Supercomputing Definition - Update It Here
Humanoido
Posts: 5,770
Let's Make a New Supercomputer (Definition)
Take one prop chip for example. Then compare. Granted, the human brain is a biological supercomputing machine. I think we've proven that no matter what the element, carbon based, biological, mechanical relays, molecular, even light constructed, supercomputers can be made from these elements. Equally important is that we have different types of supercomputers based on the instructions in software, such as FP and INT. It does not end here.
For a moment, let's take our crusty old minds, break out of our lifetime shells, and surpass the old, outdated, too rigid and too confined definition that is based upon speed only. Let speed remain as one element that can make a computer into a supercomputer but let's introduce additional elements that can also contribute to making a supercomputer.
There are definitely super computing machines that can be based on factors other than just speed. For example, a computing machine with 100,000,000 cores can be considered a supercomputer compared to one core. By contrast it is super. Maybe it has a slower clock than a one core computer. But the sheer vastness of cores and the potential new prospects it introduces make it super compared to one core.
It fits the very definition of super.
http://www.thefreedictionary.com/super-
Super is superior in size, quality, number, or degree, exceeding a norm, excessive in degree or intensity, containing a specified ingredient in an unusually high proportion, and more inclusive than a specified category.
So now we can rate a machine in terms of its capacity, computing efficienty (even if we are talking about toggling a single port), the number of cores, the ability in which it exceeds from one computer by the machine, that it has something in high proportion (ports, gates, cores, counters, clocks, size, cost, type of contruction - example is molecular or light, and more multipurposed.)
Supercomputing can be relative. Set the historical Cray-1 in your garage. Is it a supercomputer? Yes. It is as fast as the fastest supercomputer today? In fact, nearly all those supercomputers of today cannot and are not the fastest - there is only one and perhaps for only a fleeting moment.
So from our contemplative thought and consideration perspective that speed alone is NOT the single-most criteria for making a supercomputing machine, what additional new positive enlightening criteria would you introduce to qualify your computing machine as a supercomputer? This is the focus and question of this post.
Humanoido
Take one prop chip for example. Then compare. Granted, the human brain is a biological supercomputing machine. I think we've proven that no matter what the element, carbon based, biological, mechanical relays, molecular, even light constructed, supercomputers can be made from these elements. Equally important is that we have different types of supercomputers based on the instructions in software, such as FP and INT. It does not end here.
For a moment, let's take our crusty old minds, break out of our lifetime shells, and surpass the old, outdated, too rigid and too confined definition that is based upon speed only. Let speed remain as one element that can make a computer into a supercomputer but let's introduce additional elements that can also contribute to making a supercomputer.
There are definitely super computing machines that can be based on factors other than just speed. For example, a computing machine with 100,000,000 cores can be considered a supercomputer compared to one core. By contrast it is super. Maybe it has a slower clock than a one core computer. But the sheer vastness of cores and the potential new prospects it introduces make it super compared to one core.
It fits the very definition of super.
http://www.thefreedictionary.com/super-
Super is superior in size, quality, number, or degree, exceeding a norm, excessive in degree or intensity, containing a specified ingredient in an unusually high proportion, and more inclusive than a specified category.
So now we can rate a machine in terms of its capacity, computing efficienty (even if we are talking about toggling a single port), the number of cores, the ability in which it exceeds from one computer by the machine, that it has something in high proportion (ports, gates, cores, counters, clocks, size, cost, type of contruction - example is molecular or light, and more multipurposed.)
Supercomputing can be relative. Set the historical Cray-1 in your garage. Is it a supercomputer? Yes. It is as fast as the fastest supercomputer today? In fact, nearly all those supercomputers of today cannot and are not the fastest - there is only one and perhaps for only a fleeting moment.
So from our contemplative thought and consideration perspective that speed alone is NOT the single-most criteria for making a supercomputing machine, what additional new positive enlightening criteria would you introduce to qualify your computing machine as a supercomputer? This is the focus and question of this post.
Humanoido
Comments
Humanoido
We have this word "supercomputer" there are two parts to this:
"super" meaning, asy you say, "superior in size, quality, number, or degree, exceeding a norm, excessive in degree or intensity, containing a specified ingredient in an unusually high proportion, and more inclusive than a specified category."
and "computer" meaning "one that computes" Don't forget the original use of the word "computer" referred to humans employed to perform repetitive calculations.
So the obvious key point here is "super computing ability" not "super size" or "super weight" or "super power efficient". Or "super at saving your life" as in your bloodstream swimming robot example.
No, all that matters is "excessive in degree" computing.
Clearly that is not the Prop or any reasonable cluster of Props by comparison to any "excessive in degree" computer since the 1960s.
On the other hand:
I presented a two lines of PASM that will toggle a LED on the Prop at 10MHz. I would like someone to show me a LED on this machine http://en.wikipedia.org/wiki/Jaguar_%28computer%29 that can be toggled at 10MHz under software control. Fail I think. Therefore for that application a single Prop is a supercomputer already:)
P.S. Beau has a two line PASM loop that will toggle a LED at 20MHz can anyone think what it is? I won't present it here it's Beau's invention I'll wait for him to show it.
