How To Build a Supercomputer In 24 Hours 161
An anonymous reader writes with a link to this "time lapse video of students and postdocs at the University of Zurich constructing the zBox4 supercomputer. The machine has a theoretical compute capacity of ~1% of the human brain and will be used for simulating the formation of stars, planets and galaxies." That rack has "3,072 2.2GHz Intel Xeon cores and over 12TB of RAM." Also notable: for once, several of the YouTube comments are worth reading for more details on the construction and specs.
Pretty sure (Score:4, Funny)
that my old palm pre has more computing power than most human brains on this planet.
Headline is stupid (Score:1)
Build a computer in 24 hours? I guess it's possible.
Fund its costs and gather the materials? I guess not.
Re:Headline is stupid (Score:5, Informative)
Re:Headline is awesome (Score:4, Insightful)
I will ask the inevitable questions, as a system builder.
How many parts were DOA?
How many failed inside of the first month?
Re:Headline is awesome (Score:5, Informative)
Re: (Score:2)
Not to mention to mine the ore and manufacture the chips. It's all relative.
Re:Pretty sure (Score:5, Informative)
Computers have surpassed that level a loooooong time ago
Doubtful.
The computational requirements for simulating the human brain have been severely, even hilariously, underestimated in the past. To quote Wikipedia: One estimate puts the human brain at about 100 billion neurons and 100 trillion synapses.
That's... a lot.
First off, a lot of people think that 1 FLOP = 1 Neuron, which is not even close. The active points are the synapses, of which there are about a thousand per neuron! Each may receive an impulse over ten times a second, and involve dozens of parameters, such as the recent history of firings, neurotransmitter levels, hormone levels, membrane potentials, etc... A very conservative estimate would be that a single neuron, receiving impulses at around 10 Hz on 1000 synapses would require on the order of 1 megaflop to simulate. That's ONE neuron. Now multiply that by 100 billion, and you get a picture of what's required: about 100 petaflops, minimum. Storage is nothing to sneeze at either. Assuming a mere 50 single-precision floating point values per synapse to store all simulation state, you're looking at almost 18 petabytes of memory! That's over $100M for the memory sticks alone, even with a deep bulk-purchase discount. Unlike most server or HPC workloads those 18 petabytes would have to completely read out, processed, and possibly updated again at least ten times a second.
Second, consider that the first simulations won't be very optimized. We still don't really know what's relevant, and what can be simplified away. Hence, I suspect that the first attempts will be much less efficient, and may require 10x or even 100x as much computer power compared to later attempts. For example, neurons don't just fire impulses, they also grow and change shape. I don't think there's even a good model for how that works in the complex 3D environment of the brain!
We are getting closer, but expect to wait at least a decade or two before people start talking seriously about a full human brain simulation.
Re: (Score:3, Insightful)
Doing equivalent computation to a human brain, and simulating a human brain, are two very different problems.
You have to solve a non-linear coupled differential equation at 1MHz to simulate a 5kHz sawtooth wave generator in Spice. It's about 30 MFLOPS. But, functionally, all you're doing is generating a 5kHz sawtooth wave, which is 15 kFLOPS of work. This is a 2000:1 efficiency difference between simulating an analog system and running a direct digital equivalent implementation.
So divide that 100 PFLOPS
Re: (Score:3, Informative)
Re: (Score:3)
1 - White matter (neuroscientists are discovering) is actively involved in computation and it changes due to learning - there are 10x more glial cells than neurons
2 - "Type" of dendrite (in some cases) influences how the signal is transmitted and at what rate and this function plays a pa
Re: (Score:2)
http://www.sciencecodex.com/insync_brain_waves_hold_memory_of_objects_just_seen-101396 [sciencecodex.com]
Speciality (Score:2)
but i am an astrophysicist, not a neuroscientist ;)
Don't worry. Some of us here realise it and take it with the necessary amused grain of salt.
We understand what you meant actually:
a gigantic number-crunching machine with bat-shit crazy computing capability.
The reverse (Score:2)
The reverse would be comparing the number of stars in the sky with the number of hairs.
It's probably a gross under-estimation (I in turn am no astrophysicist), but the listener clearly gets the point:
There's an insane amount of other stars out there than our sun.
