backtop

Hardware World's First Morphable Computer for DARPA
Marcus Yam (Blog) - March 23, 2007 12:23 AM
Print 50 comment(s) - last by ralith.. on Apr 3 at 5:52 PM

A diagram of MONARCH
Research giant Raytheon says its created the world's first computer architecture that can adopt different forms

Raytheon Co. claims to have developed the world's first computer architecture that can adopt different forms depending on its application. Dubbed MONARCH for “Morphable Networked Micro-Architecture,” Raytheon says the chip was developed for the Department of Defense to address the large data volume of sensor systems as well as their signal and data processing throughput requirements.

The MONARCH contains six microprocessors and “a highly interconnected reconfigurable computing array,” which according to Raytheon’s estimates “provides 64 gigaflops with more than 60GB per second of memory bandwidth and more than 43GB per second of off-chip data bandwidth.”

According to Raytheon’s presentation materials, the six processors inside the MONARCH are of RISC scaler architecture and are capable of Altivec-like SIMD operations. The processor contains 96 adders (fixed and float), 96 multipliers, 124 dual port memories, 258 address generators, 12MB on-chip DRAM, 14 DMA engines and 20 differential IFL (DIFL) ports capable of 1.3 GB per second each.

The MONARCH also performs as a single system on a chip, resulting in a significant reduction of the number of processors required for computing systems. The system’s polymorphic capabilities enable the development of defense systems that need to be very small size, low power, and in some cases radiation tolerance for such purposes as global positioning systems, airborne and space radar and video processing systems.

“Typically, a chip is optimally designed either for front-end signal processing or back-end control and data processing,” explained Nick Uros, vice president for the Advanced Concepts and Technology group of Raytheon Space and Airborne Systems. “The MONARCH micro-architecture is unique in its ability to reconfigure itself to optimize processing on the fly. MONARCH provides exceptional compute capacity and highly flexible data bandwidth capability with beyond state-of-the-art power efficiency, and it's fully programmable.”

In addition to the ability to adapt its architecture for a particular objective, Raytheon believes that the MONARCH computer is one of the most power-efficient processor available. Specifically, power requirements are estimated between eight to 50 watts nominal. In relation to energy efficiency and performance, the MONARCH outputs three to six gigaflops per watt.

“In laboratory testing MONARCH outperformed the Intel quad-core Xeon chip by a factor of 10,” said Michael Vahey, the principal investigator for the company's MONARCH technology.

The MONARCH processor was developed under a Defense Advanced Research Project Agency (DARPA) polymorphous computing architecture contract from the U.S. Air Force Research Laboratory. Raytheon is currently testing prototypes of the polymorphic MONARCH processors.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
cost?
By dome1234 on 3/23/2007 6:37:15 AM , Rating: 2
quote:
“In laboratory testing MONARCH outperformed the Intel quad-core Xeon chip by a factor of 10,” said Michael Vahey, the principal investigator for the company's MONARCH technology.


nice but maybe at 100 x the cost.




RE: cost?
By SurJector on 3/23/2007 8:08:32 AM , Rating: 5
quote:
“In laboratory testing MONARCH outperformed the Intel quad-core Xeon chip by a factor of 10,” said Michael Vahey, the principal investigator for the company's MONARCH technology.

outperformed in what ? speed of computation ? of what ? memory bandwidth ? thermal ? GFLOPS/W ? GFLOPFS/$ ? There are so many things that a computer can do better (or worse) than a competitor that such a non backed affirmation is useless.

Just my .02


RE: cost?
By Ebbyman on 3/23/2007 9:33:26 AM , Rating: 2
Well it was built for the government and you know how much they pay for toilet seats. Figure it out :-)


RE: cost?
By edge929 on 3/23/2007 2:49:02 PM , Rating: 3
I'll gladly pay taxes for the stuff that comes out of DARPA.

After all, they gave us the interent not to mention Metal Gear...


RE: cost?
By throughhyperspace on 3/23/2007 7:29:22 PM , Rating: 2
I was assuming he was referring to power consumption vrs processing capabilities, considering the paragraph before hand.. But, yes, some clarification would be nice.


