supercomputer
Latest
IBM's Watson is really smart, will try to prove it on Jeopardy! this fall (video)
As much as we love our Google homepage, computer search remains a pretty rudimentary affair. You punch in keywords and you get only indirect answers in the form of relevant web results. IBM doesn't seem to be too happy with this situation and has been working for the past three years on perfecting its Watson supercomputer: an array of server racks that's been endowed with linguistic algorithms allowing it to not only recognize oddly phrased or implicative questions, but to answer them in kind, with direct and accurate responses. Stuffed with encyclopedic knowledge of the world around it, it answers on the basis of information stored within its data banks, though obviously you won't be able to tap into it any time soon for help with your homework. The latest word is that Watson's lab tests have impressed the producers of Jeopardy! enough to have the bot participate in a televised episode of the show. That could happen as early as this fall, which fits right in line with our scheduled doom at robots' hands by the end of 2012. Ah well, might as well get our popcorn and enjoy the show.
AMD launches 12-core Opteron server chips, Intel counters with the 8-core Xeon 7500
You thought six cores were nifty? This week, AMD and Intel have begun the multithreaded battle in earnest -- if only on the IT front -- with chips that have up to double that core density. First up, AMD has officially brought us that Opteron 6000 series leaked last week, a set of 8- and 12-core processors aimed at dual- and quad-CPU servers that it claims have both higher performance and lower cost than Intel's recent hex-core offerings. Not to be outdone, Intel has just introduced a 8-core processor series of its own, the Xeon 7500, that it envisions deployed in mammoth 256-processor configurations. In bulk orders of 1,000, a single 12-core Opteron costs nearly $1,200, while the cheapest single 8-core Xeon will set you back a cool $2,461 in the same quantity. We don't doubt they're powerful, and we'd kill for a pair of either in our gaming rig. At those prices though, we'll stick to building our supercomputer out of PS3s -- oh, wait.
Intel's Core i7-980X Extreme Edition hits a slew of new gaming desktops
Intel's six-core, twelve-threaded Core i7-980X Extreme Edition has turned the hardcore gaming community on its head, and just about everyone is scrounging around in a (mostly futile) attempt to locate $999. For those in dire need of an entire system replacement, it seems that today's the day to start looking. Shortly after we heard that this 32nm Gulftown chip would be landing with Alienware and Origin PC rigs, a veritable plethora of other outfits have shown up to make similar announcements. Digital Storm has popped an overclocked (4.4GHz) version into its Black|OPS machine (which conveniently starts at $5,642, while CyberPower is now offering the silicon in its Black Mamba, Black Pearl and Gamer Xtreme 3D machines. Maingear's also sliding said CPU into its world-beating Shift "supercomputer," and anyone shopping a high-end Velocity Micro system will also see the option. We suspect most every other PC maker in existence will be following suit soon, so if your prefab PC builder hasn't yet jumped on the bandwagon, just hold tight. Real tight.
EVE Online devblog addresses recent server issues
EVE Online's server is a complex beast, holding the title of biggest supercomputer in the gaming industry. The main server cluster is housed in London and serviced by a team of IBM engineers. In addition to constant hardware upgrades to take advantage of the newest technology, CCP's network programmers work around the clock to improve performance and track down bugs that will affect the game. EVE is no stranger to lag or network issues and older players know all too well that server troubles are expected around patch days. When the Dominion expansion was released, there were far more complaints of server issues than could be attributed to the usual "patch day blues". Now several months down the line we're still hearing horror stories of fleet battles lagging unbearably with only a few hundred players. The last few months have seen an increasing number of node deaths and database failovers, in some cases causing unscheduled server reboots. Read on to find out what CCP is doing to combat the issue.
