Propeller supercomputing
cgracey
Posts: 14,134
I realized over the weekend that due to the PC <-> Propeller download scheme, it should be possible to connect up many Propeller chips with common P31 and P30 connections and load them all simultaneously.
If you drove all their·XIN inputs from a single 5MHz oscillator, all their PLLs could be phase-aligned at 80MHz. So, they could run in lock-step.
You could have the chips enumerate themselves, too, with dedicated in and out enumeration pins. After enumeration, everyone would know 'who' they are and could proceed to do something differentiated from the others.
By arranging the chips in an array and connecting them with·common vertical and horizontal I/O busses, you would have conduit in place to exchange data via rows and columns of, say, 8 bits each. Once you sync'd the array, they could deterministically communicate over these busses at a the rate of one transmit/receive per assembly instruction (20MHz bursts). A single COG in each Propeller could be responsible for this data exchange so that the individual 32KB of HUB RAM in each Propeller could be globally accessible with fairly short latency.
An 8x8 array of Propellers would have the following qualities:
512 COGs
10,240 MIPS of 32-bit computing
2MB of·HUB RAM
1MB of COG RAM
768 spare I/O pins
under 3.5 x 3.5" using QFN packages
~10 watts with every COG running
Imagine writing a single assembly-language program that, when downloaded, might run in 512 COGs simultaneously, but use its enumerated ID to focus on a limited piece of a larger data set. This could be the ray-tracer that someone was talking about on another thread.
Jeff and I were estimating that an 8x8x8 array of the next generation Propeller chips would have·8,192 COGs, over 1,000,000 MIPS, and 64MB of HUB RAM! That's getting a lot closer to supercomputing.
For the current Propeller, we plan to add some intelligent multi-chip loading capability to the Propeller Tool so that groups or single chips within a group·could be loaded with unique compiled top files. In the interim, maybe we'll make a dumb 8x8 array and see what we can get it to do... Andre?
·
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
Post Edited (Chip Gracey) : 4/19/2006 7:43:45 AM GMT
If you drove all their·XIN inputs from a single 5MHz oscillator, all their PLLs could be phase-aligned at 80MHz. So, they could run in lock-step.
You could have the chips enumerate themselves, too, with dedicated in and out enumeration pins. After enumeration, everyone would know 'who' they are and could proceed to do something differentiated from the others.
By arranging the chips in an array and connecting them with·common vertical and horizontal I/O busses, you would have conduit in place to exchange data via rows and columns of, say, 8 bits each. Once you sync'd the array, they could deterministically communicate over these busses at a the rate of one transmit/receive per assembly instruction (20MHz bursts). A single COG in each Propeller could be responsible for this data exchange so that the individual 32KB of HUB RAM in each Propeller could be globally accessible with fairly short latency.
An 8x8 array of Propellers would have the following qualities:
512 COGs
10,240 MIPS of 32-bit computing
2MB of·HUB RAM
1MB of COG RAM
768 spare I/O pins
under 3.5 x 3.5" using QFN packages
~10 watts with every COG running
Imagine writing a single assembly-language program that, when downloaded, might run in 512 COGs simultaneously, but use its enumerated ID to focus on a limited piece of a larger data set. This could be the ray-tracer that someone was talking about on another thread.
Jeff and I were estimating that an 8x8x8 array of the next generation Propeller chips would have·8,192 COGs, over 1,000,000 MIPS, and 64MB of HUB RAM! That's getting a lot closer to supercomputing.
For the current Propeller, we plan to add some intelligent multi-chip loading capability to the Propeller Tool so that groups or single chips within a group·could be loaded with unique compiled top files. In the interim, maybe we'll make a dumb 8x8 array and see what we can get it to do... Andre?
·
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
Post Edited (Chip Gracey) : 4/19/2006 7:43:45 AM GMT
Comments
Wow. So you could devise a PC/104 type type stacking system, put groups of chips on each board and "build" yourself an expandable "super-propeller". More power required - more boards stacked.....
Using the tool upgrades, you can then scale your apps out to X cogs on Y Propellers.....
James
I barely have the slightest idea what you are talking about, and even MY reaction is WOW!!!
