AboutAI

UK researchers are working on fitting CCTV cameras with artificial intelligence, allowing them to more quickly respond to crimes. The technology, being developed by University of Portsmouth scientists, would allow cameras to “hear” violent sounds and react, swiveling quickly in the direction of a broken window or somebody shouting abusively for example, before alerting an operator.

The artificial intelligence powering the camera would also be able to respond to visual cues such as fights, or violent behaviour.

Scientists say the aim is to allow the camera to react just as a human might, hearing a scream and then swinging around to find the source with the same speed as a person, which is about 300 milliseconds.

Over time, the scientists claim the AI algorithms would learn, picking up key words and phrases it associates with criminal activity.

“The longer artificial intelligence is in the software the more it learns. Later versions will get cleverer as time goes on,” says Dr David Brown, director of the project.

Fuzzy thinking

Brown says the foundation of the technology is a new type of fuzzy logic: “In identifying sound we are looking for the shapes of sound. In the same way, if you close your eyes, you can trace the shape of a physical object and ‘read’ its profile with your hand we are developing shapes of sound so the software recognises them.

“The software will use an artificial intelligence template for the waveform of sound shapes and if the shape isn’t an exact fit, use fuzzy logic to determine what the sound is. For example, different types of glass will all have slightly different waveforms of sound when they smash but they will have the same generic shape which can be read using fuzzy logic.

“It’s a very fast, real-time method of identifying sounds.”

While there are clearly privacy implications inherent in the technology, Brown claims the AI will be trained “to only listen for specific words associated with violence, not full conversations.”

However, is this hardly likely to calm privacy advocates already concerned with the growing number of CCTV camera in the UK, and their potential uses.

At the moment the AI is in research stage, though scientists have been given a three-year grant by the Engineering and Physical Sciences Research Council (EPSRC) to further develop it.

Popularity: 1% [?]

The first laptops to make use of the SpursEngine, a multimedia co-processor derived from the Cell chip that powers the PlayStation 3, will go on sale in Japan in July. Toshiba will launch its Qosmio G50 and F40 machines with the chip, which contains four of the “Synergistic Processing Elements” from the Cell Broadband Engine processor. The Cell chip used in the PlayStation 3 has eight of the SPE cores plus a Power PC main processor. The SPE cores perform the heavy number-crunching that makes the console’s graphics so stunning.

The SpursEngine SE1000 will work in much the same way in the laptops.

The operating system will run on an Intel Core 2 Duo chip and the SpursEngine will be called on to handle processor-intensive tasks, such as processing of high-definition video. This arrangement means the laptop should be capable of some tricks that haven’t been seen on machines until now.

Among them, Toshiba said the two computers will be able to upscale standard-definition video to high definition; transcode in realtime digital TV to MPEG4 so that the resulting files are cut down in size by one-eighth and burn video to DVD in half the time of current machines.

A novel feature is face navigation. Faces that appear in video are recognized and displayed as thumbnail images to create a visual index to the video. Users can find the person or scene they want by glancing at the thumbnails and then click on the respective one to watch that portion of video. The computer can also divide up the scenes in user-shot video so they can be viewed one-by-one and analyze and display the volume or the clip across its entire length so, for example, excitement in a sports event can be more easily found.

Finally, by analyzing images from the computer’s built-in camera it’s possible to control video playback with hand gestures.

The Qosmio G50 is a multimedia laptop and has an 18.4-inch high-definition screen, 500G bytes of hard-disk space, NVidia GeForce 9600M graphics processor, dual digital TV tuners and wireless LAN including 802.11n. It weighs 4.9 kilograms and measures 45 centimeters by 31cms by 4.8cms. Battery life is about 4 hours.

The Qosmio G50 will be cost from ¥290,000 (US$2,700) and the F50, which has a 15-inch screen and 250G byte hard-disk drive, from ¥250,000. Toshiba plans to put the machines on sale overseas but has yet to announce launch details.

http://www.pcworld.com/businesscenter/article/147415/toshiba_unveils_laptop_with_cellderived_chip.html

Popularity: 1% [?]

Stanford University computer science professor John McCarthy coined the phrase in 1956 to mean “the science and engineering of making intelligent machines,” In the early years of the artificial intelligence movement, enthusiasm ran high and artificial intelligence pioneers made some bold predictions.

In 1965, artificial intelligence innovator Herbert Simon said that “machines will be capable, within 20 years, of doing any work a man can do.”

Two years later, MIT researcher Marvin Minsky predicted, “Within a generation … the problem of creating ‘artificial intelligence’ will substantially be solved.”

