Slashdot Mirror
'Inexact' Chips Save Power By Fudging the Math
Barence writes "Computer scientists have unveiled a computer chip that turns traditional thinking about mathematical accuracy on its head by fudging calculations. The concept works by allowing processing components — such as hardware for adding and multiplying numbers — to make a few mistakes, which means they are not working as hard, and so use less power and get through tasks more quickly. The Rice University researchers say prototypes are 15 times more efficient and could be used in some applications without having a negative effect."
37 posts about the Pentium division bug.
Don't they do this too, but fudge the maths so they can be a bit faster?
Hence the Cuda stuff needed special modes to operate in IEEE floats etc...
Meh. Close enough.
I read about that like 3 years ago, if not more.
I wish I could say reading the article gave me some insight as to where it fudges, but they kinda left it out.
He tried to kill me with a forklift!
http://xkcd.com/221/
CC.
TaijiQuan (Huang, 5 loosenings)
a basis for AI that can make quick inexact decisions?
These chips will, of course, be aimed at government markets.
Didn't Intel do something like this in 1994? :)
This is first post according to my new power-efficient computer!
See my journal for slashdot ID's by year. Mine created in 2005. http://slashdot.org/journal/289875/slashdot-ids-by-year
They could be useful in a few small circumstances, but for the vast majority of cases, I'd be interested in how a speed payoff is going to be beneficial given you don't know whether you got the correct answer. You could run a check to see whether it's correct, but then you can't trust the check to give you the right answer either... so you could run a third check...
Someone somewhere will end up killing people with this.
that strategy has always worked for me.
Please do not read this sig. Thank you.
I can already hear people arguing "no, no, I did not fudge with the numbers, it's the computer chip" :)
Never antropomorphize computers, they do not like that
At least our eventual computer overlords won't be able to count accurately to be sure they've eliminated all of us...
technical whipping boy, Occam's Strop (think about it...)
I mean, what could possibly go wrong?
From here on out, I'm requiring my chips to show their work. And, it better not look the same as the work that that northbridge chip you are sitting next to.
Seems like nothing new to me. Floating point binary math is basically used for the same reason. It gives us and answer that's close enough, without requiring too much computation time. And it causes all sorts of fun since even simple numbers like 0.1 can't be represented exactly in binary floating point. Binary floating point works well for scientific apps, but fails quite badly at financial apps. I think this is basically taking floating point to the next level where the calculations are even more off. Which might work for certain applications, but for other types of applications would be completely catastrophic. What really bothers me is languages and platforms that provide no ability to work with numbers in a decimal representation.
Anthropic principle: We see the universe the way it is because if it were different we would not be here to see it.
Before someone comes up with that stupid remark, not much. :) If the chips are 15 times as efficient as normal ones, it means that you could run for instance four in parallel and rerun each calculation in which one of them differs. That way you would both get both accurate calculations and power savings. Modify the number of chips to run in parallel depending on the accuracy and efficiency needed.
Football Odds
"This isn't so much a circle as a square, what the hell's going on?!"
"Oh, that's because the chip in your machine doesn't accurately define PI, it rounds the value up"
"To what?"
"4"
Summation 2
Anyone else think of http://en.wikipedia.org/wiki/Technology_in_The_Hitchhiker%27s_Guide_to_the_Galaxy#Bistromathic_drive when they read this?
Hmm, seems this has been used by The Fed and European Central Bank for quite a while now.
"But we decide which is right, and which is an illusion"
Speed or accuracy?
At last! Somebody who finds this as scary as I do!
http://en.wikipedia.org/wiki/Pentium_FDIV_bug/
Remember "News for Nerds, Stuff that Matters"? Help make it a reality again! http://soylentnews.org
In more recent news, computer scientists determined that monkeys can get the same job done even faster, and by using even less power, and by making, um... a lot more mistakes.
Where I work, we call this "the much-faster 'wrong' algorithm". It's frequently a side-effect of overly-enthusiastic attempts at optimization, sometimes by people, and sometimes by compilers.
2*3*3*3*3*11*251
Hasn't this concept been around since the 60's?
