Yes, whatever we learn, it becomes a slate for future experiences.
Anyone who have experience in programming the 3D GPU would understand what that means.
Consider GPU booting up in VESA mode, it have no concept of 3D, only 2D. When we feed it some meaningful math vectors, the GPU would understand what 3D is afterward, in the same way the babies learn about the 3D stereoscopic vision. it's the objects that is interacted with giving us the basic understanding.
Okay, here's a surprising fact. What's the supercomputer in the nature, only it's not Silicon-based, but Carbon-based? Our mind. It's the computing power our brain possess, learning what's around us, and calculating what we want to do with objects, or just to avoid them.
Just conjecturing the basics. If anyone think I am wrong, go ahead and flame me or any way you like.
Speaking of minds, they're working on a supercomputer with the same computing power (360PFLOPS), mostly for AI research which will benefit the future robots.
I mentioned it when I was being interviewed for a place at university to study psychology. I later found that it was a favourite study with one of the interview committee members.
True. Being passive observer is not going to you any fruitful information.
Okay, shall I comment on the FPGA? Suppose you develop a dumb, simple CPU in the FPGA and put it in a robot that you're going to subject to the greatest challenge of the time, like trying to avoid the walls? And, you put in the powerful, multicore ARM CPU in the robot of the same construction - let's go to the active experiments.
As those robot get subject to such abuse of hitting the wall, they starts to understand why: They feel the shock and see that wall, and now they go out in random distance, and those robot starts to confront the door that they don't remember seeing the door.
Okay, which one will avoid the door, and which one decides to hit the door?
... If you answered FPGA one avoided the door and the ARM SoC one hit the wall. You are right. Why? The robot with FPGA for a brain already have the "neurons" of that experience and will of course remember that any objects, no matter what, shall be avoided at all cost. The ARM-powered robot, sure it have a powerful CPU, have no real "hard" experiences, so its decision would become REAL limited. Ask yourself now, which one is an active observer or a passive observer?
Perhaps you are familiar with this cat experiment as well:
Take a new born kitten and have it live for some weeks in cylindrical container. There is light in the container but the inside walls are painted with black and white stripes running horizontally. The idea being that for those weeks the kitten sees nothing much that is not horizontal.
Result: Eventually you start to play with the kitten by waving a black stick in front of its eyes. It is found that the kitten will reach out and grab or other wise respond to a horizontal stick. But if the stick is held vertically there is no response. He just "looks trough it", can't see it at all.
Conclusion: Even a simple thing like recognizing "vertical" is learned in the developing brain, it is not pre wired. If it is not learned at a very young age it may never be learned later.
I've tried to find links to this experiment, which I saw on TV, perhaps the Open University in the late seventies, on the net with no luck so far.
Now we are way off topic for a Propeller Forum, unless someone is building a kitten simulator, perhaps Humanoido.
My wife is a cat lover. I would be in trouble if I so much as booted a cat. That kind of limits my experiments to booting and torturing Propellers nowadays:)
Perhaps you are familiar with this cat experiment as well:
Take a new born kitten and have it live for some weeks in cylindrical container. There is light in the container but the inside walls are painted with black and white stripes running horizontally. The idea being that for those weeks the kitten sees nothing much that is not horizontal.
Result: Eventually you start to play with the kitten by waving a black stick in front of its eyes. It is found that the kitten will reach out and grab or other wise respond to a horizontal stick. But if the stick is held vertically there is no response. He just "looks trough it", can't see it at all.
Conclusion: Even a simple thing like recognizing "vertical" is learned in the developing brain, it is not pre wired. If it is not learned at a very young age it may never be learned later.
I've tried to find links to this experiment, which I saw on TV, perhaps the Open University in the late seventies, on the net with no luck so far.
Now we are way off topic for a Propeller Forum, unless someone is building a kitten simulator, perhaps Humanoido.
I don't think YOU are much of topic.