RE: cost?
By splint on 3/24/2007 6:28:08 PM , Rating: 2
quote:
outperformed in what ?


They ran a Quake3 timedemo.


RE: cost?
By jak3676 on 3/23/2007 8:12:36 AM , Rating: 2
I think you are being VERY generous here. It's not like this chip will ever make it to large scale production. So the R&D won't be spread out over millions of chips. (of course the R&D may prove useful for other systems in the future.) I'd guess the costs are more like 1000x - 10,000x more than Intel.


RE: cost?
By thatguy39 on 3/23/2007 3:48:23 PM , Rating: 2
Stop guessing. None of you have enough information so you just "guess"? Childish...

You chose to guess instead of following the logical course of assumption; if they compared it to a Xeon, then it can be assumed the manufacturing cost of the chip is relative to the purchase price of a Xeon chip.

Don't let your fantasies and the need to bash out-weigh your logic.


RE: cost?
By throughhyperspace on 3/23/2007 7:42:14 PM , Rating: 1
Not enough information to guess, your absolutely correct. And, as logic dictates, even less information to assume. Especially something so absurd as the manufacturers cost being even remotely close to a Xeon. Maybe the entire operating budget, but they are obviously making a lot more Xeons.. The comparison had nothing to do with costs, but capabilities. Thus, "out perform".

And thats not some "need to bash", thats just the truth of economies of scale. As edge stated above, I gladly support such a budget. But, state of the art research isn't cheap.


RE: cost?
By ralith on 3/23/2007 8:48:45 AM , Rating: 3
What you all have to remember is that government fund research tech eventually trickles down to the consumer.


RE: cost?
By KaiserCSS on 3/23/2007 7:07:12 PM , Rating: 2
True. However, I have yet to get my hands on a stealth Cessna.


RE: cost?
By ralith on 4/3/2007 5:52:09 PM , Rating: 2
And you'd need one because...


RE: cost?
By saratoga on 3/23/2007 6:03:52 PM , Rating: 1
I love meaningless statistics like that.

Wow you mean using an unknown benchmark its 10x faster? That only sounds impressive until you realize that with the right benchmark, a GPU can be 100x faster then a CPU. Or a 100x slower.


The begining?
By dice1111 on 3/23/2007 9:41:29 AM , Rating: 2
quote:
"Raytheon Co. claims to have developed the world's first computer architecture that can adopt different forms depending on their application."
Could this be the begining of giving computers the means of becomming self aware? Is Terminators "Skynet" not far off in the distant future? Stay tuned to the "Paranoid Conspiracy Channel" to find out more...

Joking aside, I'm actually half serious here, to be honest. Maybe my imagination dial is tuned up to high... What do you gents think?




RE: The begining?
By TheDoc9 on 3/23/2007 10:57:30 AM , Rating: 2
skynet will only be formed when the programing logic is there. If a computer can become self aware at all, it would have to be programed as such either through learning itself or through programed experiences. The hardware has little to do with it. You could likely run skynet on a modern supercomputer, but since the programing isn't there that we know of, it would likely be on the retarded side of intelligence.


RE: The begining?
By Larso on 3/24/2007 5:53:45 PM , Rating: 2
I'd oppose and say that hardware has very much to do with it. The traditional CPU is way too sequential in operation. Too predictable. Our brain (which is appearently selfaware) is incredibly parallel, and slow, compared to computers. What we need for skynet is probably some new ideas at the silicium level...

*feels an irresistible itch to download a giant pulsating neural net on a fat Xilinx*


RE: The begining?
By peternelson on 3/25/2007 2:33:34 AM , Rating: 2
A conventional processor is sequentially programmed (although micro-op scheduling and maths may take place maybe a few things at once).

With pure uncommitted logic gates such as are in fpgas, millions of things can occur in parallel if you want them to. Even at a modest clock speed, the aggregate performance is great for suitable applications.

Neural nets have already been implemented in FPGAs.

Now the leap to Skynet.... Well who said humans had to program it. Humans can program computers, which can then program and configure the hardware. It just needs requirements and behavioural capture and specification. Algorithms themselves could be optimised iteratively by computers perhaps much better than a human could.