This is exactly the problem with American chips lately. They're too lazy to put any effort into their work. Sure, they're "saving energy" but that just means they're going to become even more obese. Chips from many Asian manufacturers are already much more accurate and efficient than American ones. We need to encourage American chips to be more interested in STEM fields if we're ever going to turn our economy around!
what happens when the "Minority Report" is actually CORRECT??
now if you are doing a calc with 9 significant digits and you are only using 5 then if the error is limited to the "extra" digits then it might work (or if the error is in a predictable range).
Do you really want the ropes supporting you 800 meters above the ground to have had their strength OVER estimated??
Any person using FTFY or editing my postings agrees to a US$50.00 charge
the concept works by allowing processing components — such as hardware for adding and multiplying numbers — to make a few mistakes, which means they are not working as hard
But my math teacher didn't understand the important difference between efficient and lazy.
This concept was used a lot back in my high school.
Excel has been running its numbers through an emulated version of this hardware for years.
1 (short ton / firkin) = 89.1432354 slugs / keg
For signal processing this would be great. Lost art: analog multipliers, integrators and stuff...
http://tech.slashdot.org/story/09/02/08/1716235/sacrificing-accuracy-for-speed-and-efficiency-in-processors
Of course, you might've been sacrificing speed for accuracy in that 3 year estimate.
(and for all of the nay sayers -- I could see this being great for monte carlo simulations or other modeling where you're dealing with so much imprecise inputs that minor error's not going to be significant)
Build it, and they will come^Hplain.
This device has already been invented by a certain Tom Peters. I present: THE CALCUCORN:
http://video.google.com/videoplay?docid=-7064178994016272127
(skip to 7:15)
With the first link, the chain is forged.
I write 15 times as much code by not bothering to fix the mistakes.
Of course, we can't trust that number if it was run on the chips in question. What is the margin of error? Plus or minus two to the fourth power?
If you are not allowed to question your government then the government has answered your question.
Wow, so the goal to be Green in the future is to introduce more bugs into hardware to save power. While I am sure there are limited uses of this kind of "math" in general I don't believe these chips will have widespread adoption because mathematical accuracy, at least for integer values, is kind of critical for most applications. Its hard enough for developers to predict the random an idiotic nature of the users of their software, now they have to build protection against hardware throwing them random results.
This instantly reminded me of a developer that claimed a 1200% improvement in performance after he optimized some code. The developer wasn't particularly skilled and some senior level guys had already optimized the performance about as far as it could be taken, so we were dubious. We found after a code review that basically this developer has improved the efficiency of the software by skipping some critical intensive calculations that was the point of the software.
Sure you could claim that this optimization is greener then the original code because the CPU is not working as hard, but if you are not going to get the expected results, f*ck being green.
I haven't thought of anything clever to put here, but then again most of you haven't either.
I've been interested in AI this month, and there's a really awesome application here for AI. Part of what makes people what we are is that we can't get certainty on a lot of our answers. Some guesses are good, some are spectacularly wrong. So if you build self awareness of that in the chip, it will need to use "damage control logic" to recover "socially" if it makes a mistake.
Same theme, we could get some really funny results when chips make that mistake and accidentally "get angry".
My first Journal Entry ever, in 8 years! http://slashdot.org/journal/365947/aphelion-scifi-fantasy-horror-poetry-webzine
Saves energy by not bothering to get things right!
They should name this chip "The Latino".
As much as good enough might save power, good enough doesn't cut it once you start using Currency. As long as these stay away from accountants and banks that's fine, but will they is the issue.
Call it the "Close enough for government work" chip.
It's still math, it's just in the hardware rather than the software.
I see lots of complaints, but this could be a fantastic option in GPUs. Does the calculation for the color of some pixel really need to be perfect. I'd certainly trade that for an increase in frames per second.
Video game graphics could probably benefit from this. Very few people will notice that one pixel is #FA1003 instead of #FC1102, especially when it's replaced 16ms (or, worst-case, 33ms) later with yet another color. It might actually make things "better" - making the rendering seem more analog. Many games are "wasting" power adding film grain or bokeh depth-of-field or lens flares or vignette, to try to simulate the imperfections of analog systems to try to make their graphics less artificial-looking. If you can get a "better" look while using *less* power, all the better.