It only confirm that NO mater how nice sensor's else computer You use.
If You not program it correctly You will not have results that You wanted.
I don't mind if anyone are off the topics. Anyways, it's an interesting ancodote evidence.
Alright, Humanoid... How far have you gotten with your UltraSpark 40? Just curious.
And, do you think PCIe 1x bus is okay regarding the spinal cord system in the supercomputer?
I don't mind if anyone are off the topics. Anyways, it's an interesting ancodote evidence.
Alright, Humanoid... How far have you gotten with your UltraSpark 40? Just curious. And, do you think PCIe 1x bus is okay regarding the spinal cord system in the supercomputer?
Dr. Mario, it is your thread and if you don't mind off topic then so be it. But if it gets too out of hand, I'm sure a moderator will step in. I did have some ideas for you and your idea of core farming is interesting. How do you think about mixing props with another vendor, another software, etc.?
I looked up the card, you can get one for $60 with 320 processors - there are others with more processors too. Connecting to props, well I think that's why you mentioned PCI Bus... My only thought is about programming the board and how much of a learning curve it will need. This issue was addressed in some posts a while back and there was a significant learning curve and new pay software required which upped the investment substantially. You're looking at far more than $60.
I think the prop will be ok for software. How many props are you thinking about using? The PCI bus (Peripheral Component Interconnect) was developed by Intel for the PC. What is your primary purpose for your supercomputer? Basically I started with a serial bus interface that connects all the processors through a Daisy Chain with the BASIC Stamp Supercomputer, for understanding.
There are different node arrangements, each with advantages and disadvantages - google 'network' to see some examples. Each system will require different programming. I like to make each processor self determinant, so in a future project, phasing in some AI will be doable.
Some of the faster systems require an inordinate number of wires. I didn't have enough time to implement some of the interfaces that require exponential numbers of connecting wires. For example, a true hypercomputer will have computers at every dimensional vertice leading into multiple dimensions. In one wire, every time you change one wire, you have to change 40 wires. If you have hyperspace, and 2,000 wires to change, you get the idea.
I also use color coding. For example, if one wire falls off then I just follow the color coding down from the top of the SkyScraper. It has worked in the past so now I exclusively use pre-made color coded wire cables with end pins.
To start, I recommend bit banging along one wire - the speed of this has grown recently. Review the Objects in the OBEX. There's some very good stuff there. You can also configure your props in different ways. I use a twinning interface to share boards, peripherals and to save cost and compress real estate. EEPROMs are especially cool. Multiple props can share EEPROMs and they can piggy to save space. Just use the index of the EEPROM to access it.
Another good idea is breadboarding. You can quickly setup and change wiring. I have not had any problems with breadboards and overclocking with a 6.25 MHz crystal. You may choose from several loading techniques which is directly related to the circuit and number of props. I hope this helps get you started.
Humanoid, Oops, sorry, that was a little quip too far, my apologies. I don't mean any harm or offence. My bad sense of humour gets me into trouble all the time.
Heater, apology accepted. It is really my fault for being too serious. A few times my humor gets me into trouble too. I'm not sure why the humor is lost in text. Even worse, Chinese friends don't always get my humor. Even if I take great effort to explain the joke, they don't get it. I even researched this and learned it's because there is a vast difference in culture and language. While we all have a great sense of humor, it is expressed in different ways. I must remind myself that all of us, though from the same book, are all on a different page.
I don't know how I would use 40 props, but I became interested in the parallel processing propeller when I decided to (eventually) build a mark one tricorder, parallax is the perfect company for this project as they also sell sensors, but the cost will be quite high so it won't happen soon.
I expect to use 4 propellers in a handheld TNG style case.
If I wanted to build a TOS style tricorder, there could be room for 40 props but I'm not sure what I would do with over 1200 i/o pins.
Even when the prop 2 is available I would still use prop 1's, as cogs are more important than double speed processing or i/o pins.