There is still needed expansion in terms of number of neurons in a typical brain, even a mouse brain.

We'd have to decide where to draw the line on defining "intelligent", "self-aware", "autonomous", "capable of learning" etc.

It's interesting that a baby learns from its experiences and environment. A computer capable of processing and incorporating such input might be capable of developing far beyond it's original programming.

By the time that's developed, we may have a practical holo-emitter device so it's avatar could be seen to walk around and interact with humans as Star trek's doctor.


RE: The begining?
By KristopherKubicki (blog) on 3/25/2007 2:53:30 AM , Rating: 2
quote:
We'd have to decide where to draw the line on defining "intelligent", "self-aware", "autonomous", "capable of learning" etc.

Turing Police anyone?


RE: The begining?
By thatguy39 on 3/23/2007 3:58:45 PM , Rating: 2
The construction of a computer brain is faaar too simplistic at this point. Just think of how many extremely powerful sectors are in the brain that work together to create thought. Computers are still nothing more than EXTREMELY powerful calculators.

BUT the foundation of thought is right around the corner. The key (extrapolating from current tech) piece to the AI puzzle are multiple cores. The more cores a computer has, the more likely the cores are going to be specialized for certain tasks, think GPU v. CPU. Imagine we have a computer with 64 cores, each specializing in a certain type of computational task (like sectors of our own brain)... the computer would then be able to "reason" and come to its own conclusion after receiving input from every core of its "brain".


RE: The begining?
By throughhyperspace on 3/23/2007 8:34:42 PM , Rating: 2
While multiple cores or possibly more relevant, cell type architectures would provide the headroom necessary to produce AI, it by itself will not simply occur at 64 cores. Besides, many universities have such computers. Shit, Cray is has one available with 60000 cores. http://www.dailytech.com/article.aspx?newsid=4931

Reason, implies logic, which computers have plenty of. You give them an algorithm and they give you an answer faster then we can. Self consciousness from a "formal logic system", however, is still being debated in the philosophical community. http://en.wikipedia.org/wiki/Artificial_intelligen...

"Coming to its own conclusion" has to be based on rules. Either an algorithm that is programed by a human, or one programmed by the computer itself. This infers learning, which, as the docs says, has to be programed at some point. But once the program can adapt its decision making by acquiring and deciphering information, then we will be on the cusp of truly amazing.. Perhaps scary, as the OP jokes.. But the first two posts bring up a very interesting thought: "morphing hardware" combined with self-adapting programming seem like the perfect environment to develop "artificial consciousness". Considering the brain does, in fact, adapt its "hardware", rather then being static like the 64 core x86 PC you envision.


RE: The begining?
By throughhyperspace on 3/23/2007 9:03:42 PM , Rating: 2
In interest of furthering the discussion, there are very intelligent people that believe it is only a matter of evolution for robots to be born in our image, inherit the earth, and advance in ways we cannot, since we have plateaued as a species.

http://web.media.mit.edu/~minsky/papers/sciam.inhe...

He also implies that parallel programming is the necessary foundation of artificial consciousness and goes on to say:

quote:
In order to think effectively, you need multiple processes to help you describe, predict, explain, abstract, and plan what your mind should do next. The reason we can think so well is not because we house mysterious spark-like talents and gifts, but because we employ societies of agencies that work in concert to keep us from getting stuck. When we discover how these societies work, we can put them to inside computers too. Then if one procedure in a program gets stuck, another might suggest an alternative approach. If you saw a machine do things like that, you'd certainly think it was conscious.


And for those that are still with me and will probably not read the essay, he concludes:

quote:
Will robots inherit the earth? Yes, but they will be our children. We owe our minds to the deaths and lives of all the creatures that were ever engaged in the struggle called Evolution. Our job is to see that all this work shall not end up in meaningless waste.


So, to the OP: Its not scary, its our destiny. lol..