Heater: The post question is about pointing out positive things not speed related.
Heater said, No, all that matters is "excessive in degree" computing. Clearly that is not the Prop or any reasonable cluster of Props by comparison to any "excessive in degree" computer since the 1960s.
Heater: Don't be hasty. We have already shown that a small reasonable cluster of Propeller chips is equal to numerous machines that were indeed extreme supercomputers well into 75 and undoubtedly beyond.
http://forums.parallax.com/showthread.php?t=123828
Add a few more chips to this machine and it's into 80s, then 90s. Add a lot more and it jumps ten more years. At least one man pondered a supercomputer farm, with potential power to compete with much more powerful supercomputers (after the year 2,000) yet you discouraged him for some reason, and me. Heater, do you have a bone to pick?
http://forums.parallax.com/showthread.php?t=125614
BTW, I think you want to post remarkable speed achievements at this link:
http://forums.parallax.com/showthread.php?t=125627
Humanoido
No I don't have a bone to pick, I just like a good debate. If we can disagree with each other whilst putting up good arguments, evidence and points of view then we both stand to learn something.
Still my point of view is that a "supercomputer" is not defined by it's size, number of CPU's, efficiency etc but just by it's ability to compute lots of stuff fast. We should not bend the term to encompass these other things.
Now that rather presupposes that we have defined what we mean by "compute". As I said, I don't believe that the Cray Jaguar, currently the fastest computer in the world, can actually count from 0 to 4 Billion in software and display the results on 32 LEDS faster than the Propeller can. In that computation the Prop wins.
See, I can agree with both sides of this argument at the same time:)
Beau has this cunning two PASM instruction loop that can do that counting at 20Mega Hertz. If he does not post it somewhere soon I won't be able to resist doing so.
On the other hand, if it's qualified as in "Propeller Super Computer", then it's appropriate. For a given class of computer, there are "super" ones, right?
In micro-land, the concurrency in the Propeller, is "super", but I don't know "super" what. Maybe that's something to be worked on, rather than a bending of the general term "super computer". There are lots of examples of this kind.
Most computers these days are general purpose. Some have specific design elements that nail a task. When that's done in silicon, those abilities remain relevant, even though the machine as a whole isn't.
A SGI O2, has this attribute with it's shared memory graphics, and video capture hardware. That thing --even a slow one, can put video on a surface, ripple it, spin it around, and such without hassle. It's capture works at the hardware level to dump the video right to disk, and so on... It can pan and zoom on a 1GB plus graphic image easily as well.
These days, it's kind of a crappy computer. Too slow. For doing video, the thing is as relevant as it was first produced.
I can think of many of these kinds of examples, where computers were distinctive in some way.
The Prop is like that, because of it's concurrent multi-processing nature. General purpose compute wise it's not always relevant, but where the concurrency lends an advantage, it's going to continue to be relevant.
What's the term for that kind of thing, other than "distinctive?"
http://en.wikipedia.org/wiki/Supercomputer
We can't fly, but superman can.
We can't shoot lasers from our eyes, but superman can.
Can a propeller supercomputer perform an extraordinary computational task? If not, I would say it is not super.
Ross Snider's encounter with the basic stamp supercomputer should have cast some light on what supercomputer means for those paying attention.
Which is reasonable, I bet that Cray Jaguar cannot do things with it's GPIO pins, if it even has any, that an Ultra Spark 40 can do.
Perhaps that term supercomputer is so entrenched to the element of speed and literary consumption that we need a new term to make a point.
There is no definition of super computing machine or super computing according to dictionary sources. It is either supercomputer or supercomputing.
Therefore a Prop with added power could be proclaimed as a Super Computing Machine. (3 words) No one doubts that the Prop, one or more, is super at computing. Heater, shut up.
Humanoido
This post intentionally left blank.
No I don't have anything cooking on the multiple-prop front. However I did have this bizarre idea:
Normally we might think of sending data from Prop to Prop via a link like FullDuplexSerial or whatever that exchanges actual numerical data. This tends to eat COGs as the receiver and transmitter have to build and decode protocol data and hence can't be used for other work.
Instead why not have the output of a COG be a frequency generated by a counter, or perhaps two. The counter will continuously send what ever frequency is set whilst the Cog gets on with other stuff. The Cog is not tied up generating bit patterns for the communication protocol.
On the receiving Prop the COG measures that frequency (or two) and hence as an input value(s) to work with. Again it does not have to dick around decoding protocol bit sequences, It can just read the frequency when it feels like it, the rest of the time doing useful work.
This is no good for normal numerical calculations but might be away of simulating neuron like systems.
That's the thing. Today's supercomputers, like the Jaguar, seem to be glorified arrays of almost normal PC hardware.
My PC here scores 5.5 million dhrystones/second. But if I want to flash a LED at speed under software control with it I'm never going to beat the Prop. Where do I plug the LED for a start?
Why struggle with the confusion caused by trying to re-tread an old word? Why not generate a new word for whatever new concept you (apparently?) have in mind? Unless you're trying to fulfill some kind of contractual technicality, what's the point of wrangling with dusty old words?