Re: (Score:2, Insightful)
Boats and submarines don't "simulate" fish, but they still swim.
Airplanes don't "simulate" birds, but they still fly.
Artifical intelligence may not need to "simulate" the brain to reach human level.
Just sayin'.
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
You also have to remember that the human brain also has to handle all of those useless background process like "breathing" and "heartbeats".
Fyi those are called autonomic functions and are handled by low level centers in the reptillian part of the brain that only do that stuff and things like balance and the other complex bodily coordination tasks that we normally don't need to think about. Some of it happens in the spinal cord alone. Some functions can control themselves if all connections to the central nervous system are severed (the heart is like this and has a built-in processor of its own). Actual conscious thinking happens in high l
Re: (Score:2)
Re:Pretty sure (Score:5, Insightful)
So these things happen on different (connected) machines if you will. Every neuron is a processor in itself anyway.
I think that's part of why computing power greatly surpassed humans long ago, and will not reach human levels for many years. The brain isn't digital. It holds an "infinite" number of analogue states, simultaneously. With massive errors and gaps filled in with guesses made from other parts, without even an minor error check that indicates that he information being determined to be "true" is 100% interpolation with 0% fact or actual memory. The very idea of an error check that was wrong more than right and kept no indication of where the result actually came from is so incredible that nobody would ever create a computer capable of operating that way. It won't be until we have computers many millions times more powerful where we can remake a "perfect" brain, until then, we'll never be able to match the capabilities of the human brain.
Re: (Score:2)
Re: (Score:3)
The brain isn't digital. It holds an "infinite" number of analogue states, simultaneously.
You can't approximate a very large number with infinity. The difference between a very large number and infinity is ... well ... infinity.
Re: (Score:2)
It is difficult for anyone that loves computers to accept, but a computer will never be sentient. [wikipedia.org] The best that can be hoped to achieved is that a computer appears sentient.
Then you get into definitions of sentience. Can a human be "dumb" enough to not be self aware? If a computer were to behave as if it were self aware, how would we be able to distinguish it from one that actually was?
The problem with the question is that we are working on the answer without defining the problem. We may have sufficient computing power to match the capabilities, but no program to simulate it. Maybe the problem isn't hardware, but software. The problem is that the science concerns itself
Re: (Score:2)
Nice attempt at a hand-waved dismissal, but pretending we don't have precise definitions of words is a trap.
Have you ever looked in a dictionary? There are lots of definitions. In fact, almost all words have more than one. I'm not pretending we don't have definitions of it, but that my precise definition of it may not match yours, and we'd both be correct according to every dictionary. Or perhaps you are using the psychological definition and I'm using the philosophical one, both precise, but not identical.
In the same sense a computer can simulate intelligence, understanding and subjective experience for an observer, but nothing will actually be "thought," understood, or subjectively experienced by the computer. It can, at best, only appear that way. No amount of advances in digital computer technology can change this. It is an impenitrable philosophical wall that cannot be breached.
So your opinion is that it's impossible, so that any discussion on it is like discussing the color of Sa
Re: (Score:2)
You are not arguing against what I proposed,
You proposed nothing, you just attacked my rhetorical style.
Re: (Score:2)
Will a computer ever be self aware?
Or will it always tell you that it's self-aware because that's what it's been programmed to do?
Re: (Score:2)
Re:Pretty sure (Score:4, Interesting)
Google's cars say different.
Looking at physical complexity:
A human brain has about 86 billion neurons. An Intel Core-i7 process has 731 million transistors. A neuron is more complex than a transistor. Let's say it does a job, for the sake or argument, that would take about 16 transistors. So say the Core-i7 has the equivalent of about 45 million neuron-equivalents. That's a factor of about 1900 in physical complexity.
But the brain manages to pull off a clock cycle about 200 Hz, based on the neuron's firing rate. Maybe 1000 Hz at most. The clock rate of the CPU is 3.2 GHz. It is 16 million times faster than your brain.. Since the computer can execute programs of arbitrary complexity, it can simulate your brain's operation -- if properly programmed, with a much smaller hardware set running much faster. In raw computational capacity, it apparently has 16 million / 2000 = 8000 times the computational capacity of your brain. So even if its' simulation were quite computationally inefficient, it should still be able to do the job of a number of brains, if programmed to do so.
In short, exceeding the capacity of a human brain isn't a hardware problem any more. It hasn't been for years. It's a programming exercise, albeit a particularly challenging one.
Re: (Score:3)
That's a rather simplistic analysis. The problem is that we don't even know how the brain fully works. If, for instance, neurons make use of quantum properties (a very real possibility), then classical computers may be hard-pressed to replicate even a mouse brain. We don't know whether quantum computers are more powerful than classical computers. Many believe that they are, but we have not found proof of this. They could be anywhere from equal to exponentially faster, and you know what exponentially faster
Re: (Score:2, Insightful)
That's a rather simplistic analysis. The problem is that we don't even know how the brain fully works. If, for instance, neurons make use of quantum properties (a very real possibility), then classical computers may be hard-pressed to replicate even a mouse brain.
I have only one thing to say about that: Pfffffbbbbbt! The brain is an electrochemical machine that works on a macroscopic scale. It can't distinguish and act upon the quantum state of a a particle any more than a hedgehog can sing opera.
All speculation on the brain possibly having quantum properties is complete nonsense based on equivocation and quasi-religious hoo-hah.
We don't know whether quantum computers are more powerful than classical computers. Many believe that they are, but we have not found proof of this. They could be anywhere from equal to exponentially faster, and you know what exponentially faster means with numbers like what you've been pulling.
Your entire post is based on the premise that a neuron and a transistor are even comparable. That's one hell of an assumption.
If you're going to discuss the computing power of a human brain, you are by definition comparing neurons to switches. It's one electric
Re: (Score:2)
All speculation on the brain possibly having quantum properties is complete nonsense based on equivocation and quasi-religious hoo-hah.
That may be so, but Einstein thought this was possible. Not one given to equivocation or quasi-religious hoo-hah. There are plenty of macroscopically detectable effects of quantum phenomena.
Re: (Score:2)
All speculation on the brain possibly having quantum properties is complete nonsense based on equivocation and quasi-religious hoo-hah.
That may be so, but Einstein thought this was possible. Not one given to equivocation or quasi-religious hoo-hah. There are plenty of macroscopically detectable effects of quantum phenomena.
Einstein was a physicist, not a neurobiologist. He knew less than most modern scientists about how brains work.
Biology is against you (Score:2)
A neuron is more complex than a transistor. Let's say it does a job, for the sake or argument, that would take about 16 transistors
Let's say that, you completely under estimated the computing power of a neuron.
First of all, a transistor just takes a few inputs, integrates them and give 1 signal out. To over simplify (medical doctor/researcher speaking here, so it will be an abusive over simplification) it will work like a basic boolean operator on 2 input bits.
A neuron is several order of magnitude more complex than that. It takes many more different inputs. The connection between two neuron is a synapse: neural impulse comes to one ex
Re: (Score:2)
I don't see why it's necessarily a wrong estimate. 16 bits represents 2^16 possible states and a large number of possible connnections. What if I doubled that? Would you be happy then? What if I multipled it by SIXTEEN, so it had 256 transistors, 2^256 possible states and hundreds of possible connections? IT STILL WOULD BE LESS PHYSICALLY CAPABLE THAN A MODERN MICROPROCESSOR because the microprocessor is still 16 million times faster. How many transistors would it take to make you happy and how do yo
Re: (Score:2)
The brain is both a computer and a not-computer. It is a biological system encompassing process of external stimuli and control of internal systems - including itself. The latter part of that description describes a computer, in the broad sense. The former part, an integrated, multitasking system with many, many inputs, an organic network which is constantly evolving (adapting) to meet the needs of the larger organism to fit its environment (or to allow it to change its environment to suit it). There is no
Re: (Score:2)
There is no artificial computer based on silicon that even comes close to the adaptability or complexity of the brain
That I'll grant you.
- not even considering the clock speed.
I thought we were talking about computing capacity. You can't consider computing capacity without considering the speed of operations.
That means precisely nothing when you're talking about a system that doesn't need a reboot every 47 days or grinds to a halt when it runs out of random access memory or shuts itself down when it gets too hot.
It means something. It just means that a brain's operating system has significant advantages for managing an animal's systems. But you aren't considering downtime, for example. Your brain requires a partial shutdown approximately every 8 of every 24 hours, is not capable of continuous attention AT ALL, is subject to huge perceptual errors and has many
Re: (Score:2)
you have mechanical protection in cerebrospinal fluids and a network of intracranial fibroblasts (a sort of scaffold for the brain), for a start. Thermal control by way of the circulatory system. Pain management by way of pentobarbitone-analogues and other chemicals produced by the body to reduce the risk of permanent damage following traumatic brain injury by putting the host organism to sleep.
It's already done. (Score:2, Funny)
Re: (Score:2)
Re: (Score:2)
Re: (Score:1)
I find it hard to believe a standard LAN would be sufficient. You need static routing and possibly optimized paths for gather-scatter of the data. There is a lot of work in designing and optimizing the topology of the network. A very important number is latency. Can you get a reliable microsecond-scale latency on the cloud? In these things if any node lags all the computation does.
I am sure Amazon and buddies can deploy good and even exceptional hardware, just that it is not what they are selling normally a
Re: (Score:1)
HPC is not about raw computing power anymore, Amazon or Google have way more power than most supercomputers. HPC is memory bound. CPU memory bandwith, interconnect bandwith, and filesystem bandwith.
Try writing results from 500.000 processors to disk once. Each processors writes to one file? You end up with 500.000 files. Try doing ls on 500.000 files in a distributed filesystem, it'll take hours. Actually, they write to disk 1000s of times per day, to a single file. Amazon and Google can actually wait for w
Re: (Score:2)
Title could be (Score:4, Insightful)
Price, from comments:
Just under 750,000 Swiss Francs, or about $800,000
Re: (Score:1)
Amazing things can be accomplished with cheap labor and tons of money: look at the pyramids... these people didn't aim high enough!
Comment removed (Score:4, Funny)
Re: (Score:2)
1% of the human brain ? (Score:2)
Re: (Score:2)
Yes indeed it's a ridiculous claim. If it has the computing power of 1% of the human brain, why even build it?
Re: (Score:3)
> The computing power of the human brain is infinite, by some standard definitions, as it is analogue (at least partially, according to some theories) and there are an infinite number of analogue states.
Just because the brain theoretically has a practically infinite number of possible states does not mean all (or even most) of those states are meaningful and important. People lose thousands of neurons each day without changing significantly (with respect to both personality and intelligence). Obviously t
Re: (Score:2)
> the brain is supposed to lose neurons that aren't needed because the brain uses such a large portion of the caloric intake.
That does not seem like a good explanation. The human brain has about 100 billion neurons, so even if you lose 100.000 neurons every day, that means only a 0.04 % reduction every year, which does not make a meaningful difference in brain energy consumption.
Do explain how extra cells you aren't using don't qualify as 'unnecessary detail'.
Re: (Score:2)
1. A neuron isn't really analog. A synapse either fires or it doesn't.
The speed and strength of the firing matter. Also, each neuron is, itself a DPU (distributed processing unit, as CPU has no meaning with so many "equal" processors), so the firing this time may be digitial, but it sets an internal state that is analog, affecting all future firings.
2. Claiming that an analog system's operation can't be adequately simulated by a digital system is wrong. The analog system is only different from the digital system if the digital system doesn't simulate it down to the noise level.
You answered yourself. A digital system can only "simulate" analog. Whether we are able to notice the difference is a separate question.
3. What's the noise level of a synapse anyway?
I've read they are looking into quantum entanglement as a part of the brain process. So, w
Re: (Score:2)
> I've read they are looking into quantum entanglement as a part of the brain process.
Is there actually any evidence that this is happening, or is it just someone's pet theory?
Re: (Score:2)
Re: (Score:2)
I did spend a few minutes on google before posting that reply, did not find anything substantial. The theory appears to be pushed primarily by Hameroff and Penrose, but I could not find any experiments that support their theory. Other scientists appear to be arguing that either quantum effects are not required to explain the operation of the brain, or quantum systems in the brain would decohere too quickly to have a meaningful influence on computation. Since you originally introduced the Quantum mind theory
Re: (Score:2)
It wasn't a statement that I believe it to be true, but that there are still lots of theories, and we don't know enough to eliminate them, so we obviously don't know too much.
Re: (Score:2)
there is more than simply firing and not firing as it fires different things and different combinations of those things
Yes, but... (Score:2)
Re: (Score:3)
Didn't you notice the modern miniaturized version of on board Beowulf cluster integrated in a chip chips they put it there? I would guess they put about 50 of them in the rack but since it was in fast forward, I couldn't count accurately. Anyway, if they build a Beowulf cluster of those, we will end up with a Beowulf cluster of Beowulf clusters.
News For Nerds (Score:1)
Now this is what I call news for nerds. None of this 'how would you crap'!
Go team zurich.
Re: (Score:1)
Re: (Score:2)
that... is a lot of juice... not something I could run in my garage then?
Also notable: (Score:5, Funny)
"for once, several of the YouTube comments are worth reading for more details on the construction and specs."
Yeah, unlike the impeccably high standard of comments you see on Slashdot. Mod me up if you hate Bieber!
Re:Also notable: (Score:4, Insightful)
Racist comments don't routinely stay modded up on slashdot as they do on many parts on YouTube. Use of dubious debating techniques such as the strawman usually gets noticed here. Unpopular viewpoints are often modded up to +5 Interesting if they are sufficiently well argued.
Slashdot ain't what it used to be, but it still has standards.
Re: (Score:2)
Ironically I'm modded up to 5. I assume this was from the Bieber quip.
Be Professional (Score:2, Insightful)
1) Brag about how you will succeed well before knowing what it's all about.
2) Immediately after, seek the lowest standards you should comply with.
3) Then, study rhetorics in order to getting away with even lower standards.
4) Subsequently explore the deep and dark lows and lower your standards to the absolute minimum.
5) Hiring time. Get yourself people capable of realizing your preposterous proposition and seek the lowest fee to pay.
6) With a bit of delay -bein
That pesky static discharge (Score:5, Insightful)
Re: (Score:2)
I was messing around with micro controllers a few years ago and had a few mysteriously die on me.
I started to wear an antistatic wrist strap and did not have another unexplained failure.
PC components may not be as fragile but I always ground the case before doing any work and wear my AWS when handling expensive or important hardware.
Re: (Score:2)
I always ground the case before doing any work and wear my AWS when handling expensive or important hardware.
you wear amazon web services?
Re: (Score:2)
The risk is so small that most people, rightly, choose to ignore overbearing advice about the handling of "sensitive" electronic equipment.
Re: (Score:3)
The risk of damage depends on the sensitivity of the component. For instance good luck wreaking a standard BJT with ESD. On many ICs the I/O is either suitably protected against ESD, or the nature of the design makes it less vulnerable to the effects of a sudden static spike.
It's a very different story for say working with high frequency MOSFETs, uncut silicon wafers, or pretty much any RF gear. I remember in our lab one of the PHD students was working on a silicon wafer. The instrument had the wafer suspen
Re: (Score:2)
Software (Score:2)
will be used for simulating the formation of stars, planets and galaxies
It was nice to hear about the beefy specs, but how about a bit more information about the piece above? What kind of simulation, what software applications and so on.
Re: (Score:2)
full specs and the system config: http://www.itp.uzh.ch/~stadel/doku.php?id=zbox:zbox4 [itp.uzh.ch]
Re:Software (Score:5, Informative)
Re: (Score:2)
http://www.youtube.com/watch?v=9KKgvRw1rrU&feature=endscreen&NR=1
Just asking aaaait?
Re: (Score:2)
I just logged into it and yes since bc is installed:
$ bc
'bc 1.06.95
Copyright 1991-1994, 1997, 1998, 2000, 2004, 2006 Free Software Foundation, Inc.
This is free software with ABSOLUTELY NO WARRANTY.
For details type `warranty'.
9999999999999999999999999999999999999999999999999999999999999 *\
filter trap
9999999999999999999999999998999989999989899999999999999999999976499
filter trap
99999999999999999999999999989999899999898999999999999999999989764990\
filter trap
000000000000000000001000010000010100000000000000000000
1% of Human Brain Neuro Cap. & Used For *Whar* (Score:1)
The machine has a theoretical compute capacity of ~1% of the human brain and will be used for simulating the formation of stars, planets and galaxies."
May I be the first to say; Formation of stars, planets, and galaxies my ASS!
Nominate it in a special Act for POTUS!
I mean, c'mon. Could it seriously be that much worse than the choices at present?
At least then, maybe the US populace would begin to grasp the concept of GIGO (garbage in, garbage out) and maybe begin to apply it to the other parts of government. And no, nothing at all to do with political party/ideology. Rather, more a perspective from a "CS101 basics" point of view. :)
Strat
OCZ Rebates (Score:5, Funny)
Nice rack! (Score:2)
and it actually runs Linux.
"YouTube comments are worth reading" (Score:2)
Individually, five words. Collectively, in that order, the cause of me needing a new keyboard.
WTF?! have these kids never heard of ESD? (Score:3)
NO grounding straps.
NO signs of any ESD precautions!!!
Lacking from the video is the debugging process.
Sure they built it in 2 days... but how many nodes came ready?
I was cringing through the whole video over their lack of concern for basic ESD prevention. They don't need to be wearing bunny suits or anything that extreme, but FFS.... could ya show a little bit of respect for the hardware? Heck even clipping the freaking base-plates to ground during assembly would have been more than adequate.
That video was like watching "OW MY BALLS" for geeks!
Re: (Score:2)
Maybe that 2% loss was due to sloppy ESD protocol? You'd never know, unless you take proper precautions, and vet nodes before final assembly.
Hey... but that is what ya learn in school, yes? To cross all the 'T's and dot all the 'I's?
Re: (Score:2)
Sorry for the second replay.
IF you had a 2% fail on assembly in a commercial environment... you'd be looking for a new job.
Re: (Score:2)
Re: (Score:2)
2% is below average for an OEM. (IWAOEM (I was an OEM)). Most distributors have an RMA warranty for OEMs.
Re: (Score:2)
no nodes came ready
I'm going to assume that my comment was misunderstood.
When accepting new hardware it is typical to verify that it works.... THEN assemble it into the desired configuration. THEN verify that each sub-system is fully functional... at least all the I/O. These are basic functional tests... yes?
Where I have issue with the video is the lack of precautions taken during assembly...
That is all.
Re: (Score:2)
A lot of CPUs != Supercomputer (Score:2)
A lot more goes into make a 'super' computer than just a bunch of cpus and some ram.
Re: (Score:3)
But what about all the Garbage? (Score:5, Interesting)
Do it yourself cluster (Score:2)
I sometimes look at the ads for the local computer stores and add up what it would cost to roll my own cluster. At 2012 prices a 32 core cluster (say, 8 Core i5 CPUs) would cost only a little more than my first computer, that I bought in 1986. And that's at retail prices. I'm sure if I wanted a bulk purchase, the stores would cut me a deal.
Then I wonder what I would do with it, and decide I have better things to spend my money on...
...laura
Re: (Score:2)
Then I wonder what I would do with it, and decide I have better things to spend my money on...
...laura
Blasphemy!
Turn in your Geek card on your way out.
Faster to do it on the Amazon Cloud (Score:2)
- http://www.wired.com/wiredenterprise/2011/12/nonexistent-supercomputer/ [wired.com]
Re: (Score:3, Funny)
Comment removed (Score:4, Informative)
Re: (Score:2)
I can translate that into something a bit more modern. Don't ask why, you just owe me a cold beer later.
said to him, is your will that we go and gather them up? 13:29
But he said, no; unless while you gather up the weeds, you also root up
the wheat with them.
13:30 Let both grow together until the harvest: and in the time of
harvest I will say to the harvest crew, Get the weeds pulled first,
and bind them in bundles to burn them: but put the wheat into my
barn.
13:31 Another parable he gave them, said;" The kingdo
Re: (Score:2)
You're not missing anything - and I had the same thought about the Swiss ball. It is really too bad that they didn't take basic anti-static precautions - it really isn't hard at all. It's like wearing your seat belt...
Re: (Score:2)
It supports Crysis 3 only. Crysis 1 and 2 are not supported because of some input problems related to the complexity of game controls in 1 and 2.
Re: (Score:2)
that'd make Pamela Anderson look flat chested...
Re: (Score:2)
No, it's 42.99999997
FDIV bug, for those who missed it.