“We’re making genuine progress to predict the growth and decline of cancer and reactions to various therapies,” said Oden, a member of the National Academy of Engineering.

Model Selection and Testing

Over the years, many different mathematical models of tumor growth have been proposed, but determining which is most accurate at predicting cancer progression is a challenge.

In October 2016, writing in Mathematical Models and Methods in Applied Sciences, the team used a study of cancer in rats to test 13 leading tumor growth models to determine which could predict key quantities of interest relevant to survival, and the effects of various therapies.

They applied the principle of Occam’s razor, which says that where two explanations for an occurrence exist, the simpler one is usually better. They implemented this principle through the development and application of something they call the “Occam Plausibility Algorithm,” which selects the most plausible model for a given dataset and determines if the model is a valid tool for predicting tumor growth and morphology.

The method was able to predict how large the rat tumors would grow within 5 to 10 percent of their final mass.

“We have examples where we can gather data from lab animals or human subjects and make startlingly accurate depictions about the growth of cancer and the reaction to various therapies, like radiation and chemotherapy,” Oden said.

The team analyzes patient-specific data from magnetic resonance imaging (MRI), positron emission tomography (PET), x-ray computed tomography (CT), biopsies and other factors, in order to develop their computational model.

Each factor involved in the tumor response — whether it is the speed with which chemotherapeutic drugs reach the tissue or the degree to which cells signal each other to grow — is characterized by a mathematical equation that captures its essence.

“You put mathematical models on a computer and tune them and adapt them and learn more,” Oden said. “It is, in a way, an approach that goes back to Aristotle, but it accesses the most modern levels of computing and computational science.”

The group tries to model biological behavior at the tissue, cellular and cell signaling levels. Some of their models involve 10 species of tumor cells and include elements like cell connective tissue, nutrients and factors related to the development of new blood vessels. They have to solve partial differential equations for each of these elements and then intelligently couple them to all the other equations.

“This is one of the most complicated projects in computational science. But you can do anything with a supercomputer,” Oden said. “There’s a cascading list of models at different scales that talk to each other. Ultimately, we’re going to need to learn to calibrate each and compute their interactions with each other.”

From Computer to Clinic

The research team at UT Austin — which comprises 30 faculty, students, and postdocs — doesn’t only develop mathematical and computer models. Some researchers work with cell samples in vitro; some do pre-clinical work in mice and rats. And recently, the group has begun a clinical study to predict, after one treatment, how an individual’s cancer will progress, and use that prediction to plan the future course of treatment.

At Vanderbilt University, Yankeelov’s previous institution, his group was able to predict with 87 percent accuracy whether a breast cancer patient would respond positively to treatment after just one cycle of therapy. They are trying to reproduce those results in a community setting and extend their models by adding new factors that describe how the tumor evolves.

The combination of mathematical modeling and high-performance computing may be the only way to overcome the complexity of cancer, which is not one disease but more than a hundred, each with numerous sub-types.

“There are not enough resources or patients to sort this problem out because there are too many variables. It would take until the end of time,” Yankeelov said. “But if you have a model that can recapitulate how tumors grow and respond to therapy, then it becomes a classic engineering optimization problem. ‘I have this much drug and this much time. What’s the best way to give it to minimize the number of tumor cells for the longest amount of time?'”

Computing at TACC has helped Yankeelov accelerate his research. “We can solve problems in a few minutes that would take us 3 weeks to do using the resources at our old institution,” he said. “It’s phenomenal.”

According to Oden and Yankeelov, there are very few research groups trying to sync clinical and experimental work with computational modeling and state-of-the-art resources like the UT Austin group.

“There’s a new horizon here, a more challenging future ahead where you go back to basic science and make concrete predictions about health and well-being from first principles,” Oden said.

Said Yankeelov: “The idea of taking each patient as an individual to populate these models to make a specific prediction for them and someday be able to take their model and then try on a computer a whole bunch of therapies on them to optimize their individual therapy — that’s the ultimate goal and I don’t know how you can do that without mathematizing the problem.”

The research is supported by National Science Foundation, the U.S. Department of Energy, the National Council of Technological and Scientific Development, Cancer Prevention Research Institute of Texas and the National Cancer Institute.

Source: Aaron Dubrow, TACC

 

The post TACC Supercomputers Help Researchers Design Patient-Specific Cancer Models appeared first on HPCwire.