RE: The begining?
By throughhyperspace on 3/23/2007 11:41:33 PM , Rating: 2
Alright, last one. For those interested, another essay - same subject. This one is a more specific prediction chronologically. Which is pretty funny considering it was written around the time DOS 5.0 was released..

http://www.frc.ri.cmu.edu/~hpm/project.archive/rob...


RE: The begining?
By josh65 on 3/26/2007 10:55:52 PM , Rating: 2
OMG DUDE I THOUGHT THE EXACT SAME THING MAN!
Listen to the nude guy that travels back in time to save us.


Anandtech is falling short
By ranger203 on 3/23/2007 7:00:53 AM , Rating: 1
Anand is lettings us down, I would expect a well written article like this to have extensive Benchmarks, speed vs. cost analysis, and an estimated each cost per qty of 1000. I mean they are really dropping the ball on this article. I need to see how fast Freecell will load on one of these bad boys!!! :-)




RE: Anandtech is falling short
By James Holden on 3/23/2007 8:24:54 AM , Rating: 5
It hasn't been Anand's site for a year buddy.


RE: Anandtech is falling short
By rushfan2006 on 3/23/2007 3:28:24 PM , Rating: 2
quote:
It hasn't been Anand's site for a year buddy.


WTF...do you have no sense of humor? Did you read his post being the first few words and just go all EMO?

You and everyone who voted him down have no sense of humor.

They guy was just cracking a joke..my god don't take life so serious all the time.


RE: Anandtech is falling short
By GhandiInstinct on 3/23/2007 4:41:31 PM , Rating: 3
Aye aye,

Dailytech mods and readers are uptight, anal retentive star wars geeks, Wii fanboys, Sony haters, conservative, eggheads.

Omg no /sarcasm there, time to mod me down...your only power mwhahahaha...


RE: Anandtech is falling short
By WhiteBoyFunk on 3/23/2007 5:53:29 PM , Rating: 2
Believe it or not, slapnuts, that I am convinced you are a much bigger geek than myself, the admin here and at least 50% of the DT populace.


RE: Anandtech is falling short
By shaw on 3/24/2007 1:01:14 AM , Rating: 2
I'm waiting for Anandtech's results on how well it overclocks.

Or does it evovle into a faster CPU? O_o


RE: Anandtech is falling short
By mars777 on 3/24/2007 4:07:52 AM , Rating: 2
But we all have something in common.
We are all geeks, since we read and comment this news :)

Nevertheless people here are not haters, just bashers and fanboys.
A lot of them are MS fanboys since the fact they bash the PS3 in the Folding@Home thread.

I personally hate people who bash for no reason (in fact that was one of the few things sony did good this year).


World's FIRST? Don't believe the PR hype
By peternelson on 3/25/2007 2:23:18 AM , Rating: 2
I'm sure this will give good performance, but reconfigurable hardware like FPGAs (field programmable gate array) have been around for 10-15 years.

THIS IS FAR FROM A WORLD FIRST!

Xilinx and Altera are two main suppliers of the chips, there are other players too.

Quantel have used Altera FPGAs in specialised television and print graphics computer systems (got one under my desk) to surpass what regular processors could achieve in image processing performance at the time.

As for computer systems using this tech: CRAY and Silicon Graphics both have Xilinx based supercomputer systems or options.

As for boards you can stick in a PC and cluster, check out Starbridge systems (who have many xilinx chips on a node board specifically for high performance computing). Their software (not the only solution but one adopted as an option by several leading players) makes it simpler to reconfigure the resources of the programmable hardware for the job you want it to do, even without being an elite hardware design engineer.

I've got my own custom reconfigurable computer right here on my desk. It's a PC with SIX special pci cards in each with a Xilinx FPGA. They are networked together by my custom interconnect and also communicate with the host system. I can optimise the processing resources for a particular task like finding prime numbers. A second later it can be reconfigured to perform some other task well.

PCIe or PCI cards with reconfigurable FPGA hardware are available from several vendors, off the shelf.

There is also lots of work been/being done on PARTIAL reconfiguration, and the FPGA could even reconfigure part of itself commanded by the rest of the chip.