Popular culture jumped onto the artificial intelligence bandwagon and gave us Rosie the Robot from the Jetsons, HAL from the movie 2001 and R2D2 from Star Wars.

Yet, here we are, decades later and what has artificial intelligence done for us lately? If you define artificial intelligence as self-aware, self-learning, mobile systems, then artificial intelligence has been a huge disappointment.

On the other hand, every time you search the Web, get a movie recommendation from NetFlix, or speak to a telephone voice recognition system, tools developed chasing the great promise of intelligent machines do the work. In other words, we may not have full-functioning robots that cater to our every need, but artificial intelligence is embedded in our everyday lives.

“Once tools get far enough out of the lab, they’re no longer AI, just common computer science,” says Professor George Luger of the University of New Mexico. “AI just went to work.”

One of the biggest boosts to artificial intelligence is Moore’s Law, because artificial intelligence needs CPU power. “It took 20 years to go from a 5MHz chip to a 500MHz chip, but only eight months after that to get to a 1GHz chip,” says futurist Daniel Burrus, author of the best seller Technotrends: How to Use Technology to Go Beyond Your Competition and founder of Burrus Research.

“The new Sony Playstation came out a year ago,” says Burrus, “but if it came out five years earlier it would be considered a supercomputer.” Burrus likens the growth of processing power on a graph to a hockey stick. “In the 90s, the graph was still low. In 2000, the graph started up a little. In 2008, we’re getting on the handle of the hockey stick.”

Burrus listed off multiple uses of artificial intelligence and expert systems that work behind the scenes. “The first application of successful AI was in the financial services industry for loan qualifications. Loan qualification went from one to two weeks down to minutes.” Other examples include systems that help Navy pilots land jets on aircraft carriers.

His personal favorite is the use of an expert system to manage room service orders at Marriott hotels. “AI tells them when to start cooking and when to deliver. Marriott tells me exactly when breakfast will be delivered while others give me a 15 minute window. That’s a competitive advantage for Marriott.”

While energy prices soar, Burrus noted the cost of intelligence keeps going down. “Maybe we can offset the energy trend as we make appliances more intelligent.”

Access to tools
Part of offsetting that trend will be better software tools, the type favored by Luger in his book, Artificial Intelligence: Structures and Strategies for Complex Problem Solving (Sixth Edition). “Modern languages have roots in AI research, including object oriented design, C++, C#, and Java,” Luger says. “The coolest stuff we’ve done is build a set of exciting tools.”

Yet tools and embedded intelligent systems don’t answer the “grand challenges” of artificial intelligence, including robots and language processing. Very few projects have captured the public’s imagination.

NASA got great public response with their Mars rovers, but little was made of the artificial intelligence components. Artificial intelligence techniques considered pure research 15 years earlier guided rovers Spirit and Opportunity around rocks a world away.

Defense Advanced Research Projects Agency (DARPA) provides money for “grand challenges” including Internet development in their earlier incarnation of ARPA. Now it sponsors a contest to build autonomous vehicles (see Urban Challenge). This forces teams to integrate separate discipline areas such as machine vision, learning systems and problem solving while moving through unfamiliar areas.

Yet tools and embedded intelligent systems don’t answer the “grand challenges” of artificial intelligence, including robots and language processing. Very few projects have captured the public’s imagination.

NASA got great public response with their Mars rovers, but little was made of the artificial intelligence components. Artificial intelligence techniques considered pure research 15 years earlier guided rovers Spirit and Opportunity around rocks a world away.

Defense Advanced Research Projects Agency (DARPA) provides money for “grand challenges” including Internet development in their earlier incarnation of ARPA. Now it sponsors a contest to build autonomous vehicles (see Urban Challenge). This forces teams to integrate separate discipline areas such as machine vision, learning systems and problem solving while moving through unfamiliar areas.

One of the most successful artificial intelligence products is literally underfoot. Roomba, the home vacuuming product from iRobot, has sold over 2 million units. One survey showed over half of the deployed Roombas have been given pet names by their owners.

Colin Angle, CEO and co-founder of iRobot, says, “When we started shipping Roomba in 2002, we asked focus groups if it was a robot. They said no, a robot was humanoid and this was an intelligent floor vacuum. Now people are definitely changing to accept robot appliances.”

Hollywood again set the bar high. “Since the Jetsons in 1962, they created expectations we failed to meet for over 40 years. Big AI projects have largely gone by the wayside, but you can see effective behavior that solves real world problems,” Angle says.