I also realized last week, and never posted, something that YOU realized a long time ago. The propeller could easily be a text editor. You have fonts and keyboard, why not a text editor. It even occured to me that one could write their code with a propeller, and (eventually?) program a propeller with the same chip.
This might seem trivial to you, but it seemed a big deal to me at the time.
Kudos man! I'm glad that I am able to get in at the ground floor! I see this as one of my better design and life choices.
-Parsko
Cheers,
Simon
I warned you might get kissed on the street by random programmers.
Personally, I think I just owe you a beer.
Yours, TDP, ml, msl, & pfpp
·· A single Propeller could do all of that easily.·
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chris Savage
Parallax Tech Support
csavage@parallax.com
You should apply for a patent.
You have disclosed too much info to be able now to apply for a patent.
Does that mean that there is no feedback from the Propeller back to the PC to validate the programming?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Tracy Allen
www.emesystems.com
I keep getting deja vu of the IMSI S-100 computer and how it caught the imagination of so many.
Frankly, the hardest part is to try to keep up with all of you. I suppose I will just have to accept my snail pace.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
"When all think alike, no one is thinking very much.' - Walter Lippmann (1889-1974)
······································································ Warm regards,····· G. Herzog [noparse][[/noparse]·黃鶴 ]·in Taiwan
Post Edited (Kramer) : 4/19/2006 4:40:11 PM GMT
In the mid nineties a company I eventually owned had developed a multiprocessor super computer. In fact, for a brief period we could boast about having the word's largest supercomputer with 1,036 Motorola 6800 processors; 4 per each of 256 boards 18 inches square (plus some housekeeping processors) in a total of 18 cabinets, each about 30" x 30" by 4 ft high. It filled a whole lab, and cost investors over $40 million in development work. The 1 Gigabyte "hard drive" was a cabinet full of 1 bit streaming drives, 8 bits wide plus a parity drive. Each "bit" drive cost $10,000, so the cabinet was about $100 K.
The target applications were compute-intensive; 3D seismic computing, weather and pharmaceutical modeling etc. As I recall for some demonstrations, it did a wonderful job on Mandelbrodt displays.
The toughest part was the software to keep it all coherent... the processors were not hardware lock-stepped, it was all achieved by software. As I recall there were a number of patents involved.
As we were always playing "technology catch-up", we went on to build a single cabinet of 64 Motorola 68000 processors, but these machine well ahead of their time from an application perspective, and we could not find many buyers at $5 million a pop, so eventually the company became defunct. I still have a bunch of these in a warehouse; for what I don't know!
As you can see, parallel computing is one of my favourite subjects, and I applaud you for suggesting the subject with the Propeller in mind. It could be a lot of fun..... but also a lot of pitfalls if anyone expects to "go commercial" with them.
Cheers,
Peter (pjv)
Post Edited (pjv) : 4/19/2006 4:19:58 PM GMT
http://205.158.110.70/ubbcgi/ultimatebb.cgi?ubb=get_topic&f=20&t=001322
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
These are dark days for inventors in the US and they seem to be getting darker.·The patent system has been liberalized by a small group of judges, despite copious warnings from industry·to Congress, and despite the USPTO's wishes. Now, everybody is free to claim ANYTHING and EVERYTHING as their own exclusive property, no matter how obvious, stupid, or pre-existent it is, or simply because it poses an extortion opportunity directed at the few companies bold enough to actually MAKE and sell something. At the rate things are going, there will soon be no air left to breathe. There is a massive land-grab going on·RIGHT NOW·for every idea under the sun. The NTP vs RIM judgment was a tipping point, I think. I believe the only·people that are newly-positioned to benefit from the liberalized system are thieves and lawyers. The average person has no understanding of the dire ramifications of this liberalization, not to mention·how insane software and business-method patents are. It is a fact that you cannot write a program, you cannot design hardware, without 'infringing' patents. This is what the lawyers will NEVER admit. They love to pretend that someone deliberately 'stole' something, but this is almost always a lie. They also act like everyone who doesn't familiarize themselves with every patent they might infringe is criminally negligent. The fact is this: There isn't enough time in your life to do this and even if you did, you would only come to the conclusion that you CAN'T·DO ANYTHING. The entire system is gamed to their advantage, and by creating a civil war out of what should be·public-domain ideas (pull-up resistors, driving an LED with more current than the datasheet recommends, PWM'ing an LED to dim it, etc), the lawyers are going to win in every direction. If they are not stopped, they will choke to death what's left of US industry. I've come to the personal conclusion that our government either will not or cannot do anything about·it. The lawyer cabal is too powerful, too pervasive, and the government, itself, is made up mainly of·lawyers. One out of every fourteen people in Washington D.C. is a lawyer. The only way out is for·everybody to NOT play the game. That's not going to happen, though. You have to play defensively just to protect yourself, but there are ways of doing this that are not offensive in nature: SIR's (Statutory Invention Registrations) which look like patents, but go straight into the public domain, publishing papers, or deciding not to sell into the US market. Many Japanese companies have shunned the US market for decades·because it's just too treacherous to deal with the litigious environment here.