Establishments like CERN and Fermilab have long used FPGAs in their sensor acquisition equipment. Satellites use them as cpus so the chips can be reconfigured to bypass a fault that develops because of a cosmic ray. Some set top boxes and cars use them for early time to market (better lead time than an asic) and adaptability to changes in emerging standards.

There are already Torrenza based Hypertransport enabled plug in modules eg from DRC using Xilinx or Altera FPGAs to put reconfigurability into your socket 940 or socket F AMD-based computer.

I'm sure what Raytheon made is cool, but I don't see anything that makes it in any way a world first.




By spartan014 on 3/26/2007 5:15:19 AM , Rating: 2
Good post... I got more information from this single post than all others combined. Thanks..

By the way, One doubt.. FPGA chips will take some time to be reconfigured entirely, right? The impression I got from the article was that they are trying to do it on the fly with as little delay as possible.

Anyway, nice post..


By alienbibin on 3/26/2007 5:25:09 AM , Rating: 2
“The MONARCH micro-architecture is unique in its ability to reconfigure itself to optimize processing on the fly. MONARCH provides exceptional compute capacity and highly flexible data bandwidth capability with beyond state-of-the-art power efficiency, and it's fully programmable.”

I dont think FPGAs are capable of this...


Very Interesting
By Cogman on 3/23/2007 9:16:01 AM , Rating: 1
This is a very interesting product. I doubt, however, that it is going to be adapted to the consumer market for a LONG time (if ever). The Idea is sound, and even exciting. Imagine that you are able to convert your P4 into a Core 2 Duo with just a couple of instructions. That is basically what this thing is able to do. (Further more, it can convert from a Core 2 Duo to a Geforce 8800 and back again). Of course, that is as long as there are enough transistors available to the thing. The reason I say it is bad for Intel AMD and the likes is because they will have a hard time selling these things a new when all it takes to update them will be a new driver. And with a new driver you will see considerable performance increases.

They could try and sell their new drivers. But I'm pretty sure that I would be more then tempted to "Test" the top of the line driver by simply downloading it off of the Internet (Yeah, you know you would probably do the same thing, It would almost be the same principle as overclocking. Only now you know for sure that your CPU's are the same, just configured differently).

Not to mention, The open source community would have a hay day with these things. Onces they figured out how to make drivers for them you would probably start to see linux on steroids (and Intel will become the new Microsoft, trying to make you pay for software to run their products.)

I think this is a great Idea though, something that would really give the computing industry a big processing power boost. But it simply will not happen because it would kill the market for traditional processors.




RE: Very Interesting
By nafhan on 3/23/2007 12:02:48 PM , Rating: 2
If something like this did come to market (assuming Intel/AMD produced them), it would not kill the processor market.

However, moving to a new process node would be a much bigger deal, as the main performence limiter would be the number of transistors. For example, adding transistors would allow extra processor cores, GPU pipelines, sound cards, cache, etc. to be added to the main CPU. Plus, there would always be work into developing more efficient transistors.

Driver updates would definitely play a much bigger role. Imagine going from DirectX 10 hardware to DirectX 11 hardware with a driver update!


RE: Very Interesting
By Cogman on 3/24/2007 12:08:49 AM , Rating: 2
I can't really see it helping the processor market all that much. As things stand right now, A new processor core is developed once every 6 months (sometimes longer) and many new cores are simply minor changes to the old ones. Then the only real difference with a new processor release is somebody that monkeys with the Multiplier of FSB. With this though, You would be almost garenteed to have the top of the line processor every time you buy a new core, there would be no middle ground to really rest on as the major differences could potentially be easily changed with a architect re-write.

You are right that it would not completely kill the market, But it would harm it as people would be far less inclined to buy a new server then they are now. (average Joe would probably not be effected by the shift as much, in fact they probably would not even notice the difference)


Nice
By Hypernova on 3/23/2007 6:06:45 AM , Rating: 2
The spec looks like a FPGA on steriods.




RE: Nice
By hpdeskjet697c on 3/23/2007 8:46:56 AM , Rating: 2
heh, exactly what i was thinking


Price tag?
By leidegre on 3/23/2007 7:44:32 AM , Rating: 2
How much does it cost? Becuase if it out performs an Intel Xeon by a factor of 10, (including the price), I don't see the point. Of course we always have those with money, and the purposes, but if an adaptive architectures is that much better and a desktop feasability, I say go for it!




