VizWorld.com

This is a bit off-topic for us here at VizWorld, but interesting nonetheless.  CogniMem will be launching and demonstrating their new “CogniBlox” module at SC11, an FPGA-based stackable module targeted at large “pattern recognition” tasks.  It’s effectively a specialized neural-network processing system that offers about 4000 “neurons” per block, and the blocks can be stacked in all 3 dimensions for surprising density.  It’s an excellent candidate for the first exascale computing systems, but will be highly specialized in what it can do.  This makes it similar to the early MDGRAPE system, which hit 1 Petaflop well before other general-purpose computers, but was specialized and not capable of running LINPACK (thus making it ineligible for the Top500).

“CogniBlox architecture is based on multiple CM1K (1024 neuron) components allowing for constant parallel matching of vectors in 10 microseconds regardless of the number of vectors being compared at the same time,” said Bruce McCormick, co-founder, president and CEO of CogniMem.  “Recent announcements from IBM about its DARPA SyNAPSE project have rekindled interest in this exciting technology.  Based on multiple generations of IBM patented ZISC technology, we have perfected this approach for practical commercial use, providing unmatched performance at low power, and made it available now.”

With this making Exascale feasible, one has to wonder if such special-purpose machines may open the door to more mainstream general-purpose machines soon.  You could almost argue that NVidia’s development in the Tesla space could be a step towards bridging the cap (putting special-purpose hardware alongside general-purpose hardware).

Get the full press release after the break.

Read more…

Hardware, Science cognimem, hpc

With SC11 on the horizon, you can expect many more such announcements, but today China and NVidia are excited about the use of 2,200 Nvidia Tesla cards to run the first ever computer simulation of a complete H1N1 flu virus at atomic levels.

The CAS-IPE researchers made the simulation breakthrough by developing a molecular dynamics simulation application that takes advantage of GPU acceleration2. It was run on the Mole-8.5 GPU supercomputer, which is comprised of 288 server nodes. The system was able to simulate 770 picoseconds per day with an integration time step of 1 femtosecond for 300 million atoms or radicals1.

via Chinese Researchers Tap GPU Supercomputer for World’s First Simulation of Complete H1N1 Virus – NVIDIA Newsroom.

Hardware, Science hpc, molecular, nvidia

The National Center for Atmospheric Research (NCAR) is getting ready for a big new IBM iDataPlex system named “YellowStone” that will give them 1.6 Petaflop of HPC goodness along with 149.2 TB of RAM across 74,592 Intel cores (with Sandy Bridge).  Press releases like these come regularly, but this one actually makes note of how they plan to manage the massive amounts of simulation data coming from the system.  They’ve setup a dedicated 17PetaByte filesystem shared between YellowStone and a pair of dedicated Analysis & Visualization systems.

The DAV resource is made up of two systems, one designed to facilitate large-scale data analysis, and the other for parallel processing and visualization activities. Taken together, these components will dramatically improve capabilities central to NCAR’s mission, such as climate modeling, forecasting, and preservation of critical research data. The NWSC will serve researchers across the United States and around the world who will interact with its systems remotely.

Nice to see the visualization aspect in the same press release as the computational announcement.  Apparently we have their director to thank for some of this:

CISL director Al Kellie emphasizes the importance of the integrated computing resource, explaining that what makes this system exceptional for geoscience research is the linking of a very large centralized file and data storage system to a high-performance computational cluster and visualization resource. “While we wanted to make sure we had adequate computing capacity, we knew that it would be of limited use if we didn’t ensure easy access to the data and appropriate resources for storing and analyzing it,” Kellie says. “In addition to high-performance machines, researchers need quick access to their data and a way to analyze it, to see what it means. This system addresses those needs elegantly.”

via HPCwire: NCAR to Install 1.6 Petaflop IBM Supercomputer.

Hardware, Science hpc, ncar

ORNL is making a play to take back the #1 spot on the Top 500 with another upgrade to their Jaguar frankenputer, I mean supercomputer.  Already part Cray XT4 and part XT5, the new “wing” will be XK6 systems running with NVidia GPU’s.  The initial parts will be Tesla M2090′s, but then later parts will be based on the new Kepler design.  The final system is designed to hold 18,000 GPU’s.