As you might expect from someone making work tools for the real world, Angle takes a practical look at artificial intelligence and robotics. “In general, software is algorithms and code that can be reused across platforms. The more low-level tasks used to handle different situations, such as obstruction avoidance, the more successful. We call it bottom-heavy cognition,” he says

See me, feel me, touch me
Seeing and avoiding obstacles remains tough. “Years ago, researchers had the idea that machine vision was a straightforward problem, and was given to a graduate student for a summer project. Turns out things are radically harder than what people in the field though,” Angle says.

Many remember Phillipe Kahn from his high profile days running Borland, but now he’s CEO of Fullpower Technologies. The company provides an operating environment for sensors in camera phones and consumer electronic devices.

“What we do is all about sensors. Imaging sensors, proximity sensors, and touch sensors are all part of what needs to be put to work. Sensors produce piles of organized data. Great software turns that raw data into actionable information. Fullpower is working on such solutions,” Kahn says.

Micro-controllers often only have 8KB of RAM, so Fullpower writes in C and Assembler. “In the real world of next-generation intelligent devices, small, lean and frugal rule,” Kahn says. “I predict that most of the successful and useful advances will come from sensor-enabled devices and networks of such sensor-enabled devices.”

The language barrier
If machine vision remains a barrier for robot movement and navigation through the environment, the language barrier still looms large but is shrinking. Workable systems are appearing, particularly when a voice-recognition system can be trained or remains limited to certain vocabulary word groupings

Larry Harris founded Artificial Intelligence Corporation in 1975, then founded Linguistic Technology Corporation in 1994, which became EasyAsk Software. Now vice president and general manager for the EasyAsk division of Progress Software, Harris continues to help machines solve language problems.

“We translate over 60,000 natural language questions per month into queries,” Harris says. When people type more than two or three words into an e-commerce search field, the system has to understand enough to search the product database accurately.

“The base work for Ask Jeeves was at the AI Lab at MIT,” Harris says. “They were at the top until Google came out.” Google uses artificial intelligence techniques for word stemming (getting the word down to the root), language analysis, and applying the results to the index.

As an example of artificial intelligence tools becoming commonplace programming modules, Harris listed word stemmers. “You can now buy them off the shelf and plug them in. And you choose stemming rules for the language you need, since the rules for German are different than French and English.”

Harris warns there are no silver bullets in artificial intelligence, just incremental advances. “People don’t want to claim their product is AI,” Harris says. “They just focus on the voice recognition angle. There’s no real advantage to calling it AI, and even some baggage. Once you have a high proficiency example, you don’t mention AI.”

University of New Mexico’s Luger says “language processing is a big area. We’re working with a small company to answer questions in the context of a knowledge base that knows the area of inquiry.” Asking machine language processing to understand all words and speech idioms still leads to failure, but building in a knowledge base of a topic area works.

“Go to the Next I.T. Web site and check Ask Jenn from Alaska Airlines and Ask SGT STAR from the U.S. Army, two natural-language bots we put together,” Luger says. “We want to give the same answers to the same questions, which you don’t always get with people.”

Research yields results
Eric Horvitz, manager of the Adaptive Systems group at Microsoft, says “about a quarter of all Microsoft research is focused on AI efforts.” Microsoft Research includes close to 1,000 Ph.D level researchers spread across eight campuses around the world, and a completely open research and publication environment. “It’s a think tank, but not a captive one. We have an open publication model.”

“Microsoft Research’s No. 1 goal is to push the state of the art forward without regard to Microsoft,” Horvitz says. “Researchers do their best work, publish in journals, and then work with product teams to build the best software or service.” One project that started in Microsoft Research became the new SYNC voice recognition technology used by audio systems in Ford vehicles.

Horvitz and fellow researchers also have the ability to turn thousands of Microsoft employees into guinea pigs. The kernel of the Vista operating system includes machine learning to predict, by user, the next application that will be opened, based on past use and the time of the day and week. “We looked at over 200 million application launches within the company,” Horvitz says. “Vista fetches the two or three most likely applications into memory, and the probability accuracy is around 85 to 90%.”

Desktop application traffic is one thing, but city traffic prediction is another. ClearFlow, a project born of the frustration of sitting in Seattle traffic, examined thousands of routes for people based on the inference of local street traffic flow reacting to highway accidents. Realizing side streets become clogged when drivers seek to escape highway congestion, Microsoft’s maps.live.com site includes side street congestion history in rerouting suggestions. Microsoft rolled out this free service for 72 cities in early April.

The excessive hype over artificial intelligence promises in the 1950s, 1960, 1970s, 1980s, and 1990s have made the public weary of unfulfilled promises. While almost every consumer electronic device includes some artificial intelligence tools inside, the box labels never include artificial intelligence in the parts list.