Here's an example of business-method patent mayhem: Netflix got a patent on the idea of having customers sign up online for what movies they want to see, and·in what order. When Blockbuster wanted to compete - LAWSUIT! PATENT INFRINGEMENT! This is absurd to most people, but this is the emerging future.
There may come a day when you hear from your neighbor over the fence that he's 'working on' some patents with his lawyer, and they're pretty confident that they are about to hit it big. It won't be that they're creating anything, they are just laying a trap to spring on someone. When that time comes, what's left of innovation·will flee the US just like manufacturing (and·quite a bit of engineering) already·has.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
Post Edited (Chip Gracey) : 4/19/2006 6:22:31 PM GMT
viva el parallax!
boo
I dont think the judges are specifically out to mangle the system, its a difference in experience. Judges are experts in the legal system, and it would be impossible for them to become technical experts on every field of invention that a particular patent dispute could come across thier bench. So something that is obvious to one of expertise in the field, is not obvious to them, simply because they are not experts. Its an unfortunate situation that technology has become so complex that only those well trained in the field can fully understand. The burden lies upon the examining core to provide clear, consistant, and fully illustated means for showing someone who isn't an expert why something is an obvious modification to an existing idea, and sometimes thats a very hard thing to do.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
1+1=10
Post Edited (Paul Baker) : 4/19/2006 6:06:08 PM GMT
I say there's nothing patent-worthy about the whole Propeller, but people can get and do wield patents on far-simpler ideas. This is not right.
Today everyone thinks they're staking their claim in the coming IP Gold Rush, as one guy gets his gun that could never really shoot,·and raises it to scare everybody else, some other person assumes 'ownership' of the ground from right under his feet. And on and on it goes. Only the lawyers profit from this. A screwed-up patent system and a public that buys into this nonsense (think NASDAQ) must be a dream-come-true for lawyers. One guy currently embroiled in a patent mess told me that he's heard of personal-injury attorneys migrating over to intellectual property. It's the future!
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
Post Edited (Chip Gracey) : 4/19/2006 6:23:51 PM GMT
BTW, that will cost you about $25USD. So why not put a copyright mark on it, along with the Propeller trademark.
You may not want to litigate, but it still is nice to own.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
"When all think alike, no one is thinking very much.' - Walter Lippmann (1889-1974)
······································································ Warm regards,····· G. Herzog [noparse][[/noparse]·黃鶴 ]·in Taiwan
http://invention.smithsonian.org/about/about_bio_jerome.aspx
Lemelson generally found that he had more success asserting his patent rights against foreign companies than against domestic companies. American corporations would go to great lengths and great legal expense to avoid paying royalties, perhaps to avoid setting a precedent in favor of other independent inventors. Sometimes the cost of their legal fees over the years of litigation would far exceed what they might have paid to license the patented product or process that they were fighting in court. The cereal company that printed masks on the backs of its boxes spent $150,000 to $200,000 in legal expenses to avoid paying $15,000 for licensing the idea from Lemelson.