University of Antwerp stuffs 13 GPUs into FASTRA II supercomputer
The researchers at the University of Antwerp's Vision Lab caused quite a stir last year when they built a supercomputer with four high-end NVIDIA graphics cards, but it looks like they've truly stepped up their game for their followup: a supercomputer that packs no less than thirteen GPUs. That, as you might have guessed, presented a few new challenges, but after wrangling some flexible PCI cables into a specially-made case and loading up a custom BIOS courtesy of ASUS, they were apparently able to get six dual-GPU NVIDIA GTX295 cards and one single-GPU GTX275 card up and running with only a few hiccups. As before, the big advantage with this approach is that you get an enormous amount of computing power for a relatively small cost -- twelve teraflops for less than €6,000, to be specific. Head on past the break for a pair of videos showing the thing off, and hit up the link below for the complete details (including some jaw dropping benchmarks).
US Air Force to purchase 2200 more PS3s
You may laugh at the "It Only Does Everything" ads but those of us with a cot in the Joystiq Bunker understand that they're a sign that the end is nigh. Why? Because the US military is looking to make a super computer made up of 2500 interconnected PS3 systems -- wait, isn't that how Skynet is made? The US Air Force already has over 300 launch PS3s in its arsenal, but an upcoming purchase of 2200 additional systems will greatly expand the number-crunching capabilities of its networked supercomputer. According to Information Week, the US Air Force will use Sony's Cell-powered hardware for radar imagery, HD video processing and "neuromorphic computing" or, to those of us without a PhD in Armageddon-ology, "building computers with brain-like properties." Yeah, we told you the end was coming.
IBM PowerXCell-8i processor said to be last of its kind, but Cell will live on
IBM's Cell processor may have helped break a few records and find its way into everything from video game consoles to TVs at the same time, but it looks like things could be about to change in a fairly big way for Big Blue's groundbreaking chip. According to Heise Online, IBM Vice President of Deep Computing David Turek has confirmed that the company's current PowerXCell-8i processor will be the last of its kind, and that there will not be a successor with dual PowerPC processors and 32 SPEs as originally planned. Slightly less clear, however, is the future of the Cell program in general, which will apparently live on in "another form" -- to which Turek reportedly added, somewhat vaguely, "the future is hybrid," although we're fairly certain he's not talking about cars.
NVIDIA Tesla 20-series GPUs promise to dramatically cut supercomputing costs
Sure, you've been hearing NVIDIA toss around names like CUDA, Fermi and Tesla for what seems like ages now, but we're guessing this is the sort of thing that'll get most folks to really take notice: a promise to cut supercomputing costs by a factor of ten. That rather impressive feat comes courtesy of the company's new Tesla 20-series GPUs, which come in the form of both single GPU PCI-Express Gen-2 cards and full-fledged GPU computing systems, and promise a whole host of cost-saving benefits for everything from ray tracing to 3D cloud computing to data analytics. Of course we are still talking about "cheap" in supercomputing terms -- look for these to run between $2,499 and $18,995 when they roll out sometime in the second quarter of 2010.
Cray Jaguar leaps past IBM Roadrunner as world's fastest supercomputer and pun generator (video)
Cray has finally clawed IBM back from the lead position on the Top500 Supercomputer chip-measuring contest. After just missing out on the title to IBM's Roadrunner last year, Cray's XT5 supercomputer (aka, Jaguar) at Oak Ridge National Lab in Tennessee received an update from quad- to six-core Opteron processors to boast a 2.3 petaflop per second performance peak (theoretical) and 1.75 petaflops as measured by the Linpack benchmark; a feat requiring almost a quarter million AMD cores. IBM's Roadrunner, the very first supercomputer to race past the petaflop per second threshold, managed just 1.042 petaflops by comparison. Remember, one petaflop per second is equivalent to one quadrillion calculations per second. Of course, chip makers put their own spins on the list by noting that 4 of the top 5 systems depend on AMD for performance while Intel can be found powering 402 of the Top500. Video of the AMD processor upgrade procedure can be found after the break.