Artificial intelligence is not only still around, but in more places than ever. Rather than calling the tools artificial intelligence, manufacturers just call technologies developed by artificial intelligence research “tools.” Just remember that the next time you perform a Web search, write an address on an envelope the Post Office sorts automatically, or ask Microsoft Word for a grammar check, artificial intelligence does the heavy lifting.

Gaskin writes books (16 so far), articles, and jokes about technology and real life from his home office in the Dallas area. Gaskin has been helping small and midsize businesses use technology intelligently since 1986. He can be reached at readers@gaskin.com

http://www.networkworld.com/research/2008/062308-artificial-intelligence.html?nwwpkg=slideshows

Popularity: 2% [?]

English researchers have developed a robot that can not only create 3-D replicas of objects like shoes and door handles – it also can replicate itself. Scientists from the University of Bath in England unveiled an open-source machine that acts like a three-dimensional printer. Instead of printing out documents or pictures on paper, this printer uses blueprints to produce 3-D plastic objects.

The machine has been dubbed RepRap, which is short for replicating rapid-prototyper.

The goal is to eventually build a robot that can produce individual processors and circuit boards so people can build their own computers, according to Zack Smith, director of the RepRap Research Foundation.

“It’s a printing press for the digital age,” Smith told Computerworld. “The goal is to have one on everyone’s desk. If it could build circuit boards, someone could design and build their own at home. Open-source electronics is a movement that’s really taking hold.”

While 3D printers have been commercially available for about 25 years, RepRap is the first that can essentially create its own structural parts, said team member, Vik Olliver, in a written statement.

Smith explained that unlike a regular printer that uses ink, RepRap heats up plastic and then squeezes it out into a line. The lines are built up into usable forms as they solidify. So far, the robot has made everyday plastic objects, like door handles, sandals and coat hooks. The machine has also successfully copied all of its own structural pieces.

For a full replication of all its own parts, Smith said that might be as far away as 20 years down the road. “Being able to replicate a computer chip would take a whole lot of precision,” he added. “For me, the exciting part is we’re building a tool that can build other things.”

Popularity: 1% [?]

Romantic human-robot relationships are no longer the stuff of science fiction – researchers expect them to become reality within four decades. And they do not mean simply, mechanical sex. “I am talking about loving relationships about 40 years from now,” David Levy, author of the book Love + sex with robots, said at an international conference held last week at the University of Maastricht in the south east of the country.

“… when there are robots that have also emotions, personality, consciousness. They can talk to you, they can make you laugh. They can … say they love you just like a human would say ‘I love you’, and say it as though they mean it …”

Robots as sex toys should already be on the market within five years, predicted Levy, “a sort of an upgrade of the sex dolls on sale now”.

These would have electronic speech and sensors that make them utter “nice sounds” when a human caresses their “erogenous zones”.

But to build robots as real partners would take a bit longer, with conversation skills being the main obstacle for developers.

Scientists were working on artificial personality, emotion and consciousness, said Levy, and some robots already appear lifelike.

“But for loving relationships – that is something completely different. In loving relationships there are many more things that are important. And the most difficult of all is conversation,” said the author.

http://timesofindia.indiatimes.com/HealthSci/In_2050_your_lover_may_be_a_robot/rssarticleshow/3148524.cms

Popularity: 1% [?]

Udi Manber sums up Google’s core challenge with this description of people’s expectations: “Here’s what I say, now give me what I need.” In other words, the company must use computers to comprehend humans, said Manber, the vice president of engineering in charge of Google search, in a speech at the Gilbane Conference here Wednesday.

“Ideally, we would understand your question, we would understand all knowledge, and match the two,” Manber said.

That’s not possible today, though, so Google takes a shortcut: Google tries to analyze and summarize all content, extend a user’s query into a summary version, and then match the two.

That sounds like a pretty long shortcut, but clearly Google has set its standards and goals very high. “We strive to answer every question, in every language, in a personalized fashion, in less than 100 milliseconds, for free,” Manber said.

In Manber’s view, humans are a puzzle only beginning to be unlocked. “The 20th century was about conquering nature. The 21st will be about understanding people,” he said, and computing is following suit. “The largest computing clusters in operation today are doing search, e-mail, social networking.”

Google starts opening up
Google is notoriously secretive about exactly how it decides which results to show in response to a particular query–a subject of high interest to companies counting on high placement or people hoping embarrassing Web pages will fade away–but the company has begun opening up. Manber promised in a blog posting in May to shed more light on search quality in coming months.