My prediction: Sooner or later, you'll need a real OS.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Richard "Still thinks the PDP-8 was a great machine" Cook
The Japanese have been manipulated countless time by inventors squeezing them for royalties because thier cultural bias towards cooperation rather than adversarial. There is one guy in particular (Chip knows his brother) that got millions from employing such a tatic, that when he tried it against an American firm, they sucessfully nullified his patents. The Japanese firms filed a complaint with the office that they wanted thier royalties back, to which the answer was: you should have contested his patents·yourself. (actually it was closer to: we do not have the jurisdiction over the assesment of royalties garnered from patents)
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
1+1=10
Post Edited (Paul Baker) : 4/19/2006 7:01:30 PM GMT
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
1+1=10
One possible way to survive in such an environment is to stay under the radar and keep your assets to a minimum and/or out of reach. If you invent something disruptive enough that a large company takes notice and goes after you, they may be able to get an injunction to stop you from selling it; but not even the most egregious patent trolls will bother with a lawsuit if it means squeezing blood from a stone.
If you have soaring ambitions for world domination with your invention, it's inevitable: you'll get shot at. But if your goal is just to have fun and make a modest living, chances are much less that anyone's gonna mess with you.
-Phil
Yes, we at Parallax are quite content to have fun and make a modest (and honest) living. That's all we ask for in the world. The ability to do that is·what I feel is being stolen out from under us. We do not need or want·to lay exclusive claim to any ideas, and then get ugly with other people who want to do something similar. Copyrights and trademarks are quite satisfactory protections for us. We do not want to go public, either. If we can continue to work on fun stuff,·deliver to our·customers,·and feed our families, then we are a great success.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
Words spoken from a man of many years and expierences.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Just tossing my two bits worth into the bit bucket
KK
·
Okay. If I make an 8x8 array to experiment with, I will build one for you, too. I know there could be benefits to a proper OS, but I think that you would be surprised to find what a single COG per Propeller would accomplish in the way of inter-chip communication. By loading every Propeller with the same code and then enumerating them to give them uniqueness, you would have symbolic access to every variable in every Propeller by using the same symbolic names and addresses that exist in any local Propeller. If one chip needed to do something altogether different, you could query it's enumerated ID and branch accordingly.
I notice a recurring theme in multi-processor discussions where it is supposed that a program must be 'parallelized' to reap any benefit from multiple processors. This leads to talk about complicated compilers which infer parallelism from regular source code. I think this is totally missing the point! The great thing about having multiple processors is that you can assign them to completely different tasks, but have their variables related back 'live' to any other process.
We are trying to hire a new PCB layout person, and this 8x8 array is probably the first job for them to tackle. The schematic is almost nothing. It's just a lot of wiring.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
This is true, but as long as the chip has passed production testing and it's not connected to an EEPROM, it should always work. It's deterministic enough that it would make as much sense to verify the loading as it would to verify writes to an SRAM.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
You calculated that an 8X8 array of propellers would be capable of 10,240 MIPS.
It would also cost $1600 for the chips alone.
An AMD Athlon FX-57 dual core processor is capable of 12,000 MIPS of 32 bit instructions and costs $800.
It's also one chip instead of 64, which simplifies the routing a bit.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
I wonder if this wire is hot...
There is value, though, in having many concurrent and deterministic processors executing code together. I'd rather be tasked with bringing up 64 Propellers than a single Athlon, myself, and I know the development tool would be much simpler. The Athlon is a system chip, with caches, predictive branching, and multiple execution units. It would be tough to write assembly language programs that would have deterministic behavior all the way out to I/O pins. The Propellers would be simplistic, but at least regular in their behavior. BTW, there's a price break at 25 chips.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Chip Gracey
Parallax, Inc.
I would rather work with Parallax microcontrollers then Intel CPU's or I wouldn't be here, but at some point reality comes into play.
The Propeller will be huge in robotics, and has a fighting chance in the handheld market.
I hope to make fun toys for myself and my children.
Keep up the good work!
-Alexander
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
I wonder if this wire is hot...
Could we keep in mind that one does not necessarily HAVE to use an array of 8x8 Props. Am I correct in saying that all the same theories apply to a 1x2 array of Props?
Chip, would you please explain the theories of enumeration, parallelization to me, being a laymen (I implied this question in my first post on this topic, but should have asked outright)? I'm mostly curious about the "10 words or less" version, or should I say, what does this mean to me? I personally really like the idea of linking two Props for data aquisition in control systems (though I have no clue what even one Prop will do yet!)
-Parsko