The first computer system incorporating IBM’s TrueNorth computer chip was conceived in Deer Valley, Utah last month, in a mere four days at the Future in Review conference.
Under the direction of Mark R. Anderson, a handpicked group of computer executives from different companies designed a desktop supercomputer system as part of a CTO challenge at Anderson’s annual conference, also known as FiRE.
They designed the Pattern Computer as a general purpose computer, one that runs on any operating system, and is highly efficient, extensible, scalable and unbelievably fast. It will be built using commodity hardware, connect to the new 100Gbps Pacific Research Platform, and run TrueNorth and other new brain-inspired chips as co-processors.
The Pattern Computer is designed to discover patterns in big data – not just the ones expected or postulated but revealing those “where we might otherwise not see them,” said Anderson, who is well known within tech circles for his accurate predictions. Anderson is CEO of Friday Harbor-based Strategic News Service and publisher of the SNS Global Report, a weekly newsletter devoted to technology and its global economic impact.
“Think of it as a camera for data rather than a camera for images,” he said. “Unlike past computers, the (Pattern) computer will say, ‘I have found a pattern you haven’t seen before ‘– it could be in physics, in climate change, or in anything. Our challenge will be to interpret rather than instruct, which is a big, big deal.”
“This could be the advent of an entirely new computer age, and a revolutionary change in human and computer interaction.”
FiRe attendee Aaron Contorer, who formerly served as a tech advisor to Microsoft co-founder Bill Gates, served as general manager of Visual C++ and helped build Windows, MSN, and Visual Studio, immediately expressed interest in working with Anderson and the CTO team.
The Pattern Computer would mean “users could have access to dramatically more powerful and less expensive computer capacity for their larger problems — basically a desktop supercomputer,” said Contorer, the founder and chief executive of FP Complete, based in San Diego. “Having access to a huge number of compute cores on a single machine with an internal bus would slash costs and raise speed, compared to the same number of cores on a cloud connected with a traditional network.
“It would also mean that high-performance computing could be done in anyone’s office or spare room, rather than needing a specialized data center with load of cooling and a huge power supply.”
With the Pattern Computer, FP Complete could put its programming technology on a much more cost-effective hardware platform, Contorer said. But it would mean the most to chip makers struggling to survive against cheaper Chinese chips, he added. “If a research platform like this, with help from software tool makers like us, can get software apps to really use multicore hardware, then suddenly the market for higher-end chips might explode.”
Larry Smarr, a physicist and leader in scientific computing, supercomputer applications, and Internet infrastructure, agrees that the core Pattern Computer technology will be ubiquitous in 10 years, not only in desktop computers but in corporate data centers, smartphones and other devices much like the DSP or digital signal processor is today.
“This is a new capability in the computer ecology of our world,” said Smarr, founding director of Calit2, a UC San Diego/UC Irvine partnership. Smarr has tracked brain-inspired computing developments since 2000, and helped Anderson create the CTO challenge. Envision, Smarr said, “real time speech recognition where you can tap into deep learning requiring vast amounts of data, giving you the ability to reach back into the cloud to tap into hundreds of millions of sounds and hundreds of languages to refine what goes on in the smartphone – in real time.”
Commercializing the Design
The Pattern Computer will be built at Calit2’s newly formed Pattern Recognition Laboratory, dubbed “The FiRe Lab,” which also made its debut at FiRE. Under the direction of CTO team member Ken Kreutz-Delgado, professor of Intelligent Systems, Signal Processing and Robotics, Jacobs School of Engineering at UCSD, the goal is to have a working a working prototype completed in the next 12 months.
Building the computer at the new Calit2 lab is important because while IBM’s TrueNorth chip, first unveiled in mid-2014, is not yet commercially available, Kreutz-Delgado is one of about 30 researchers and institutions that have the chip. The new computer will also connect to the Pacific Research Platform, the ultra-fast, separate Internet-like network linking more than 20 research universities that Smarr is pioneering.
Anderson said the efforts of the FiRe team were analogous to the advent of IBM’s personal computer and most likely would have taken a large enterprise two years and some $50 million to do what his team accomplished in just four days. “I’m blown away,” Anderson told the team after they shared the design and computer system specs at the conclusion of the conference.
Smarr, who has helped Anderson with FiRE and the CTO challenge for more than a decade, said this challenge was the most difficult yet, and “the most unlikely to be successful.” He didn’t think they would do it.
Within hours of the Pattern Computer debut, several investors at the conference offered funding and urged Anderson to form a company to commercialize the machine. Anderson did so last week, forming Pattern Computer, Inc., with headquarters in Friday Harbor, WA.
Publicly, IBM has not announced that it is making a computer based on the TrueNorth chip. The company declined to comment for this story. John Kelly, a senior vice president at IBM who oversees IBM Research, told the MIT Technology Review last month that IBM is talking to major computer and phone manufacturers about commercializing the TrueNorth chip, and working on the next version.
Using the TrueNorth Chip
The Pattern Computer would not have materialized if not for several recent technology developments, the first being IBM’s TrueNorth “brain-inspired” computer chip. TrueNorth operates much like neurons and synapses in a human brain, storing and processing information in a distributed and parallel way rather than the traditional way most computers operate today. The radically new chip, which is also extremely energy efficient, has been under development by a team led by Dhamendra Modha, IBM Fellow, Chief Scientist, and IBM Senior Manager of Cognitive Computing, for at least seven years.
TrueNorth “is a supercomputer on a postage stamp with the battery power of a hearing aid,” Modha explained to FiRE attendees. Modha was a member of the initial FiRE team tasked with Anderson’s challenge, but he declined to comment for this story. While he was at the conference, he was not part of the final CTO design presentation.