Well, yeah... Humanoido, I'm thinking about using PCI-express bus for spinal cord of the supercomputer (to reduce som eof wiring hassles and to be able to transfer daat 2 - 4 GB/s (Yes, Gigabytes per seconds - PCI express's that fast!) - and to be able to easily swap out due to the nature of PCI express bus itself). And, PCI express bus is handled primarily by the branch microcontroller (which control the Propeller I / II chips and download the buffer data onto HUB RAM on selected Prop), on each of the boards. PCI-e is to be converted into PCI, something PIC32 can understand, and without breaking MIPS vector interrupts linked to the propeller chips.
Well, yeah... Humanoido, I'm thinking about using PCI-express bus for spinal cord of the supercomputer (to reduce som eof wiring hassles and to be able to transfer daat 2 - 4 GB/s (Yes, Gigabytes per seconds - PCI express's that fast!) - and to be able to easily swap out due to the nature of PCI express bus itself). And, PCI express bus is handled primarily by the branch microcontroller (which control the Propeller I / II chips and download the buffer data onto HUB RAM on selected Prop), on each of the boards. PCI-e is to be converted into PCI, something PIC32 can understand, and without breaking MIPS vector interrupts linked to the propeller chips.
Hi Dr. Mario, sounds like you have a good Hybrid design already sketched. How many FPG boards are you now thinking about? There's a computer skyscraper (business building) that is filled with boards from top to bottom. Maybe there's a few hundred stores. They have all kinds of these boards on sale. If you bring a local friend, they will talk the price down from 200 to 50 or from 50 to 10. But that's all you get, a board, and then you must know what to do with it.
Yeah. I'm thinking about like 32 boards (roughly) - to finalize the software, but I will settle on maybe one or few boards, even plugging it in my own workstation to finalize the management software as I'm writing the uCLinux driver in the uCLinux based firmware on PowerPC seed computer. After things gets nice (I will have to buy my own server since I am interested in Anime cel graphics) - and money well held over, I will do maybe 1,024 board on U-rack. The only limit is, (sigh) the wall outlet currents being sucked up by multiple ATX power supplies...
Yeah. I'm thinking about like 32 boards (roughly) - to finalize the software, but I will settle on maybe one or few boards, even plugging it in my own workstation to finalize the management software as I'm writing the uCLinux driver in the uCLinux based firmware on PowerPC seed computer. After things gets nice (I will have to buy my own server since I am interested in Anime cel graphics) - and money well held over, I will do maybe 1,024 board on U-rack. The only limit is, (sigh) the wall outlet currents being sucked up by multiple ATX power supplies...
Dr. Mario, what amp service do you have? I have a direct 220 volt service feeding my rooms and the trick to pulling more amperage power for a supercomputer is to use the kitchen outlets which can deliver more amps, and then a second tap from the other rooms. Just make sure it's not out of phase or you know what will happen. :skull: That's a healthy number of boards planned and sounds like you will have a very robust supercomputer. How soon can you start? I know what you're saying about the current draw of so many boards. I have to go around cutting out those power hungry components to save every bit of power on the aggregate. In the BSS, I had to cut power and cut out the voltage dog. In the US40, it was a matter of bypassing the PB's power circuits. In the next PSC it will be a massive distributed Chinese PS and absolutely no voltage dogs using custom designed boards. I have already wired these out, at least in my mind. But... who is responsible for the light flickering in the building? .... er... why is everyone looking at me?
Of course. But it may not happen within few months... 1. I may probably move soon. 2. Not enough cash after paying what I owe. (Bills are a real dog... T____T Boo hoo...) 3. Something I need to do (doing someone a favor).
BTW, I may move to Japan soon. (Yes, you still can use USA computer ATX power supply there in Japan - just leave it at 115V at the voltage selector switch - at one expense, it will eat even more Amps at 100V than with 115V. This is one issue I have - trying to do so without wrecking the power bills - either that or I can do that at work...) What about 50 or 60Hz? Nah, don't worry about it, 200 - 400 Volts line filter capacitors are very good at masking the line frequency (after recifited - it would look like 100Hz or 120Hz pulses, hence the reason for low ESR high voltage capacitors).
