I am getting really tired of these articles that come out every other month touting some new technology as the replacement to the standard npn or pnp transistor model. First it was the cross-switch then it was some stupid crystal, then a strand of DNA...give me a break already. All these exotic solutions exist solely in academia and never end up being viable in the commercial realm. Yes yes, I hear you yelling "But a lot of past inventions came from academia!!". I am referring just to transistors here, and the tendency to hype up every new concept as a potential replacement to existing fabrication techniques.
A real breakthrough actually came in 2007 when the industry replaced polysilicon gates with high-k metal ones, a breakthrough that was accomplished through research at IBM, not a university.
In addition, I am sick of every transistor article mentioning Moore's "Law". Originally Gordon Moore noticed a trend, and made a prediction. Now companies are forcing themselves to innovate so as not to "fall behind" the Moore curve. This is a good thing, as it causes innovation, but the trend he noticed should not be considered a law by any means.
They will have a great opportunity at year's end when Intel gets hit with an antitrust lawsuit in Europe, which should take away some of their Kentsfield momentum. Hope AMD can capitalize.
"I'm in the 30-40% camp that believes that there's really not anything magical about the human brain."
This statement is astonishing, particulary because it is coming from an AI "expert". Aside from the fact that the brain has allowed humans to create the most complex tools and technology the planet has ever seen, all the while not having to think about breathing or keeping our blood flowing, monitoring insane amounts of chemicals, repairing itself after injury, retaining intricate memories from decades ago, and even attempting to build machines to do the same thing, the brain is not that complicated or magical(note the searing sarcasm). Pearson has completely discredited the humain brain. Considering our research on the brain has only really been excelling in the past 30 years, and that we are nowhere near understanding its full complexity, Pearson instantly loses credibility IMO.
It has been long understood in AI that the only way to get a machine to truly behave like a human is through pure volume of information. People know that if they leave some salad out in the open and don't refrigerate it, it will spoil. To get a machine to come to the same conclusion, it needs to be programmed to know that specific fact, but still doesn't know anything about why or what a refrigerator does. It is only through brute force of millions of facts that we can get a machine to even come close to the range of knowledge that the human brain contains which is the basis for projects like Cyc. This leads to another point: that the human body itself is perhaps the most complicated organism on earth. Simply getting a robot to write its name in cursive is an enormous task to accomplish, as mechanical parts are no where near as efficient as human joints (although this is improving, they are still not sending the type of precise electrical signals to the brain like our bodies do). No, we don't need to be programmed to be able to catch a fly ball at a baseball game, we just judge where the ball is going. A robot must be told trajectory, wind speed, velocity of the ball, and all sorts of vector math just to be able to accomplish what we see to be a simple task. And on top of that, it can't learn how to do it again.
A fully functional "learning" AI computer is at minumum 50 years into the future, more likely 80-100 years, not 10 to 15. People in the 80s were saying the same thing back then as well, and when people wanted to know where there intelligent robotic math tutor was, they found out that it was a near impossible task to create such a robot. Either way, nothing will even come close to mirroring a human being completely in terms of complexity in our lifetimes.
What needs to be remembered is that Cell is a strictly in-order CPU, meaning that it cannot execute instuctions out of order. Therefore, hand-optimization can only get you so much. If you are waiting on an add, you are waiting on an add; not much can be done in that sense. It really comes down to compiler logistics and how the various tasks of branch prediction, locality and "cache" come into play/are optimized. Remember that Cell has no real cache but a virtual one. It ultimately comes down to the compiler, which I am sure is pretty beefy in this case.
The only component of the Cell that would remotely require an FPU is the PPC core which handles the physics and AI. However it does the job fine with out one due primarily to compiler optimization. The compiler is key for Cell because all of it's cores are "in-order", meaning that instructions must be executed in the order they are written and recieved. An out-of-order core requires much more complex circuitry along with surface area and power consumption, and in turn heat dissipation. With an optimized compiler, one can make an in-order core behave similar to an out-of-order one. Branch prediction also improves. Keep in mind that these cores are working with no cache and we can see why the compiler is crucial.