TrueNorth is “tailored for a new class of distributed, highly interconnected, asynchronous, parallel, large-scale cognitive computing architectures,” IBM has stated in press releases.
While the TrueNorth chip is said to be “brain-inspired,” the Pattern Recognition computer will harness its power with a different goal: To find natural and irregular patterns “not only in n-dimensional physical and mathematical spaces, but also in n-dimensional time,” Anderson wrote in his 2013 newsletter on his idea for a “pattern recognition processor.” His vision of “The Most Important Chip Not-Yet Invented” led Modha to call it “prophetic.” And it led to an invitation to see Modha’s work prior to TrueNorth’s 2014 unveiling and to speak at its launch.
Pattern recognition is a key technique Anderson has refined to accurately see the world and make predictions about future outcomes. Recognizing patterns and when patterns are broken, is a natural part of the brain, and anyone can teach himself how to see the world this way, he believes. Creating a computer to aid this process is the next evolution of his vision.
Anderson didn’t start thinking about making a pattern computer until six months ago after talking with Smarr about TrueNorth, brain inspired computing and how the world was “reaching a phase transition point in technology where exponentials cross a threshold of some qualitatively new capability,” said Smarr, who has followed brain-inspired computing efforts for the past 15 years.
Birth of the Pattern Computer
Armed with two requirements – that they build it to Anderson’s specifications and that Calit2 could actually build the machine – the group sketched the plan over four days, said FiRE CTO Challenge spokesman Brad Holtz, CEO at Cyon Research Corp., a Bethesda, MD-based engineering and software think tank.
After group discussions on the specs and aspects of big data and pattern recognition, the team decided to focus on hardware and only to offer a strategy for software rather than diving into any additional programming issues, said Holtz.
Mike Riddle, director, Meta Model Architecture at The Center for Understanding Change, based in Silver Spring, MD and a co-founder of Autodesk, quickly formulated an idea of how to proceed. He “took the lead on hardware and software design. The sketch he did that first night formed the basis of the design,” said Holtz.
“When we looked at his sketch and started fleshing it out, we recognized where the special stuff was,” Holtz said. They recognized what one of the hard problems with bringing the design to life but also realized they knew how to solve it. “People got excited when we realized we didn’t have to invent anything,” he said.
“This design would not be pragmatic or rational without the Biomimetic Programming toolkit that Mike has been working on for the last 15 years, Holtz said. (Biomimetic refers to technologies that mimic biochemical processes.) “This really becomes practical because of that toolkit. It might be possible without it, but (developing) it would be years and years away.”
Riddle’s work began with rethinking the structure of software design. Much of the fragility of complex software comes from the way parts of the software are connected and linked, particularly as complex software is patched and grown over time. Riddle’s approach has much in common with biological systems, which communicate internally via “messages.”
In this approach, messages need know nothing about the objects that receive them, but the objects know which messages they can recognize and operate on, just as cells “know” what to do when receiving a chemical or electrical stimulation. The extent of Riddle’s toolkit is similar in scope and scale to Microsoft .Net. The toolkit and methodologies are based on a generic subset of C++ code, which allows it to run on any hardware.
Biomimetic programming’s message-passing architecture will give the Pattern Computer the control necessary to parse large quantities of data and information as quickly as it needs, and to unearth patterns following biological principles rather than the traditional approach where a programmer tells the computer what to look for.
The Need for a High Speed Traffic Cop
One of the biggest challenges in making the Pattern Computer work will be configuring the high-speed memory access needed for multiprocessors in order to ensure tasks are prioritized and the computer doesn’t get stuck in an endless loop searching for patterns. “It’s going to be interesting to design how tasks are prioritized, but Mike (Riddle) has solved it in a similar situation at least once before so it really is solvable,” Anderson said.
The design of the Pattern Computer is open and flexible, and will accommodate many chips as co-processors, not just TrueNorth. The Pattern Computer could harness Micron’s Automata Processor, a purpose built architecture that addresses processing challenges related to pattern matching and data analytics, for example. And it could use the newly announced stealth chip under development by Dan Goldin, past NASA administrator, founder and chairman of Intellisis Corp., who discussed his new “brain-inspired” chip at FiRE.
“This is the advent of an entirely new computing era,” said Anderson, adding that commercialization for the Pattern Computer might be one to two years away.
“Mark pointed out, and I agree with him, that this could be as significant as the invention of the PC,” said Holtz. “But only if we execute on it.”
The Pattern Recognition Computer CTO Challenge team included:
- Mark Anderson, CEO, Strategic News Service
- Larry Smarr – founding director, Calit2, a UCSD/UCI partnership
- Hugh Bradlow, chief scientist, Telstra Australia’s leading telecommunications and information services company
- Ty Carlson, Senior Manager, Technical Program Management, Amazon, Seattle
- Steve Coy, President, TimeLike Systems, Albuquerque, NM
- Brad Holtz, CEO at Cyon Research Corp., Bethesda, MD
- Ken Kreutz-Delgado, professor of Intelligent Systems, Signal Processing and Robotics, Jacobs School of Engineering at University of California San Diego
- Dhamendra Modha, IBM Fellow & IBM Chief Scientist, who led the IBM team creating TrueNorth, San Jose
- Mike Riddle, co-founder of Autodesk and Director, Meta Model Architecture, Center for Understanding Change, Silver Spring, MD
- Greg Schmergel, CEO and co-founder of Nantero, Woburn, MA
- David Schoenberger ,CIO CertainSafe, Colorado Springs, CO
- Tim Stonehocker, CTO Soundhound Inc., Santa Clara, CA