Manber shared several details about Google’s search quality process in his speech. For one thing, he said, there are more than 100 “signals” the company uses to determine the order of search results. Signals can be anything from language to location to a person’s previous search behavior–the latter only if the user enabled Google’s search history feature that personalizes results.

He also said the company has a team of “dozens” who do nothing but analyze the quality of search results, where quality is measured by hundreds of charts. These employees support the engineers who try to improve the search results, and Google wants those engineers to experiment with new search quality methods, Manber said.

Frictionless engineering
“The basic idea is to remove friction from engineers…An engineer with an idea does not ask for permission,” he said. Instead, the engineer tries the experiment, and Google meets once or twice a week to judge by the data whether the changes should be incorporated into Google’s main search results.

These experiments take place on a dedicated cluster of servers, Manber said. “My group at Google has at its disposal many thousands of machines, with storage measured in petabytes,” Manber said. “This is just for our own use, not for satisfying your queries.”

Google also tests search algorithm changes on users, different groups of whom receive different search results through a comparison process called split A/B testing. The end result: Google adopts search changes quickly and frequently. Google made 450 search algorithm changes in 2007, for example.

“We opened the way for any engineer to go improve things. Mostly because it’s based on data,” Manber said. “There is no separation of research and development. Everyone does both.”

Tough nuts to crack
Manber appears to take a perverse pleasure in difficult searches, relishing the fact that expectations for search match the rising capability and size of Google’s infrastructure.

He cited as examples out a series of searches whose intent generally seemed clear enough to a human: southeast utah news-airplane crash 10/25/06, hairstyles for ears that stick out, inflammation and pain under my rib, what is answer to this math problem 6x/10x, how many calories in a pound, if real number show else error blank excel. Of that collection, Google only provided good answers to the inflamed rib query, he said.

Straightforward queries also can be tricky. Google uses context to gauge what exactly “GM” stands for General Motors in the query “GM cars” but genetically modified in the query “GM foods.” Google offers various advanced search options, but its general policy is to use its single search box for everything.

“We have to understand as much as we can user intent and give them the answer they need,” Manber said.

Source:

http://news.cnet.com/8301-10784_3-9972034-7.html?tag=nefd.top

Popularity: 1% [?]

Nvidia today announced its second generation of Tesla floating point accelerators based on the GT200 series of graphics processors. It is the first big upgrade for the company’s supercomputing product portfolio – streamlining the offering and introducing double precision support as well as much more performance than the original 8-series, which was introduced one year ago.

High-performance computing (HPC) applications are likely to see several new technologies this week. In the hardware arena, AMD already announced its 1+ TFlop GPU earlier today and Nvidia is following with a GT200 based GPGPU, also claiming to be capable of hitting 1 TFlop per processing unit in single precision applications. Compared to the first generation, the floating point performance is up from 518 GFlops.

The new T10P processing unit represents a massive die, integrating 1.4 billion transistors and 240 processing cores, which is up from 128 cores in the 8-series of GPUs.

Nvidia has cut the deskside supercomputer (D870), answering to trends of customers who have been purchasing workstation graphics cards rather than an expensive external add-on, and is now limiting the product portfolio to a 4-GPU 1U blade and a Tesla add-in card. The S1070 blade integrates GPUs clocked at 1.5 GHz, a total of 960 processing cores, 4 GB of GDDR3 800 memory per GPU for a 16 GB total, 408 GB/s memory bandwidth and a total processing capability of 4 TFlops. Power consumption is up from 550 watts in the first generation to 700 watts

The blade will be offered with either 2 PCIe interfaces ($7995) or one PCIe connect ($8295), both of which are slightly more expensive than the S870 blade, which sold for $7500 at introduction.

The entry-level Tesla product remains an add-in card, in this case the C1060, which essentially represents Quadro graphics card on steroids. The card includes on T10P processor, 102 GB/s memory bandwidth and a power consumption rating of 160 watts, down from 170 watts of the previous generation. Nvidia said that thermal restrictions forced the company to clock the C1060 GPUs at 1.33 GHz instead of the 1.5 GHz in the blade. As a result, the C1060 will not hit 1 TFlops and is estimated to check in at about 900 GFlops.

The C1060 will be offered for $1699 MSRP, up from the $1500 price tag of the original C870.

Besides performance improvements, the T10P also delivers 64-bit or double-precision capability, which is required for most fluid dynamics and financial stream processing applications. Double precision is substantially more intensive than single precision calculations and with decrease the performance of the card dramatically. Nvidia told us that double-precision calculations will result in a 90% speed penalty and deliver only 100 GFlops per T10P processor.