Just curious.
What I mean is something along the lines of:
If you take a bunch of props and somehow connect them together, that, by itself, does not make the collection of stuff a "supercomputer".
If you find a way to connect the props so that they all work cooperativley on a given problem in a way that is more cohesive than just some props sharing a communication bus, that might be a supercomputer. If this same assembly of stuff doesn't really make things work any different that the props running "stand alone", or with only basic communication between them, then you don't have a supercomputer.
Just shareing I/O shouldn't count.
Speed (at least relative to "something") has to be a factor.
I think that somehow, you'd also have to be able to load the system with a single program, and loading a single program that loads very specific programs for each "piece" (prop and/or cog) may not really be on target either.
At some point, there needs to be a significant difference between just a bunch of props (or whatever) connected in order to be a "super computer".
John R.
As far as finding something useful, this will be different from app to app. For example, if one's lot in life is creating algorithms, and a multiprop machine can do it a hundred thousand times more efficiently than one prop, would that would be a good point to qualify it?
As some threads are progressing with evolving a single prop, and massively increasing its speed with new overlaying frequency technology and/or function, we get into a new question, that is, can one prop become a supercomputer? Do you think, if it does some super computing thing other than doing it fast, that it can still qualify as a supercomputer? Is a supercomputer only speed dependent? Can 1000 cores be arranged in parallel over a single chip to divide up and work on a problem in super ways? What if the efficiency of a chip is raised a thousand million cycles?
Humanoido
The hardware side... does it matter if they are a cluster of glorified pcs or a cluster of interconnected execution units (like ALUs) ?... not really....
The software side... is what it will make it super. Look at the propeller, it has 8 cores, it is already a "super" microcontroller when compared (like always) with other uC. But if we write software to take advantage of all cores then it is more than a singlecore uC.
With the Jaguar happens the same one single(maybe dual)-core aware Crysis game can only get maybe 20 fps. When a 20,000-core-aware Crysis gets developed then we get 20x20,000 fps and then the machine is super !.
Humanoido's skycraper looks cool, it is cool but it needs the proper software, then we have a propeller supercomputer !. The only problem with the propeller is the slow IO
We know we will not win any race with the propeller supercomputer but the development of appropriate software does not die with the change of architecture: in the case the propII brings rapid IO methods for interconnecting props for instance.
Did I say we should concentrate on the software ? I do now
I hope it all makes sense...
Everyone has different ideas about what constitutes a useful software to make a cluster into a supercomputer. All these people cannot be wrong. There are many ways to make the cluster super. I will say it again, I really enjoy seeing small algorithms that make a single prop chip more powerful, even when it can make a cluster super.
I like to run many instances of these small enhancements, thus, the collective of instances as the sum of their parts is vastly more than a single chip, taking it into supercomputer region. There may be no immediate direct use such algorithms, but their development in itself is vital, necessary, and can be super - even taking a single chip or a skyscraper of chips to new heights.
Humanoido
Humanoido
By that criteria the prop (or any number of props) will never be a supercomputer. What the prop can be however is a supercontroller. IMHO there is no other microcontroller with the combination of speed, number of processors, and flexibility the prop has.
Humanoido
Ok, your idea is that a prop machine can be a Supermicrocontroller!
Do you know, a kind of supercomputer (yes it is frequently called a supercomputer) that is the most powerful in the world, is unmatched by any machine in I, can pass the Turing test repeatedly, runs primarily INT (does not calculate high speed floating point), is biological, and called the human brain?
It's a cool idea.
Humanoido
@Humanido, I am not saying an integer only (hardware) system cannot be considered a supercomputer. The question depends on how you define “supercomputer”, and that definition is pretty vague right now.
From looking at what are currently considered supercomputers, the criteria in order of importance seem to be: the number of floating point operations per second, the number of instructions per second, and the I/O bytes per second.
Perhaps what we need to do is break down supercomputers into several categories with more precise definitions for each category.
Computational supercomputers for problems that require a lot of math/floating point operations. Database supercomputers for accessing and sorting through large amounts of data. I/O supercomputers for storing and retrieving large volumes of data. Last, but not least, supercontroller or supermicrocontrollers for handling lots of high speed I/O control signals.
Problem is most, if not all computers contain elements of multiple categories, and how do you decide if it is “super” in a category.
Stonehenge.
Thank goodness it's never been run with Windows. We'd forever be groping through the dark, waiting eternities for the sun to rise, our orbits leaping from one system to another, crashing, blanking out, inadvertently deleting entire galaxies right and left.
As for Stonehenge.... it's just another example of antiquated silicon-based technology. The real super computers to envy are those inside the 3 billion year old microbes that cover the planet. They've been using quantum computing for a very long time. How come we primates haven't got with the program? Still messing around with the stony stuff, are we?
Humanoido
On the other hand, for those microbial endosporulated supercomputers that managed to hitch a ride on either of the Pioneer spacecraft, that deadline has been extended... indefinitely. Thank you, oh primates of earth. Hasta la vista!