Microsoft to Acquire Avere Systems

Wed, 01/03/2018 - 07:17

Jan. 3, 2018 — Microsoft has signed an agreement to acquire Avere Systems, a leading provider of high-performance NFS and SMB file-based storage for Linux and Windows clients running in cloud, hybrid and on-premises environments.

Avere uses an innovative combination of file system and caching technologies to support the performance requirements for customers who run large-scale compute workloads. In the media and entertainment industry, Avere has worked with global brands including Sony Pictures Imageworks, animation studio Illumination Mac Guff and Moving Picture Company (MPC) to decrease production time and lower costs in a world where innovation and time to market is more critical than ever.

High performance computing needs however do not stop there. Customers in life sciences, education, oil and gas, financial services, manufacturing and more are increasingly looking for these types of solutions to help transform their businesses. The Library of Congress, John Hopkins University and Teradyne, a developer and supplier of automatic test equipment for the semiconductor industry, are great examples where Avere has helped scale datacenter performance and capacity, and optimize infrastructure placement.

Microsoft says that by bringing together Avere’s storage expertise with the power of Microsoft’s cloud, customers will benefit from industry-leading innovations that enable the largest, most complex high-performance workloads to run in Microsoft Azure.

Source: Microsoft

The post Microsoft to Acquire Avere Systems appeared first on HPCwire.

ASC18 Competition Timeline Released

Tue, 01/02/2018 - 07:14

Jan. 2, 2018 — The 2018 ASC Student Supercomputer Challenge (ASC18) has released its competition timeline (official website: http://www.asc-events.org/ASC18/index.php), with a starting date set for January 16, 2018. Making the announcement, the ASC18 organizers also promised that the upcoming competition would include a well-known AI application in one of its challenges, a tantalizing clue to the dozens of entrants from China, America, Europe and around the world who are expected to attend.

The timeline sets January 15, 2018, as the deadline for registration, before which all entrants must submit an application on the ASC website. Each team entering the competition must include one mentor and five undergraduate students, with multiple teams from a single university allowed. Competition will begin with the preliminary round, set to kick off on January 16 and conclude on March 20. Within this timeframe, participating teams must submit their written solutions and application optimization according to ASC requirements. Each entry will be scored by the judging panel. With the top 20 teams advancing to the final round. For five days, from May 5 to May 9, these teams will compete face-to-face for the ASC18 Grand Prix.

But while releasing the ASC18 timeline, the competition’s organizers have yet to reveal several key details, hoping to spark the curiosity of young supercomputing enthusiasts around the world. The organizing committee noted ASC18 will continue with past precedent in cooperating with an ultra-large-scale supercomputer system to serve as the competition platform. But no indication was given as to whether this would be the Gordon Bell Prize-winning Sunway TaihuLight featured in last year’s competition or another system altogether. Organizers have likewise not yet revealed the location of the upcoming competition.

Whatever surprises organizers have in store, ASC18 is sure to make waves as it follows past competitions in breaking new ground. The inaugural ASC in 2012 was the first event of its kind to use a world-class supercomputer system as its competition platform, while in 2014 ASC introduced Tianhe II, the world’s fastest supercomputer.

ASC has also given participating teams the chance to engage in major international science projects. In 2015, the competition partnered with SKA, the world’s largest radio telescope project. In 2016, ASC used the Gordon Bell Prize-winning numerical simulation of high-resolution waves to present teams with a challenge relating to driverless vehicles.

With its ever-increasing challenges and unmatched opportunities for participants, the ASC has won praise from numerous leading names in the supercomputing field. “The US’ SC is like a marathon, testing participants’ hard work and perseverance; Germany’s ISC is a sprint, testing innovation and adaptability. China’s ASC is a combination of both,” said OrionX partner and HPCwire correspondent Dan Olds, who has covered all three major supercomputing challenges.

Jack Dongarra, founder of TOP500, a ranking of the world’s most powerful supercomputer systems, and researcher at the Oak Ridge National Laboratory and University of Tennessee, has described ASC as having “by far the most intense competition” of any student supercomputer contests he has witnessed.

About ASC

Sponsored by China, ASC is one of the world’s three major supercomputer contests, alongside the US’ SC and Germany’s ISC. Held annually since 2012, the competition is devoted to cultivating young talent in the field of supercomputing.

Source: ASC

The post ASC18 Competition Timeline Released appeared first on HPCwire.

Pages