There is also news surrounding the CUDA application platform, which Nvidia says can be used more any multi-core processing environment out there: This summer, the company will release a beta version of CUDA that developers can apply to multi-core CPUs. Nvidia claims that CUDA has been downloaded 60,000 times so far, but it is safe to say that there aren’t 60,000 developers working on HPC applications – and even Nvidia admits that most of those 60,000 developers are “playing” with CUDA trying to create “consumer applications.” The expansion into the CPU area could help the company reach a far greater developer base than it is able to attract with a GPU-only software foundation.

Technically, there is nothing that prevents from CUDA to also be used for ATI’s GPU products, Nvidia told us. However, not surprisingly, Nvidia said that it won’t be offering CUDA for ATI products and stated that “someone else can do that.” ATI offers its own high-level development tools called Brook+.

Source:

http://www.tgdaily.com/html_tmp/content-view-37955-135.html

Popularity: 1% [?]

A new paper from a group at Lawrence Berkeley National Laboratory, “The Potential of the Cell Processor Scientific Computing,” explores the performance of IBM’s Cell processor on some specific types of code commonly found in high-performance computing (HPC) applications. The programs used in the study are essentially smallish code blocks called kernels (see this older article for more on kernels and benchmarking) that implement typical algorithms like FFTs, stencil computations, and matrix multiplication. The paper compare Cell’s performance on these kernels to the performance of the Cray X1E, AMD Opteron, and Intel’s Itanium2.

The idea here is that Cell will be a commodity processor (at least that’s what the authors and IBM hope), so it’ll be a viable HPC alternative for the cost-sensitive academic research market. This paper represents the first formal academic attempt to decide if Cell hardware is something that researchers will want to invest in.

So how does Cell stack up in comparison to these three competitors? In a word, it screams.

First, the good news
Take a look at the following results for single-precision dense matrix multiplication, or GEMM (all numbers are Gflop/s):

Cellpm: 204.7
Cray X1E: 29.5
AMD64 7.8:
Itanium2: 3.0

The “pm” above means “performance model.” Because Cell hardware isn’t generally available for tests like this, the paper’s authors used a combination of performance projections and benchmarks on a cycle-accurate simulation of Cell that IBM has released. Real-world results should be very comparable to those in the paper, if not even better.

Note that the above results aren’t exactly typical. In some of the rest of the tests, Cell is only a mere ten times faster than the competition. Also, I should mention that the paper also looks into power consumption, and Cell still manages to trounce the other guys at performance/watt.

Needless to say, these results are extremely promising, and the authors of the paper clearly believe that Cell could change the HPC game if it is available in quantity and at commodity prices. I personally think that Cell’s “commodity” status outside of the PS3 is a bigger “if” than the paper presumes, but we’ll see soon enough.

Now for the caveats
So now that we’ve seen that Cell blows away the competition for these HPC kernels, that means that it’s going to completely dominate the next-gen console market and kill Itanium, right? Not exactly.

First, single-precision (SP) is the place where Cell really blows the doors off the barn, because SP is what game developers need. IBM made some compromises on double-precision (DP) performance, with the result that such performance is a fraction of what it is for SP. On DP code, Cell merely leads the pack for most of the tests.

The paper’s authors propose a microarchitectural improvement to Cell’s DP capabilities that they call Cell+, and they’re clearly hoping IBM will adopt their suggestion. Cell+ significantly enhances DP throughput with minimal changes, so we’ll see if IBM bites.

Another thing that should be pointed out is that the Cell used in the paper has full access to all eight SPEs, and not the six SPEs of the PS3. (Remember, one SPE is disabled for yield reasons, and the other is reserved for the system.) So keep this in mind when fantasizing about how these results are going to extrapolate to the PS3 hardware.

More important than the eight vs. six SPE issue is the fact that, due to the nature of the kernels used and the way that they were implemented for these tests, taking these results and trying to think about how a future iteration of Gran Turismo will look on the PS3 is a bit like comparing apples to cucumbers. Here’s why.

Programming models and the big picture
To get the kinds of mind-blowing results found in the paper, the Berkeley team took each kernel and custom-fit it to the bare Cell hardware using labor-intensive intrinsics and extensive hand optimization. They didn’t rely on IBM’s higher-level development tools, and they didn’t even code the kernels in C. In other words, they were operating at “Tier I” of the Cell programming complexity hierarchy. By taking into account things like the deterministic load latencies at the various levels of the memory hierarchy, this code was tuned and timed, cycle by cycle and word by word, to fit the cell hardware.