As a stabilizing mechanism you can use magnetorquers. They are inductive coils implanted into the side panels of the satellite. The navigation system detects rotation and fires off the coils when neccessary to counteract the magnetic field of the earth and slow down the rotation long enough to take a picture. It is quite an intelligent solution as it takes up no internal payload space.
I am a member of a team building a pico satellite at the University of California. The actual price for our entire project hovers around $50,000, of which 40k go to launch costs. This is the price tag for a Russian launch, I am sure the US would charge quite a bit more. The problem with these pico satellites is actually integrating a full fledged OS onto OTS parts. Data memory and program memory are in limited supply, along the lines of KB. We are using an Atmel CPU and are running AVR library from the http://www.avrfreaks.net/ website. The actual module we are writing is for the CPU scheduler. The interupt handler, bus protocols, etc are already written. Running this minimal amount of code is beneficial and doesnt overload the small amount of data memory we can utilize, so basically you can run less than a full on OS like linux and still get away with it.
"What imaging can you do for cheap -- what resolution, what wavelengths?"
The imaging comes down to the type of camera and it's power consumption. Image processing really loads the CPU and draws a lot of power, so compression algorithms are important. JPEG does the job fine, but is limited to around 320x res (in color) given our power constraints. The camera we are using is a CMOS imager, which works through the charging of capacitors relative to the intensity of incoming light. The wavelength of operation ultimately comes down to the FCC. They allocate you a bandwidth to use, most likely 2m or 440MHz. Other functions the satellite may serve is as a HAMsat, or a HAM radio repeater in space. You can tune you radio to the sat freq, and hit any place that the satellite covers with your signal. The preferred mode of operation is digital, so sat comm utilizes packet radio, which loosely resembles the TCP/IP stack. As far as a webserver goes, Im sure you could do it, but it would be limited to the length of one satellite path. Guess thats enough time for a 30 second pr0n clip...
Slashdot Mirror
I am getting really tired of these articles that come out every other month touting some new technology as the replacement to the standard npn or pnp transistor model. First it was the cross-switch then it was some stupid crystal, then a strand of DNA...give me a break already. All these exotic solutions exist solely in academia and never end up being viable in the commercial realm. Yes yes, I hear you yelling "But a lot of past inventions came from academia!!". I am referring just to transistors here, and the tendency to hype up every new concept as a potential replacement to existing fabrication techniques.
A real breakthrough actually came in 2007 when the industry replaced polysilicon gates with high-k metal ones, a breakthrough that was accomplished through research at IBM, not a university.
In addition, I am sick of every transistor article mentioning Moore's "Law". Originally Gordon Moore noticed a trend, and made a prediction. Now companies are forcing themselves to innovate so as not to "fall behind" the Moore curve. This is a good thing, as it causes innovation, but the trend he noticed should not be considered a law by any means.
According to the computers, it should be day 9338.
They will have a great opportunity at year's end when Intel gets hit with an antitrust lawsuit in Europe, which should take away some of their Kentsfield momentum. Hope AMD can capitalize.
God FORBID that it actually takes time to do things.
Pearson states:
"I'm in the 30-40% camp that believes that there's really not anything magical about the human brain."
This statement is astonishing, particulary because it is coming from an AI "expert". Aside from the fact that the brain has allowed humans to create the most complex tools and technology the planet has ever seen, all the while not having to think about breathing or keeping our blood flowing, monitoring insane amounts of chemicals, repairing itself after injury, retaining intricate memories from decades ago, and even attempting to build machines to do the same thing, the brain is not that complicated or magical(note the searing sarcasm). Pearson has completely discredited the humain brain. Considering our research on the brain has only really been excelling in the past 30 years, and that we are nowhere near understanding its full complexity, Pearson instantly loses credibility IMO.