Maingear unveils Core i7-packin' SHIFT, your own 'personal supercomputer'
We'll be straight with ya -- we're betting these "personal supercomputer" claims are just a bit out of line with reality, but even still, there's no denying that Maingear has shoved an insane amount of horsepower beneath the (admittedly large) hood of its newest rig. The beastly SHIFT does away with copious LED lighting and blinging accents found on many modern gaming PCs and instead opts for a classier, more ominous tower. Within, you'll find a vertical airflow system, a Core i7 processor, your choice of ATI or NVIDIA graphics, 8GB (and up) of DDR3-1600 RAM, up to 6 HDDs or 12 SSDs, DVD and Blu-ray options, an Asetek liquid-cooling solution, Razer peripherals, an optional Killer NIC Xeno Pro and Windows 7 running the show. The Intel P55 rig gets going at $2,199, while the X58 model starts $400 higher; for those in creative design fields, Maingear's expected to unveil a SHIFT just for you in the near future. Head on past the break for the full release. %Gallery-77033%
Asustek announces a 1.1 Teraflop, Tesla GPU powered supercomputer
Some of us love nothing more than a portable and convenient netbook -- something that Asustek knows all too well -- but how about those of us who need real computing power? To that end, Taipei's choice for all things ultraportable has just announced its very own 1.1 Teraflop supercomputer. Dubbed the ESC 1000, this (albeit large) desktop-sized machine sports a 3.33GHz Intel LGA1366 Xeon W3580 microprocessor and three CUDA-based Tesla C1060 GPUs, the likes of which we last saw in Dell's Precision "personal supercomputer" line. Shipping with 24GB of DDR3 DRAM (1333MHz) and a 500GB SATA II hard drive, the machine is said to have a cost structure of $14,519 over five years. We're guessing that you'll be able to both surf the net and watch HD quality video on the thing, although you probably won't be taking it along with you to Crazy Mocha any time soon. According to a company spokesperson, this thing is ready to ship now, although a launch date and street price have yet to be determined. One more pic after the break.
SGI announces Octane III personal supercomputer
We've seen companies throw in with personal supercomputing platforms in the past, but if anybody makes a splash with the things we hope it's SGI. The new SGI -- that is, the mash-up of the old SGI and Rackable -- has just announced the Octane III, an office-friendly (one-by-two-foot) tower that can sit under a (fairly large) desk while permitting up to 80 high-performance cores with its nearly 1TB of memory. Options include Intel Xeon 5500 series, Dual-core Atom, or Xeon 3400 processors, and an NVIDIA GP-GPU card. If you're a system administrator with money to throw around and a burning desire to add a machine running Microsoft HPC Server 2008, SUSE Linux Enterprise Server, or Red Hat Enterprise Linux to your network (and who isn't, really?) hit up the read link to talk to a sales rep. Prices start at $7995. Show full PR text SGI Unveils Octane™ III Personal Supercomputer Octane III Delivers Unparalleled Performance, Energy Efficiency and Ease of Use to HPC Users in Office Environments Intel Developer Forum 2009, San Francisco FREMONT, Calif.--(BUSINESS WIRE)--SGI® (NASDAQ: SGI) today announced the immediate availability of Octane™ III, the company's first personal supercomputer. This new product takes high-performance computing to a new level by combining the immense power and performance capabilities of a high-performance deskside cluster with the portability and usability of a workstation. The Octane III is uniquely suited for workplace environments and supports a vast range of distributed technical computing applications. Octane III is office-ready with a pedestal, one-by-two-foot form factor, whisper-quiet operations, easy-to-use features, low maintenance requirements and support for standard office power outlets. While a typical workstation has only eight cores and moderate memory capacity, the superior design of the Octane III permits up to 80 high-performance cores and nearly 1TB of memory for unparalleled performance. "Octane III makes supercomputing personal again," said Mark J. Barrenechea, president and CEO of SGI. "Our customers have been asking for office environment products with large core counts that are easy to use and whisper-quiet. Octane III brings all of this to the HPC professional, and enables a new era of personal innovation in strategic science, research, development and visualization." Octane III is easily configurable with single- and dual-socket node choices, and offers a wide selection of performance, storage, graphics, GP-GPU and integrated networking options. Yielding the same leading power efficiencies inherent in all SGI Eco-Logical™ compute designs, Octane III supports the latest Intel® processors to capitalize on greater levels of performance, flexibility and scalability. "IDC research has shown a growing demand for high-performance deskside supercomputers, and the new Octane III personal supercomputer is designed to directly address the requirements of both the entry level HPC technical server market and the desktop workstation market with a high-performance cluster designed to help close the gap between the desktop and the data center," said Earl