“Oak Ridge’s decision to base Titan on Tesla GPUs underscores the growing belief that GPU-based heterogeneous computing is the best approach to reach exascale computing levels within the next decade,” said Steve Scott, chief technology officer of Tesla products at NVIDIA, referring to computing performance levels of 1,000 petaflops. ”The Tesla GPUs will provide over 85 percent of the peak performance of Titan. You simply can’t get this level of performance in a power- and cost-efficient way with CPUs alone.”

I’ve personally come to believe that adding GPU’s to HPC’s is a short-term solution to a bigger problem.  Until we can come up with some operating systems and better tools for programming these behemoths, the major hindrance will be the development cycle.

via Oak Ridge National Lab Turns to NVIDIA Tesla GPUs to Deploy World’s Leading Supercomputer – NVIDIA Newsroom.

Hardware cray, hpc, nvidia, ornl

TACC is bringing a new Supercomputing online called “Stampede” which aims to push the limits of Linux Cluster designs and further the NSF’s “Extreme Digital” program.  Providing a peak 2 Petaflops of performance, with another 8 petaflops of possible performance thanks to Intel’s Knights Ferry chips, you may be wondering what they plan to use for visualization and analysis.

In addition, Stampede will offer 128 next-generation Nvidia graphics processing units (GPUs) for remote visualization, 16 Dell servers with 1 terabyte of shared memory and 2 GPUs each for large data analysis, and a high-performance Lustre file system for data-intensive computing.

Many people in the field (myself included) believe that the days of dedicated graphics processing systems are numbered, but Stampede seems to indicate they have some life left in them.  However, I have to wonder what percentage of the $50M cost of the system is locked up in these 16 1-terabyte nodes.

via Intel, Dell Powering Stampede Supercomputer in Texas – IT Infrastructure – News & Reviews – eWeek.com.

Hardware, Science hpc, nsf, tacc

An article in the Times of India talks about the growth of SGI, the doubly-failed supercomputing company, in the wake of certain flubs by HP and IBM.  Before I begin the dissection, I think these paragraphs are worth reading closely:

“Ninety days back, Leo announced HP’s new future to be in mobility, cloud and connectivity . Ninety days later, he withdraws mobility and connectivity, recalling its TouchPad and a planned spinoff of its PC business.” HP’s uncertainty is turning out to be good for Barrenechea’s USbased supercomputing company which recently posted better than expected results even in an uncertain economic environment.

…

“IBM just walked away from a customer at the University of Illinois, a $200-million dollar project. I don’t know how you will build a trusted relationship with a customer when you are willing to stand up to an important system and say just kidding,” added Barrenechea , who was the CTO of CA Inc before joining the SGI board in 2006.

Got to give them credit, I think HP’s uncertainty will be a boon for many companies, not just SGI.  However, I don’t see how HP pulling out of tablet computing is really going to do much for SGI’s HPC business.

As for the 2nd one, I had to stop myself from laughing out loud.  Talking smack about IBM and Blue Waters after they busted the $30M PSC deal (More here and here) really shows how short their own memory is.  Of course now NCSA is looking for another system to fill their floor, but they’ve first got to secure funding and such so it’s probably a good year away.  The Blue Waters system was an NSF system (targeted for DARPA research) but I bet DARPA won’t be paying for another system hosted at NCSA anytime soon, so that leaves NSF or other agencies to pickup the tab.  With NSF focusing on heavy petascale+ systems, they tend to be offering bigger awards less frequently.

Of course, none of anything mentioned here has anything to do with Graphics.  SGI really needs to find a new meaning for that G.

Update 12:43pm – Clarified that Blue Waters was funded by NSF, but targeted for DARPA research.

via Times of India and HPCwire

Hardware hpc, sgi

A few weeks ago Cray CTO Steve Scott made waves in the HPC News media with the announcement of his resignation and move to a “computing partner” that wouldn’t be named.  Most people figured they meant AMD.  Well, it’s been named now and it’s a shocker: He’s the new CTO of NVidia’s Tesla business unit.