And, I would still have ambition for that supercomputer even after i move there - it's useful to have lot of COGs to do fine-grained integer AI. Oh, and if you want really clean clock for your Propeller, may I recommend the MEMS oscillator, SiT8002 from SiTime? They're real cheap and robust (way better than crystal, owing to MEMS structure). And, yes, I have few 5 MHz SiT8002s from them, I will be using with Propeller - the first one I am going to use in is the Diode-Pumped Solid-State laser controller board (yes, with Propeller).
They make those power supplies that work from 100 to 220 volts, handy if you travel from country to country. I need to recommend those to Parallax. They have lots of international customers. I used a common inverter to build up an external oscillator clock. I doubt we could find the SiTime's MEMS oscillator SiT8002 over here. Do you have any smaller circuits with props running it for a test? How many can you chain to it?
Yes. I may move there. And, of course, Mouser's also available to Japanese customers too. http://jp.mouser.com - Japanese website for Mouser Electronics, Inc. (English by default - and yes, they still sell Propeller there, and my favorite oscillators, SiTime SiT8003 too.)
Don't be shocked if it looks like it's expensive, it's not - it's around $7.99, in USA Dollar. (Of course, Customer Electronics Taxes do apply here - around 10% or so. Not really painful compared to federal taxes. Federal taxes are huge there, compared to here - you can ask any Japanese peoples who have been paying that taxes. At least it won't take big bite out of my project budget... Thankfully.) After the taxes, it would be priced at $9.88, not so bad.
And, if no such luck with finding 5 MHz SiT8003, you can try 10MHz and set the PLL multiplier at 8x for 80MHz.
BTW, another word of warning, be careful around Yen money - they're really strange, they always trail around with stock market... At least you won't have to shell out way too much for the part in certain stores, like Mouser.
I have a friend who is doing fly work at UPenn, where they quantify the flies activity level and we have been kicking around better ideas for software or hardware to measure the movement of the flies.
Well, that's the limit of current system, if that gotta to do with Propeller. It's the memory and core frequency that's the limiting factor.
Of course, if you want to track the flies on-the-fly, you may want something more powerful like AMD Radeon HD 4000 or faster - just don't want to discourage anyone, it's just that the software and hardware are the limiting factor in tracking anything moving and in 3D.
Of course, if you get the supercomputer board together and give the Propeller II much larger RAM (32MB SRAM for 128KB HUB - similar to TLB cache SRAM...) and everything goes along smoothly without coughing a hairball, then yes, you can try that. However, you will want to only start the SPIN once, and afterward in C++ (some self-modifiying codes are necessary...) - like I said, the software are also the limiting factor (but like they says, "The programming rules that cannot be understood can be axed down.")
And, the supercomputer names I have been pondering for a while... Shall be... Electric Demon. If you think it's a bit distasteful, I shall change it, anyways.
Here's a diagram of the Propeller II cluster board.
And, the supercomputer names I have been pondering for a while... Shall be... Electric Demon. If you think it's a bit distasteful, I shall change it, anyways.
Here's a diagram of the Propeller II cluster board.
That looks quite cool. Why wait for Propeller II though? Build a version with Propeller I as a prototype and sell me a board! :-)
Why am I waiting for Propeller II? There are obivious benefits with SOI (Silicon-On-Insulator - Silicon Dioxide, in other word, glass layer between transistor substrate and compressed / strained Silicon gate), such as higher core frequency - up to 160MHz with this MCU, and lower current consumptions (VERY important here!!! -W- I am not trying to beat the world record for the world's hungriest supercomputer in term of wattage - COMPARED like Las Vegas, The Sin City!!!)