Joseph, HPC Program Vice President at IDC. "SGI workstations and servers have been highly regarded by HPC scientists, engineers and analysts, and the new system with up to 80 high-performance cores and 1TB of memory is certain to capture the attention of many HPC computing professionals." Octane III is available in a variety of configurations: * Ten dual-socket, Quad-Core Intel® Xeon® processor 5500 series-based nodes * One dual-socket, Quad-Core Intel® Xeon® processor 5500 series-based workstation with advanced NVIDIA graphics and/or GP-GPU card support * Nineteen single-socket, Quad-Core Intel® Xeon® processor 3400 series-based nodes * Nineteen single-socket, Dual-Core Intel® Atom™ processor-based nodes "SGI's Octane III offers a solution for workstation users wanting to reach dramatically higher levels of performance with minimal investment. With the support of up to 19 Intel® Xeon® processors, HPC users looking to enhance their workgroup productivity outside the data center can also utilize the benefits of the Octane III for faster discovery and quicker simulations," said Richard Dracott, General Manager of High Performance Computing at Intel. "With certification from the Intel® Cluster Ready program, Octane III will provide a powerful, easy-to-deploy personal supercomputer." For a simple, out-of-the-box experience, Octane III ships as a factory-tested, pre-integrated platform with broad HPC application support – powerful enough for the most complex applications in the world. These include fluid dynamics, quantum mechanics, molecular dynamics, seismic processing, data analytics, rendering, visualization and computer-aided design, among many other HPC applications. Octane III supports Microsoft HPC Server 2008, SUSE® Linux® Enterprise Server and Red Hat® Enterprise Linux operating systems. Linux configurations include SGI ProPack™ and ISLE™ cluster management software. Octane III will be on display at the Intel Developer Forum (IDF) at Moscone Center, San Francisco, Calif., on September 22-24, 2009, in Booth # 718 at the Technology Showcase in the HyperScale Community. Availability and Pricing: Octane III is immediately available with Intel® Xeon® processor 5500 series or Intel® Atom™ configurations. The base configuration price starts at $7,995. For more information about Octane III, please visit http://www.sgi.com/OctaneIII.
Water-cooled Aquasar supercomputer does math, heats dorm rooms
Not that we haven't seen this trick pulled before, but there's still something magical about the forthcoming Aquasar. Said supercomputer, which will feature two IBM BladeCenter servers in each rack, should be completed by 2010 and reach a top speed of ten teraflops. Such a number pales in comparison to the likes of IBM's Roadrunner, but it's the energy factor here that makes it a star. If all goes well, this machine will suck down just 10KW of energy, while the average power consumption of a supercomputer in the top 500 list is 257KW. The secret lies in the new approach to chip-level water cooling, which will utilize a "fine network of capillaries" to bring the water dangerously close to the processors without actually frying any silicon. While it's crunching numbers, waste heat will also be channeled throughout the heating system at the Swiss Federal Institute of Technology, giving students and dorm room crashers a good feel for the usefulness of recycled warmth.
Fujitsu's supercomputer-ready Venus CPU said to be "world's fastest"
Due to the intrinsic limitations of machine translation, it's hard to say exactly what makes Fujitsu's latest supercomputer the "world's fastest," but we'll hesitantly believe for the time being. We're told that the SPARC64 VIIIfx (codename Venus) can churn through 128 billion calculations per second, which supposedly bests the current champ -- a chip from Intel -- by 2.5 times. An AP report on the matter states that Fujitsu shrunk the size of each central circuit, which in turn doubled the number of circuits per chip. 'Course, this beast won't be ready for supercomputer work for several years yet, giving the chip maker's biggest rivals plenty of time to sabotage its moment in the limelight.[Via Physorg]
NVIDIA Tesla GPUs now shipping with Dell 'personal supercomputers'
Been itching to get your hands on a personal supercomputer, as NVIDIA's ad wizards put it? The company has just announced that its CUDA-based Tesla C1060 GPU is now available in Dell's Precision R5400, T5500 and T7500 workstations. And just to put things into perspective, NVIDIA points out that a Dell workstation rockin' a single Tesla C1060 has enough going on under the hood to power the control system for the European Extremely Large Telescope project ("the world's largest," apparently). According to one of the developers, Jeff Meisel at National Instruments, a workstation "equipped with a single Tesla C1060 can achieve near real-time control of the mirror simulation and controller, which before wouldn't be possible in a single machine without the computational density offered by GPUs." Wild, huh? If you're curious about the workout that Tesla GPUs are getting on a wide range of projects, from Bio-Informatics to Computational Chemistry to Molecular Dynamics and more -- or if you're merely a glutton for long-winded PR -- check out the good stuff after the break.