“There are few people on the planet that have Steve’s deep system level understanding of high performance computing,” said Bill Dally, Nvidia’s chief scientist. ”Steve’s decision to join Nvidia is a resounding endorsement that GPU accelerated computing is the future of HPC. He will play a central role in architecting the world’s most powerful supercomputers.”

via Nvidia Signs On Cray Exec As New Tesla CTO.

Hardware hpc, nvidia, tesla

The annual SciDAC Visualization Night was last month and another 10 great HPC Scientific Visualizations brought home awards.  Wired magazine has the list of winners, complete with their videos.

“The human eye can pick out patterns in simulations that are are otherwise hard to describe, and they can do it better than any computer,” said visualization scientist Joseph Insley of Argonne National Laboratory. “Plus, with the incredible amount of data gathered these days, it’s difficult to analyze it any other way.”

Making a useful scientific simulation isn’t light work. If field researchers want to do it themselves, they must learn to code instructions for computer processing and control advanced 3-D animation software. Because of these hurdles, and the increasing sophistication of modeling methods, most team up with computer and visualization scientists to get the job done.

Disclosure: Me & My Team are included among the winners, the Overhead Threat Protection System video included below.

via 10 Award-Winning Scientific Simulation Videos | Wired Science | Wired.com.

Graphics, Science hpc, scidac, sciviz

The latest issue of Scientific Computing has a great article from some NASA researchers on analyzing and visualizing airflow around landing gear, in hopes of redesigning them to reduce vibration and “aeroacoustic” effects (eg, Loud rumbling).  If you see their “Digital magazine” Version you can see some movies of their visualizations.

To generate the flow animations presented here required saving a small portion (12,000 snapshots or time steps) of the flow simulation record. With each snapshot resulting in a file size on the order of 4 to 5 gigabytes, the total time record saved is in excess of 50 to 70 terabytes of data. Although such an aggregated file size is not excessively large by today’s standards, it is still too large for routine visualization of the results. The push toward much larger simulations (a nose gear computation on a grid twice as large as the current grid is ongoing) precludes relying on traditional methods for post-processing of CFD data; that is, saving the volumetric information at each time step for analysis at a later time, as these are highly inefficient and no longer practical. Such large datasets demand concurrent real-time simulation, analysis and visualization of the flow field without the need to save countless terabytes of information that would soon tax the storage capacity of even the largest supercomputers.

Scientific visualization of high-fidelity, large-scale flow simulations such as these has become an indispensable tool for providing global insights and knowledge that enable the development of viable engineering solutions to pressing environmental issues affecting the public good. The landing gear simulations, for example, together with those from other disciplines relevant to aircraft design, will soon be used to help develop a new breed of subsonic aircraft that will not only reduce noise pollution, but will burn less fuel and produce fewer harmful emissions — all to improve life on our planet.

via High-Fidelity Simulation of Landing Gear Noise.

Science hpc, nasa, research, scivis

I’ve got to admit, I’m impressed by SGI.  I really thought the UV system was going to be the albatross that drug SGI to Davy Jones’ locker, but they’re actually doing ok.  A press-release out today announces 500 Altix UV systems delivered.  Now, most companies wouldn’t even bother announcing 500 systems (Dell probably ships 500 systems in a day or less), but these are significantly bigger and more expensive that your average computer.

As the world’s most scalable x86 platform, customers can deploy standard Windows Server® 2008 R2 and Windows-based applications to the maximum scale possible of 256 physical cores and 2 terabytes of memory. Altix UV 1000 with 256 cores of Intel® Xeon® processor 7500 series and 2 TB of memory has four times the x86 scalability of HP and 2.6 times the scalability of IBM.

Their partnership with Microsoft for Windows HPC Server is probably a big reason they’ve done as well as they have, as Windows HPC Server really seems to be taking off in Financial HPC.   I’m not aware of anyone doing petascale work with UV’s, but maybe they’re on to something in the high-but-not-extreme end of HPC.

Of course, their financials still seem a little fishy to me.  We’ll see how they look at the end of the year.

via SGI Marks Supercomputing Milestone with over 500 Altix UV Systems Delivered | Business Wire.

Hardware hpc, sgi, ultraviolet, uv