And, there are better ALU pipelines in Propeller II, as I want to extract out every MIPS out of integer computing.
Propeller I? Maybe, as well as trying out the theory, but there are something that still won't add up with Propeller II, as they chose to go with clean slate, with the opcode LUTs (the "scratch" ROM which contains special look-up data for the ALU instructions and microcodes). Here, I would doubt it will even execute the original ROM data without having to rewrite the OS or everything, even if it will still work, the OS will suspect that it still doesn't add up so it will spew out "ERROR = INCORRECT CPU OPCODE" flags. It would be a bit annoying...
Selling a board? I will have to think about it, since building a supercomputer won't be cheap.
(Not trying to discourage anyone... -___-" )
The problem is the software, the hardware is there. The question is which problem are you trying to solve ?
I have one:
Think of the Borg collective (!) as a group of nodes interconnected, they share information, they can work independetly on tasks, they can suddently go off-line. That is a task for your machine.
Dr. Mario, get yourself a XMP-64 from ... just google it. It is ready to run, has 64 cores and 512 threads and needs some 30 watts. Only 1500 us $. No need to wait for PropII
Good point, however even if a source can sell the board at $1500, that's still a hefty $23.44 per chip.
You could use eight prop I chips for less than $8 ea. If you're on a hobby budget, 64 prop cores are less than $64.
Comments
Anyone who have experience in programming the 3D GPU would understand what that means.
Consider GPU booting up in VESA mode, it have no concept of 3D, only 2D. When we feed it some meaningful math vectors, the GPU would understand what 3D is afterward, in the same way the babies learn about the 3D stereoscopic vision. it's the objects that is interacted with giving us the basic understanding.
Okay, here's a surprising fact. What's the supercomputer in the nature, only it's not Silicon-based, but Carbon-based? Our mind. It's the computing power our brain possess, learning what's around us, and calculating what we want to do with objects, or just to avoid them.
Just conjecturing the basics. If anyone think I am wrong, go ahead and flame me or any way you like.
Speaking of minds, they're working on a supercomputer with the same computing power (360PFLOPS), mostly for AI research which will benefit the future robots.
http://www.goes.com/hovancik/exp/kit_caro.htm
I mentioned it when I was being interviewed for a place at university to study psychology. I later found that it was a favourite study with one of the interview committee members.
Okay, shall I comment on the FPGA? Suppose you develop a dumb, simple CPU in the FPGA and put it in a robot that you're going to subject to the greatest challenge of the time, like trying to avoid the walls? And, you put in the powerful, multicore ARM CPU in the robot of the same construction - let's go to the active experiments.
As those robot get subject to such abuse of hitting the wall, they starts to understand why: They feel the shock and see that wall, and now they go out in random distance, and those robot starts to confront the door that they don't remember seeing the door.
Okay, which one will avoid the door, and which one decides to hit the door?
... If you answered FPGA one avoided the door and the ARM SoC one hit the wall. You are right. Why? The robot with FPGA for a brain already have the "neurons" of that experience and will of course remember that any objects, no matter what, shall be avoided at all cost. The ARM-powered robot, sure it have a powerful CPU, have no real "hard" experiences, so its decision would become REAL limited. Ask yourself now, which one is an active observer or a passive observer?
Perhaps you are familiar with this cat experiment as well:
Take a new born kitten and have it live for some weeks in cylindrical container. There is light in the container but the inside walls are painted with black and white stripes running horizontally. The idea being that for those weeks the kitten sees nothing much that is not horizontal.
Result: Eventually you start to play with the kitten by waving a black stick in front of its eyes. It is found that the kitten will reach out and grab or other wise respond to a horizontal stick. But if the stick is held vertically there is no response. He just "looks trough it", can't see it at all.
Conclusion: Even a simple thing like recognizing "vertical" is learned in the developing brain, it is not pre wired. If it is not learned at a very young age it may never be learned later.