Our first Cell implementation, SpMV, required about a month of learning the programming model, the architecture, the compiler, the tools, and deciding on a final algorithmic strategy. The final implementation required about 600 lines of code. The next code development examined two flavors of double precision stencil-based algorithms. These implementations required one week of work and are each about 250 lines, with an additional 200 lines of common code. The programming overhead of these kernels on Cell required significantly more effort than the scalar version’s 15 lines, due mainly to loop unrolling and intrinsics use. Although the stencils are a simpler kernel, the SpMV learning experience accelerated the coding process.

Having become experienced Cell programmers, the single precision time skewed stencil — although virtually a complete rewrite from the double precision single step version — required only a single day to code, debug, benchmark, and attain spectacular results of over 65 Gflop/s. This implementation consists of about 450 lines, due once again to unrolling and the heavy use of intrinsics.

The authors were able to do this kind of custom fit because they picked a programming model based on data parallelism. What this means is that they had the eight SPEs doing identical work on different parts of a highly parallel dataset. When you’ve got all eight SPEs marching in lock-step through a large, parallel dataset, then you can really put all of the hardware on that chip to work in a dramatic way, as the paper indeed shows.

IBM, however, is pushing a task-based approach to parallel programming the Cell, where there are many individual tasks running concurrently on the different SPEs. This is way harder to code for and optimize than the data parallism-based approach used in the paper, but it’s also where the money’s at in the consumer and game markets.

In the end, what the paper demonstrates is that, for the HPC kernels that are amenable to a data parallelism programming model, then Cell’s particular combination of a software-controlled memory hierarchy (with deterministic load latencies) and an obscene amount of parallel execution hardware is clearly the way to go. This approach is dramatically superior to a general-purpose computing architecture with a hardware-controlled memory hierarchy from both performance and performance/watt perspectives.

If Cell doesn’t really catch on as a commodity part outside the PS3, I expect we’ll eventually be posting a news item about a lab somewhere (Iran?) that placed an order for 200 PS3 consoles, with plans to cluster them.

Speaking of the PS3, that’s going to feature mostly task-based programming, which as I just said is a different beast than what was done in the Berkeley paper. Also, the programming will be done at higher levels of abstraction from the hardware. So please, don’t read this and then assume that Cell will administer a similar drubbing to general-purpose architectures like Opteron, Itanium, and Conroe on all game, physics, and AI code.

Source:

http://arstechnica.com/news.ars/post/20060615-7071.html

Popularity: 1% [?]

Less than a week after Los Alamos National Laboratory’s Roadrunner supercomputer began operating at world-record petaflop/s data-processing speeds, Los Alamos researchers are already using the computer to mimic extremely complex neurological processes. Welcome to the new frontier of research at Los Alamos: science at the petascale.

The prefix “peta” stands for a million billion, also known as a quadrillion. For the Roadrunner supercomputer, operating at petaflop/s performance means the machine can process a million billion calculations each second. In other words, Roadrunner gives scientists the ability to quickly render mountainous problems into mere molehills, or model systems that previously were unthinkably complex.

Late last week and early this week while verifying Roadrunner’s performance, Los Alamos and IBM researchers used three different computational codes to test the machine. Among those codes was one dubbed “PetaVision” by its developers and the research team using it.

PetaVision models the human visual system—mimicking more than 1 billion visual neurons and trillions of synapses. Neurons are nerve cells that process information in the brain. Neurons communicate with each other using synaptic connections, analogous to what transistors are in modern computer chips. Synapses store memories and play a vital role in learning.

Synapses set the scale for computations performed by the brain while undertaking such tasks as locomotion, hearing or vision. Because there are about a quadrillion synapses in the human brain, human cognition is a petaflop/s computational problem.

To date, computers have been unable to match human performance on such visual tasks as flawlessly detecting an oncoming automobile on the highway or distinguishing a friend from a stranger in a crowd of people. Roadrunner is now changing the game.

On Saturday, Los Alamos researchers used PetaVision to model more than a billion visual neurons surpassing the scale of 1 quadrillion computations a second (a petaflop/s). On Monday scientists used PetaVision to reach a new computing performance record of 1.144 petaflop/s. The achievement throws open the door to eventually achieving human-like cognitive performance in electronic computers. PetaVision only requires single precision arithmetic, whereas the official LINPACK code used to officially verify Roadrunner’s speed uses double precision arithmetic.

“Roadrunner ushers in a new era for science at Los Alamos National Laboratory,” said Terry Wallace, associate director for Science, Technology and Engineering at Los Alamos. “Just a week after formal introduction of the machine to the world, we are already doing computational tasks that existed only in the realm of imagination a year ago.”