It has been long understood in AI that the only way to get a machine to truly behave like a human is through pure volume of information. People know that if they leave some salad out in the open and don't refrigerate it, it will spoil. To get a machine to come to the same conclusion, it needs to be programmed to know that specific fact, but still doesn't know anything about why or what a refrigerator does. It is only through brute force of millions of facts that we can get a machine to even come close to the range of knowledge that the human brain contains which is the basis for projects like Cyc. This leads to another point: that the human body itself is perhaps the most complicated organism on earth. Simply getting a robot to write its name in cursive is an enormous task to accomplish, as mechanical parts are no where near as efficient as human joints (although this is improving, they are still not sending the type of precise electrical signals to the brain like our bodies do). No, we don't need to be programmed to be able to catch a fly ball at a baseball game, we just judge where the ball is going. A robot must be told trajectory, wind speed, velocity of the ball, and all sorts of vector math just to be able to accomplish what we see to be a simple task. And on top of that, it can't learn how to do it again.
A fully functional "learning" AI computer is at minumum 50 years into the future, more likely 80-100 years, not 10 to 15. People in the 80s were saying the same thing back then as well, and when people wanted to know where there intelligent robotic math tutor was, they found out that it was a near impossible task to create such a robot. Either way, nothing will even come close to mirroring a human being completely in terms of complexity in our lifetimes.
I can finally play Wolf3d on 200x AA. Took long enough...
Can you find me now? Good.
Has anybody googled the author?
"Dunno if it's a joke, but it seems in earnest. Mark Twain must be spinning around somewhere."
The poster seems to ignore the fact that he himself is using shortened spelling in his description of the issue: "Dunno" as opposed to "Don't know".
What needs to be remembered is that Cell is a strictly in-order CPU, meaning that it cannot execute instuctions out of order. Therefore, hand-optimization can only get you so much. If you are waiting on an add, you are waiting on an add; not much can be done in that sense. It really comes down to compiler logistics and how the various tasks of branch prediction, locality and "cache" come into play/are optimized. Remember that Cell has no real cache but a virtual one. It ultimately comes down to the compiler, which I am sure is pretty beefy in this case.
crap...they found my pr0n.
Express PCB is a decent program, and it auto routes for you. Freeware.
The only component of the Cell that would remotely require an FPU is the PPC core which handles the physics and AI. However it does the job fine with out one due primarily to compiler optimization. The compiler is key for Cell because all of it's cores are "in-order", meaning that instructions must be executed in the order they are written and recieved. An out-of-order core requires much more complex circuitry along with surface area and power consumption, and in turn heat dissipation. With an optimized compiler, one can make an in-order core behave similar to an out-of-order one. Branch prediction also improves. Keep in mind that these cores are working with no cache and we can see why the compiler is crucial.
As a stabilizing mechanism you can use magnetorquers. They are inductive coils implanted into the side panels of the satellite. The navigation system detects rotation and fires off the coils when neccessary to counteract the magnetic field of the earth and slow down the rotation long enough to take a picture. It is quite an intelligent solution as it takes up no internal payload space.
I am a member of a team building a pico satellite at the University of California. The actual price for our entire project hovers around $50,000, of which 40k go to launch costs. This is the price tag for a Russian launch, I am sure the US would charge quite a bit more. The problem with these pico satellites is actually integrating a full fledged OS onto OTS parts. Data memory and program memory are in limited supply, along the lines of KB. We are using an Atmel CPU and are running AVR library from the http://www.avrfreaks.net/ website. The actual module we are writing is for the CPU scheduler. The interupt handler, bus protocols, etc are already written. Running this minimal amount of code is beneficial and doesnt overload the small amount of data memory we can utilize, so basically you can run less than a full on OS like linux and still get away with it.
"What imaging can you do for cheap -- what resolution, what wavelengths?"
The imaging comes down to the type of camera and it's power consumption. Image processing really loads the CPU and draws a lot of power, so compression algorithms are important. JPEG does the job fine, but is limited to around 320x res (in color) given our power constraints. The camera we are using is a CMOS imager, which works through the charging of capacitors relative to the intensity of incoming light. The wavelength of operation ultimately comes down to the FCC. They allocate you a bandwidth to use, most likely 2m or 440MHz. Other functions the satellite may serve is as a HAMsat, or a HAM radio repeater in space. You can tune you radio to the sat freq, and hit any place that the satellite covers with your signal. The preferred mode of operation is digital, so sat comm utilizes packet radio, which loosely resembles the TCP/IP stack. As far as a webserver goes, Im sure you could do it, but it would be limited to the length of one satellite path. Guess thats enough time for a 30 second pr0n clip...