I've tried to find links to this experiment, which I saw on TV, perhaps the Open University in the late seventies, on the net with no luck so far.
Now we are way off topic for a Propeller Forum, unless someone is building a kitten simulator, perhaps Humanoido.
I don't think YOU are much of topic.
It only confirm that NO mater how nice sensor's else computer You use.
If You not program it correctly You will not have results that You wanted.
Alright, Humanoid... How far have you gotten with your UltraSpark 40? Just curious.
And, do you think PCIe 1x bus is okay regarding the spinal cord system in the supercomputer?
Humanoido
Oops, sorry, that was a little quip too far, my apologies. I don't mean any harm or offence.
My bad sense of humour gets me into trouble all the time.
I looked up the card, you can get one for $60 with 320 processors - there are others with more processors too. Connecting to props, well I think that's why you mentioned PCI Bus... My only thought is about programming the board and how much of a learning curve it will need. This issue was addressed in some posts a while back and there was a significant learning curve and new pay software required which upped the investment substantially. You're looking at far more than $60.
I think the prop will be ok for software. How many props are you thinking about using? The PCI bus (Peripheral Component Interconnect) was developed by Intel for the PC. What is your primary purpose for your supercomputer? Basically I started with a serial bus interface that connects all the processors through a Daisy Chain with the BASIC Stamp Supercomputer, for understanding.
There are different node arrangements, each with advantages and disadvantages - google 'network' to see some examples. Each system will require different programming. I like to make each processor self determinant, so in a future project, phasing in some AI will be doable.
Some of the faster systems require an inordinate number of wires. I didn't have enough time to implement some of the interfaces that require exponential numbers of connecting wires. For example, a true hypercomputer will have computers at every dimensional vertice leading into multiple dimensions. In one wire, every time you change one wire, you have to change 40 wires. If you have hyperspace, and 2,000 wires to change, you get the idea.
I also use color coding. For example, if one wire falls off then I just follow the color coding down from the top of the SkyScraper. It has worked in the past so now I exclusively use pre-made color coded wire cables with end pins.
To start, I recommend bit banging along one wire - the speed of this has grown recently. Review the Objects in the OBEX. There's some very good stuff there. You can also configure your props in different ways. I use a twinning interface to share boards, peripherals and to save cost and compress real estate. EEPROMs are especially cool. Multiple props can share EEPROMs and they can piggy to save space. Just use the index of the EEPROM to access it.
Another good idea is breadboarding. You can quickly setup and change wiring. I have not had any problems with breadboards and overclocking with a 6.25 MHz crystal. You may choose from several loading techniques which is directly related to the circuit and number of props. I hope this helps get you started.
Humanoido
Humanoido
I expect to use 4 propellers in a handheld TNG style case.
If I wanted to build a TOS style tricorder, there could be room for 40 props but I'm not sure what I would do with over 1200 i/o pins.
Even when the prop 2 is available I would still use prop 1's, as cogs are more important than double speed processing or i/o pins.
Humanoido
Humanoido
BTW, I may move to Japan soon. (Yes, you still can use USA computer ATX power supply there in Japan - just leave it at 115V at the voltage selector switch - at one expense, it will eat even more Amps at 100V than with 115V. This is one issue I have - trying to do so without wrecking the power bills - either that or I can do that at work...) What about 50 or 60Hz? Nah, don't worry about it, 200 - 400 Volts line filter capacitors are very good at masking the line frequency (after recifited - it would look like 100Hz or 120Hz pulses, hence the reason for low ESR high voltage capacitors).
And, I would still have ambition for that supercomputer even after i move there - it's useful to have lot of COGs to do fine-grained integer AI. Oh, and if you want really clean clock for your Propeller, may I recommend the MEMS oscillator, SiT8002 from SiTime? They're real cheap and robust (way better than crystal, owing to MEMS structure). And, yes, I have few 5 MHz SiT8002s from them, I will be using with Propeller - the first one I am going to use in is the Diode-Pumped Solid-State laser controller board (yes, with Propeller).