Based on the results of PetaVision’s inaugural trials, Los Alamos researchers believe they can study in real time the entire human visual cortex—arguably a human being’s most important sensory apparatus.

The ability to achieve human levels of cognitive performance on a digital computer could lead to important insights and revolutionary technological applications. Such applications include “smart” cameras that can recognize danger or an autopilot system for automobiles that could take over for incapacitated drivers in complex situations such as navigating dense urban traffic.

Los Alamos National Laboratory’s computation science team working with Roadrunner includes: Craig Rasmussen, Charles Ferenbaugh, Sriram Swaminarayan, Pallab Datta, all of Los Alamos; and Cornell Wright of IBM.

The PetaVision Synthetic Cognition team responsible for the theory and codes run on Roadrunner includes: Luis Bettencourt, Garrett Kenyon, Ilya Nemenman, John George, Steven Brumby, Kevin Sanbonmatsu, and John Galbraith, all of Los Alamos; Steven Zuker of Yale University; and James DiCarlo from Massachusetts Institute of Technology.

The Roadrunner is the world’s first supercomputer to achieve sustained operating performance speeds of one petaflop/s. In partnership with Los Alamos and the National Nuclear Security Administration, Roadrunner was built by IBM and will be housed at Los Alamos National Laboratory, where it will be used to perform calculations that will vastly improve the nation’s ability to certify that the United States nuclear weapons stockpile is reliable without conducting underground nuclear tests. Roadrunner also will be used for science and engineering such as energy research, understanding dark energy and dark matter, materials properties and response, understanding complex neural and biological systems, and biomedical applications.

Roadrunner was built using commercially available hardware, including aspects of commercial game console technologies. Roadrunner has a unique hybrid design comprised of nodes containing two AMD OpteronTM dual-core processors plus four PowerXCell 8iTM processors used as computational accelerators. The accelerators are a special IBM-developed variant of the Cell processors used in the Sony PlayStation® 3. Roadrunner uses a Linux operating system. The project’s total cost is approximately $120 million.

Los Alamos National Laboratory is a multidisciplinary research institution engaged in strategic science on behalf of national security. The Laboratory is operated by a team composed of Bechtel National, the University of California, BWX Technologies, and Washington Group International for the Department of Energy’s National Nuclear Security Administration.

Los Alamos enhances national security by ensuring the safety and reliability of the U.S. nuclear stockpile, developing technologies to reduce threats from weapons of mass destruction, and solving problems related to energy, environment, infrastructure, health and global security concerns.

Source:

http://www.lanl.gov/news/index.php/fuseaction/home.story/story_id/13602

Popularity: 1% [?]

Up until yesterday, Intel was showcasing its ray-tracing research using Quake 3 and Quake 4, which were nice demos, but did not exactly reflect fresh software. Over past several months, the company has been working one converting Enemy Territory: Quake Wars and we have to admit that the results are quite impressive.

Intel demonstrated ET: Quake Wars running in basic HD (720p) resolution, which is, according to our knowledge, the first time the company was able to render the game using a standard video resolution, instead of 1024 x 1024 or 512 x 512 pixels. Seeing ETQW running in 14-29 frames per second in 1280×720 has brought up our hopes for Intel’s CPU architecture, since we do not believe that CPUs would deliver a similar performance when rasterizing graphics. For the record, the demonstration ran on a 16-core (4 socket, 4 core) Tigerton system running at 2.93 GHz.

The game itself was vastly expanded when compared to original title. Intel’s Daniel Pohl showed how the engine now shoots three million rays in all directions, enabling collision detection based on rays alone.

Also, during the conversion, some effects were integrated by default, even if they had not been planned. One of those effects was fog shadow on the floor and physically-correct refractions of water. If you ever dived into a swimming pool or sea and looked up, you could have seen that the world is distorted. Now, ET: Quake Wars has the very same effect.

An impressive part of demonstration was looking at glass surfaces. Glass now reflects the environment to the tiniest detail – no LOD trickery here. Seeing a 200-window portal was quite an impressive demonstration of a situation when you are shooting rays into the environment. Check out our gallery to get more detail on this demo.

The icing on the cake was that the game was actually demonstrated running on a 64-bit Linux operating system. Intel stated that with ray-tracing, the company now supports 32-bit and 64-bit versions of Linux and Windows operating systems. We’ll see what will happen with Mac OS X support, but that should be on the cards as well.

Source:

http://www.tgdaily.com/html_tmp/content-view-37925-113.html

Popularity: 2% [?]