IBM's Watson to rival humans in round of Jeopardy!
IBM's already proven that a computer from its labs can take on the world's best at chess, but what'll happen when the boundaries of a square-filled board are removed? Researchers at the outfit are obviously excited to find out, today revealing that its Watson system will be pitted against brilliant Earthlings on Jeopardy! in an attempt to further artificial intelligence when it comes to semantics and searching for indexed information. Essentially, the machine will have to be remarkably labile in order to understand "analogies, puns, double entendres and relationships like size and location," something that robotic linguists have long struggled with. There's no mention of a solid date when it comes to the competition itself, but you can bet we'll be setting our DVRs whenever it's announced. Check out a video of the progress after the break.[Via The New York Times]
24 Samsung SSDs get strung together for supercomputer fun
It wasn't all that long ago when a mere nine SSD drives in a RAID array was enough to cause most folks' jaws to drop, but the world of ridiculous technology exercises moves quickly, and we can only be thankful that a select few continually feel the need to one-up each other and share their results with all of us. This latest effort comes from a group enlisted by Samsung (in a not too thinly disguised marketing exercise), who paired up 24 SSDs in a RAID array totaling 6TB in size. Even more impressive than that, however, is the 2GB per second throughput speed they managed to achieve, which they naturally spared no expense in demonstrating -- as you can see in the video after the break.[Via Reddit]
IBM ThamesBlue supercomputer uncovers antediluvian English words
Granted, we could personally think of much more amazing ways to put supercomputers to work, but maybe there is some sort of benefit to humanity by knowing precisely what our ancestors' first words were. All that aside, the IBM ThamesBlue supercomputer has been tapped by language masters at the University of Reading in order to find that 'I,' 'we,' 'who' and the numbers '1,' '2' and '3' are amongst the most ancient across all Indo-European languages. Comically enough, it was also found that words like 'squeeze,' 'guts,' 'stick,' 'throw' and 'dirty' were also markedly archaic, which sure says a lot about how men in particular, um, don't evolve. At any rate, these new computational powers have reportedly opened up another 25,000 years or so of language study, so we suspect the folks on this project will be occupied for some time to come.
IBM's Sequoia: 20x faster than the world's fastest supercomputer
Roadrunner? Pfff, your chart-topping 1.105 petaflops are laughable. IBM just announced its 20-freaking-petaflop Sequoia supercomputer due for delivery by 2012. While supercomputer speeds have steadily increased year-over-year, a near 20x jump in calculations per second since the last world ranking is unheard of, even if the system has yet to come on-line. Slated to spend its life simulating nuclear explosions, Sequoia will use 45-nm (PowerPC, presumably) processors with 16 cores per chip for as many as 4,096 processors per rack. That's a total of 1.6 million cores assisted by 1.6 petabytes of memory. Perhaps all this processing power might help IBM understand the futility of its Lotus Notes strategy.
AMD talks up mildly ambiguous graphics supercomputer
It's not entirely clear what the purpose of AMD's newest supercomputer is, but it's pretty safe to say that the company is making it out to be something better than it is. All harshness aside, the so-called Fusion Render Cloud will reportedly be the planet's fastest graphics supercomputer. It sounds like the machine will be good for farming out tasks, and more specifically, for "real-time rendering of film and visual effects graphics on an unprecedented scale." Evidently AMD's hoping that gaming firms tap into the beast in order to develop and deploy next-generation game content, so that probably means it's worthless for you. But hey, it sounds awesome, right?