Humanoido
Humanoido
http://jp.mouser.com - Japanese website for Mouser Electronics, Inc. (English by default - and yes, they still sell Propeller there, and my favorite oscillators, SiTime SiT8003 too.)
And as a proof: http://jp.mouser.com/ProductDetail/Parallax/P8X32A-M44/?qs=sGAEpiMZZMvu0Nwh4cA1wX5jwmbK8bxp6N4lodrTqHY%3d
Don't be shocked if it looks like it's expensive, it's not - it's around $7.99, in USA Dollar. (Of course, Customer Electronics Taxes do apply here - around 10% or so. Not really painful compared to federal taxes. Federal taxes are huge there, compared to here - you can ask any Japanese peoples who have been paying that taxes. At least it won't take big bite out of my project budget... Thankfully.) After the taxes, it would be priced at $9.88, not so bad.
And, if no such luck with finding 5 MHz SiT8003, you can try 10MHz and set the PLL multiplier at 8x for 80MHz.
BTW, another word of warning, be careful around Yen money - they're really strange, they always trail around with stock market... At least you won't have to shell out way too much for the part in certain stores, like Mouser.
I have a friend who is doing fly work at UPenn, where they quantify the flies activity level and we have been kicking around better ideas for software or hardware to measure the movement of the flies.
http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5439118
http://domino.research.ibm.com/comm/research_people.nsf/pages/gerasimos.potamianos.pubs.html/$FILE/WMVC07_3D_TRACKING.pdf
That system I mentioned couldn't do tracking in 3D, the objects had to be stationary.
Of course, if you want to track the flies on-the-fly, you may want something more powerful like AMD Radeon HD 4000 or faster - just don't want to discourage anyone, it's just that the software and hardware are the limiting factor in tracking anything moving and in 3D.
Of course, if you get the supercomputer board together and give the Propeller II much larger RAM (32MB SRAM for 128KB HUB - similar to TLB cache SRAM...) and everything goes along smoothly without coughing a hairball, then yes, you can try that. However, you will want to only start the SPIN once, and afterward in C++ (some self-modifiying codes are necessary...) - like I said, the software are also the limiting factor (but like they says, "The programming rules that cannot be understood can be axed down.")
Here's a diagram of the Propeller II cluster board.
Why am I waiting for Propeller II? There are obivious benefits with SOI (Silicon-On-Insulator - Silicon Dioxide, in other word, glass layer between transistor substrate and compressed / strained Silicon gate), such as higher core frequency - up to 160MHz with this MCU, and lower current consumptions (VERY important here!!! -W- I am not trying to beat the world record for the world's hungriest supercomputer in term of wattage - COMPARED like Las Vegas, The Sin City!!!)
And, there are better ALU pipelines in Propeller II, as I want to extract out every MIPS out of integer computing.
Propeller I? Maybe, as well as trying out the theory, but there are something that still won't add up with Propeller II, as they chose to go with clean slate, with the opcode LUTs (the "scratch" ROM which contains special look-up data for the ALU instructions and microcodes). Here, I would doubt it will even execute the original ROM data without having to rewrite the OS or everything, even if it will still work, the OS will suspect that it still doesn't add up so it will spew out "ERROR = INCORRECT CPU OPCODE" flags. It would be a bit annoying...
Selling a board? I will have to think about it, since building a supercomputer won't be cheap.
(Not trying to discourage anyone... -___-" )
get yourself a XMP-64 from ... just google it. It is ready to run, has 64 cores and 512 threads and needs some 30 watts. Only 1500 us $.
No need to wait for PropII
I have one:
Think of the Borg collective (!) as a group of nodes interconnected, they share information, they can work independetly on tasks, they can suddently go off-line. That is a task for your machine.
You could use eight prop I chips for less than $8 ea. If you're on a hobby budget, 64 prop cores are less than $64.