RE: Price tag?
By ralith on 3/23/2007 8:51:03 AM , Rating: 2
They said that they had a specific purpose in mind in the article i.e. a processor for radars.


To Burst Your Bubble.
By rupaniii on 3/23/2007 10:30:21 PM , Rating: 2
I guess some people don't get Government Budgets, projects, and design and development.
There are fabs the US gov in particular have on their own.
Or have their own lines made for them by IBM or other companies.
32bit with 64bit extension chips are childs toys. Sun UltraSPARC T1s are nice, but, still.. Cell is a bit of a glimpse.
There are RISC and PA-RISC like designs that violate every patent on earth,and unless you are really IN, you ain't going to use it.
This, in particular, is one of those kinds of devices.




RE: To Burst Your Bubble.
By Larso on 3/26/2007 6:05:16 PM , Rating: 2
So, if I understand you correctly, you hint that the government has secret CPU's so powerful that our current top of the line Intel CPU looks like a child toy?

Why don't you enlighten us a little bit more, because UltraSPARC T1 and Cell does not impress much. PA-RISC? Thats kind of oldschool... RISC? Yeah, thats a nice concept. For microcontrollers for instance.


Johnny!
By Mitch101 on 3/23/2007 9:28:50 AM , Rating: 3
Chief:
Johnny what can you make of this?

Johnny:
I can make a Hat, or a Broach or a Pterodactyl!




WOW
By scrapsma54 on 3/23/2007 1:05:41 PM , Rating: 1
Ok the wattage is impressive, but what fab is this made on? 130, 90, 65nm? It seems like cell may have a soon to be rival. It is a shame that this may never make it to consumers. Lol, programmers will no longer have to optimize their coding so it runs specifically for it. This probably is a much cheaper alternative to super computers, or at least multifunction platforms. well at least the multi-platform pc that is.




RE: WOW
By scrapsma54 on 3/24/2007 2:40:08 PM , Rating: 2
Ok, seriously, quit down ranking me. I know its the same person everytime and you just need to quit.


jawesome
By Howard on 3/24/2007 8:24:15 PM , Rating: 2
It's morphin' time!




By Hawkido on 3/27/2007 3:47:01 PM , Rating: 2
I hope part of this tech trickles down to the console market. Could you imagine game consoles, only being outdated when the silicon is maxed out because of lack of transistors? Each piece of software could reconfig the CPU for every different scene with dynamic hardware load balancing and achieve maximum use out of the Silicon available.

You gotta know the PS3 has alot of wasted silicon in it, not all of it can be used at once. But imagine if the same amount of silicon were used with this tech. Backwards compatibility issue, just adjust the hardware for that particular software. Latest software out, configure it to maximize the performance.

Your game console would have a 10 year life (well till the silicon burnt out). Games would be Dev'd to perfection by the time the system was mature, and future products would retain full backwards compatibility through reconfigured hardware. Heck the entire system could be scrunched onto that chip.




10 times perfomance explained
By ceman on 3/31/2007 3:53:25 AM , Rating: 2
From an article on Newswire
Current estimates by Raytheon put the MONARCH chip at somewhere between three and six GigaFLOPs per watt, depending on the application, with an average of five GigaFLOPs. The company claims the Cell processor in the Sony Playstation 3 runs at an estimated 2.2 GigaFLOPS per watt and the Intel Xeon runs at around 0.5 GigaFLOPS per watt, making MONARCH twice as power efficient as Cell and 10 times more efficient than Xeon.




Added benchmark
By ranger203 on 3/23/2007 7:02:17 AM , Rating: 1
.....And if there is any performance degragation while play quake 3 next to the old Chernobyl reactor




So,
By darkavatar on 3/23/2007 9:39:06 AM , Rating: 1
Is that what CELL is supposed to be?




"The Space Elevator will be built about 50 years after everyone stops laughing" -- Sir Arthur C. Clarke









botimage
Copyright 2012 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki