With the seemingly rapid increase in supercomputer development, what exactly is the State of Computing Power relative to the human mind?
First, is there a direct comparison? Not really.
Supercomputers, such as those ranked in the semi-annual Top500 listing are typically designed to calculate math-intensive equations (e.g. decrypting codes, nuclear simulations, climate modeling).
In fact, among supercomputers there is a diverse ecosystem of computing specialities. For instance, the Japanese MDGRAPE-3 is capable of performing approximately 1 petaflop, which is more than 3 times faster than the IBM Blue Gene/L, the fastest computer listed on the Top500 chart.
Yet, despite this feat, the MDGRAPE-3 is not considered a “general purpose” computer due to its use of specialized pipelines designed to simulate molecular dynamics.
So, what are these FLOPS?
It stands for Floating Point Operations Per Second. It is used to measure a computer’s performance, especially those that make heavy use of floating point calculations. A floating point is just a series of numbers with a decimal point somewhere in it.
If you are familiar with the nomenclature heirarchy used to measure RAM or hard drive space (e.g. 16 megabytes, 120 gigabytes), you can see the similarity below:
megaFLOPS (MFLOPS, 106 FLOPS)
gigaFLOPS (GFLOPS, 109 FLOPS)
teraFLOPS (TFLOPS, 1012 FLOPS)
petaFLOPS (PFLOPS, 1015 FLOPS)
exaFLOPS (EFLOPS, 1018 FLOPS)
Here is an idea of what kind of performance popular chips are rated:
AMD Athlon @ 600 mhz – 2.4 gigaflops (single precision), 1 gigaflop (double precision)
Pentium 4 @ 2 ghz – 8 gigaflops (single precision)
Pentium 4 @ 3 ghz – 12 gigaflops
Athlon 64 X2 4600 – 14.7 gigaflops, 17400 MIPS
G5 Dual 2.3GHz – 30 gigaflops
XBox 360 Xenon chip -115 gigaflops
XBOX 360 Xenos graphics chip – 240 gigaflops
nVIDIA 7800 GTX 512 – 200 gigaflops
ATi X1900 – 553.8 gigaflops
A couple of things to keep in mind. These are theoretical speeds on paper, not what is necessarily done in the “real world.” System buses, RAM type, and I/O speeds all will effect the overall system performance. Furthermore, performance-per-watt and MIPS should be considered as well.
MIPS stands for million of instructions per second. This is typically measured by how long it takes to execute a mix of sequences including various synthetically designed programs and real-world applications.
Pentium 4 – 1500 MIPS at 1.5 GHz – released in 2000
AMD Athlon FX-57 – 12000 MIPS at 2.8 GHz – released in 2005
AMD Athlon 64 3800+ X2 (Dual Core) – 18900 MIPS at 2.2 GHz – released in 2005
AMD Athlon FX-60 (Dual Core) – 22150 MIPS at 2.6 GHz – released in 2006
A watt is a measurement of energy use relative to time (e.g. 1 Watt is 1 joule used in 1 second). Products you might be familiar with:
Xbox 360 – 160 Watt
GeForce 7800 GTX – 100 Watt
ATi X1900 XT – 132 Watt
Athlon 64 @ 2.4 Ghz – 62 Watt
Intel Core Duo 2 @ 2.67 Ghz – around 65 Watt
By comparison, the human brain uses about 20 watts or about 100 watts to power the entire body. And the average household light bulb uses 40-100 watts.
Furthermore, there is no real consensus regarding the computational power of the human brain. Some estimates suggest that it is capable of 1017 FLOPS, or 100 petaflops.
According to Hans Moravec of Carnegie Mellon University (whom used the processing speed of the retina), estimates that the brain operates at 100 million MIPS (or is it 100 teraflops?). Nick Bostrom of Yale estimates it to be 1,000 times higher than Moravec’s number (or 100 petaflops). [Note: one must not conflate specialized hardware designed to defeat "grandmaster" chess players with the wide ranging cognitive abilities human brains are capable of].
Also, Japanese politicians have decided to fund the construction of a 10 petaflop computer system by 2011. Contrast that with IBM’s private partnership with a Swiss Institute, to develop a 28 teraflop computer to emulate the activity of brain neurons — appropriately called Blue Brain.
This is part one of a discussion surrounding synthetically created intelligence, mind uploading and the technological singularity. And I am not betting on these phenomenon happening anytime soon.
[...] See my previous post on the human brain, FLOPS, MIPS, and Watts. Be sure to also check out Geordie Rose’s blog as well as “Programming the Universe” by Seth Lloyd. « There’s a sucker born every minute [...]
Pingback by Seth Lloyd’s Million Megahertz CPU » Doctor Recommended — August 24, 2006 @ 12:19 am
I am looking at the monitor, and out of the corner of my eye, a poker program on TV. I am typing, and keeping track of the position of my fingers and the things they type, and watching the words just to make sure I make no mistakes. I am simultaneously processing what I hear from the poker program, understanding that Daniel Negreanu just won a hand. I also know that I probably spelled that name wrong. Simultaneously, I am rocking slowly in my neat rocking computer chair here, and someone just flopped a nut straight. I am also breathing, my heart is beating, and various other bodily processes no doubt are transpiring. I am also processing that I would prefer it to be slightly cooler in my room, and wondering what I will do tomorrow. My brain is also handling all the various other senses – the press of the chair against my back, the tactile feedback of my fingertips on the keyboard, the sensation of air entering my nostrils, and of hair upon my head.
Millions upon millions of inputs – nerve cells in my skin, rods and cones in my eyes, cilia in my ears, and the feedback routines of my own thoughts and emotions, transferring into thousands of simultaneous inputs, including every twitch of every muscle I am presently using, with precise placement and strength, and every letter I am typing at this moment.
My computer, at any moment, is doing a single calculation.
Since it can do so many in a single second – millions and millions – it seems simultaneous to us. But it can only do one thing at a time; we actually are processing things in parallel. So while I can’t calculate 2.3×10^238 to the 14th power, I can quickly and easily identify what those numbers in that font are – optical character recognition programs remain crude and imperfect. So while the calculation would be a single operation, recognizing the numbers is not – you have to look at the shape, the configuration, the colors, perhaps even the font, the context, and figure out just what is being asked.
It will be a very, very, very long time, if ever, until a computer can match the power of the human brain. And then we have to wonder if human intelligence is an emergent property that cannot be duplicated – but this is not a question for this generation.
Comment by Andy — August 24, 2006 @ 12:43 am
I find your finding interesting as Apples website lists 2 G5 CPU’s can push out 10.1 Gigaflops per sec, and not 30… Is there a difference in their results from yours?
Comment by Sledgehammer70 — August 24, 2006 @ 3:54 pm
Sledgehammer,
That is a good question. I’ve read several advertisements, such as Amazon.com, that suggest that the quad-core setup (2 dual-core chips) hits roughly 76.6 gigaflops when operating at 2.5 ghz (double-precision too): http://www.amazon.com/gp/product/B0009U7YFK/002-2409018-5215206?v=glance&n=541966
However, Apple Insider notes that the dual 2.3 ghz version Xserve hit only 35 gigaflops: http://www.appleinsider.com/article.php?id=807
The official Apple press release for the Xserve system also states 35 gigaflops: http://www.apple.com/pr/library/2005/jan/04xserve.html
So to answer your question, perhaps they were using some other test to measure the synthetic benchmark (maybe it is the theoretical paper value versus whatever test you found listed on their site).
Comment by Tim — August 25, 2006 @ 5:01 am
[...] To put that into perspective, I have discussed GPU and CPU theoretical benchmarks this past week and noted that no desktop CPU is currently capable of achieving anything near that number (the newest ones, such as Intel’s Core Duo achieve about half of that). [Note: the IBM-designed Xenos CPU found in the XBox 360 hits over 100 gigaflops on paper, however it is not a general purpose CPU] It should also be noted that with chipmaker AMD’s purchase of ATI, several reports have predicted that GPU’s and CPU’s will be manufactured on the same die before the end of the decade. [...]
Pingback by GPU versatility » Doctor Recommended — August 26, 2006 @ 8:25 pm
[...] In the past I have quoted several numbers which supposedly gauge the computational power that a human brain is capable of. Several of these estimates come from Hans Moravec, a computer scientist and robot specialist at Carnegie Mellon. [...]
Pingback by Going back to the source » Doctor Recommended — September 4, 2006 @ 12:32 am
[...] Last month I mentioned Project Blue Brain in passing. One of the aims of the research project is to simulate the neocortical column of a rat, which contains around 10,000 neurons. [...]
Pingback by Brain simulation wrap-up » Doctor Recommended — September 7, 2006 @ 12:02 am
how to calculate Mflops.Is the formulafor calculating mflops is for matrix 2*2 is
2*matrix size*matrix size ie 2*2*2;
if not then what is the formula for calculating mflops.And after calculating mflops how to get performance of program if we know the clock speed of computer.?
Comment by shweta — September 11, 2006 @ 5:05 am
[...] See also: FLOPS, MIPS, Watts and the Human Brain. « Weird Al versus Chamillionaire [...]
Pingback by High Definition Eyeballs » Doctor Recommended — September 25, 2006 @ 12:51 am
anyone knows how to calculate a system flops??? I mean if a have a desktop pc, calculating all latency, how can I calculate this????
thanx!
Comment by bangerang — October 11, 2006 @ 5:02 pm
[...] Academics such as Hans Moravec, that study brain processing power, suggest that you need 20 petaflops or so to “brute force” the emulation between the human eye and brain. [...]
Pingback by Specialization, Centralization, and the Future of Chip Integration » Doctor Recommended — December 18, 2006 @ 4:21 pm
While considering the full computational requirement for human emulation is an interesting excercise, it should be considered just that. The emulation of a human will not be a sinlge processor but a chained set of processors with specialized tasks, sharing a common memory and additional memory for specialized purposes. the trick we havent figured out is which specialized tasks and how is the memory shared. Recent announcements of new computers from ibm, intel and others indicate that these specialized capacities may in fact already be available or will very shortly be.
Comment by didi_d — February 15, 2007 @ 1:30 pm
This was a SWEET article.
It feels great to know what an asset the brain is.
It is also dual core. And in mammals, it could be said that the brain is optimised for 64 bit code.
You have your left and your right lobes, having millions of sensory cells is like running a billion programs simultaneously.
The brain being “dual core” has to do with being able to hear in 3d (from two different directions), to escape from predaters, and hunt for food.
Comment by MAX — March 1, 2007 @ 11:52 am
Why they cant deside how much flops has human brain. 100 TFLOPS or 100 PFLOPS?
Comment by rat — May 4, 2007 @ 4:42 pm
Andy, I beg to differ, in a sense at least. While some estimates may be correct, placing the capacity of the human mind in the vicinity of 100 petaFLOPS, think of all the things you mentioned that a computer needn’t be ‘fettered’ with:
typing…
keeping track of the position of fingers…
knowing it probably spelled that name wrong…
rocking slowly in a chair…
breathing…
beating a heart…
various other bodily processes…
processing that I would prefer to be slightly cooler,
(okay, well, what computer wouldn’t want that, ;o)
My point is that, at least in order to pass the Turing test, such a supercomputer would only need to focus on logic, virtual emotion, natural language, and perhaps visual recognition. No mean feats, to be sure, but a comparatively small part of the entire human condition.
I believe that once the Turing test is passed, the singularity will follow shortly thereafter. I also believe the Turing test will be passed in less than ten years.
Comment by Jack — September 2, 2007 @ 5:12 pm
[...] also: FLOPS, MIPS, Watts and the Human Brain Seth Lloyd’s Million Megahertz CPU Specialization, Centralization, and the Future of Chip [...]
Pingback by What is wrong with Moore’s Law? » Doctor Recommended — September 9, 2007 @ 7:00 am
great review. any updates as for 2008?
Comment by Valery — April 22, 2008 @ 4:00 pm
[...] and the Future of Chip Integration The Revolution Will Be Televised, Via IPTV And Micromachines FLOPS, MIPS, Watts and the Human Brain 10 years from now, where will they be? Can the Future Do Without Economic Logic? Seth Lloyd’s [...]
Pingback by The rapture is coming, look busy » Doctor Recommended — June 4, 2008 @ 2:08 am
I think its important to keep in mind several things when it comes to the area of synthetic intelligence. Firstly, while a useful tool, the Turing test is just that a bench mark. I’d wager depending on the test’s judge, you’d have a hard time getting some ‘humans’ to pass. Secondly, the criterion we utilize for both computers and human intelligence are vastly different. We expect that computers will indeed be capable of ‘perfect’ recall, we as a species generally aren’t. The search for an truely artifical intelligence won’t be based only on the ability to mimic the computational ‘power’ of the human mind, but to mimic its developmental process as well. That in and of itself will be quite a challenge as that wide array of sensory input that forms the basis of our cognative framework is largely missing from computation.
Given 10 years, and the ability to create moderately specialized but interlaced multiple cpu systems, and the ability to ‘feed’ those systems sensory information, we’ll find that we will be capable of growing up a truely artificial intelligence. Sadly,
I think that will happen before we decide what our ethical and moral responsibilities to such an intelligence should/can/would be.
Comment by Eric Scudder — June 26, 2008 @ 4:31 pm
Still, the entire genetic code for a human being is only 3GB. Your DNA would easily fit on some $30 Flash cards.
Part of the equation is not how fast the computer is, but how efficiently it’s coded. Additionally, there is no reason why a computer needs to exactly mimic a brain to give rise to intelligence. Plus, a lot of the sensory I/O humans have isn’t even necessary to include.
If a machine is smart enough to improve upon itself, it can guide it’s own evolution faster than we could design it in. And that’s exactly what some AI professionals are trying to do. All it takes is one breakthrough, and that could feasibly happen at anytime now.
Comment by tonedef — November 20, 2008 @ 5:57 pm
Look up Raymond Kurzweil if you are interested in this stuff. We are not too far off.
Comment by PeterFlop — December 3, 2008 @ 12:05 am