Actually, I seem to recall hearing about this earlier. For some reason I want to say nVidia specifically has been looking into this.
Will CPU's come with +- ratings like Capacitors?
Game programmers have already been making use of approximate math to speed calculations, like http://en.wikipedia.org/wiki/Fast_inverse_square_root
It seems like a hardware implementation could be a nice win for certain applications, assuming it can be turned on and off. But, I feel the level of accuracy for each application is different, so it will be hard to balance general purpose requirements in a chip. The chip maker might need to implement several different approximations for each function. Is the increase in power efficiency worth the increase in chip size?
Don't we already have hardware-accelerated "inexact" math?
It's referred to as "Floating Point."
GLaDOS for President 2016! "Well here we are again. It's always such a pleasure." -- GLaDOS, 2011
It seems to me that this may make repeatability difficult. What if you want to recreate the situation for debugging, court cases, etc? Perhaps there can be a "testing mode" where full accuracy is on, but switch to "efficiency mode" for low-power production. Still, losing repeatability makes me noivus, to quote the 3 Stooges.
Table-ized A.I.
That's how my brane werks for spailing. Good enuf is good enuf becuz itz an effishent brane.
Table-ized A.I.
Picture they using this to iterate through a chaotic system's evolution, e.g. a climate model.
http://www.imho.com/grae/chaos/chaos.html
I went through high school math class giving incorrect answers because I was lazy.
Have gnu, will travel.
At last! Someone who finds this as confusing as parent, and fails to grasp the notion that these would be used in applications where accuracy is not important (but speed or power savings is), and not in processing financial transactions!
What!? That rocket was NO WHERE NEAR ME. Wait, why is everything FROZEN?!
Connection Terminated. Desynch error rate exceeded.
Oh sure we'll just snapshot the whole flippin' gamestate to the clients and do reconciliation -- But that's just wrong.
Error propagation, Non-determinism, etc etc. This is OK for GPU stuff that ONLY draws pixels. Anything that affects gameplay could only be done server side with dumb clients, but not for any real detailed worlds (just ask second life devs) -- Without deterministic client side prediction you need MUCH higher bandwidth and latency of less than 30ms to get equivalent experience. The size of game state in game worlds has been increasing geometrically (in PCs it still grows, consoles hit limits due to ridiculously long cycles and outdated HW), determinism and pseudo randomness helps keep the required synch state bandwith low. Oh, I guess I could use less precise computations for SOME particle effects (non damaging stuff), but you know what? I'M ALREADY DOING THAT.
What's that you say? The errors could be deterministic? Oh really... well, then what the hell is the point? Why not just use SMALLER NUMBERS and let the PROGRAMMER decide what the precision should be. It's like no one's heard of short int or 16bit processors. Give a dedicated path for smaller numbers, and keep us from being penalised when we use them (currently, 16 bit instructions are performed in 32bit or 64bit then trimmed back down). Some GPU stuff already has HALF PRECISION floats. Optimise that path and STFU about fuzzy math, you sound moronic...
As many have said below, your brain is indeed doing math - what it's not doing is "computation".
Most of the discussions in this thread are forgetting that important difference. The applications for which this type of chip will be useful are those in which the exact value of something is not important, but the relationships between values are. For instance, if you're implementing a control system algorithm, you don't care that the value of your integration is something specific, but you do care that it will always increase in proportion to the inputs and time. This is more akin to how your brain works - it doesn't care how much force it has to apply to your arm to make it move to catch a ball - it just knows that it needs "more" or "less".
For things like finance or engineering design that actually require computation this chip would be a poor choice.
"There are a dozen opinions on a matter until you know the truth. Then there is only one." - CS Lewis (paraprhase)
"Computer scientists have unveiled a computer chip that turns traditional thinking about mathematical accuracy on its head by fudging calculations." "Computer scientists" and software engineers deal almost exclusively in software. Computer Engineers are the one's who do the above tasks.
So my computer let me down when I was using a poorly written app on this new fangled computer. I am suing the computer manufacturer, the app maker and gosh darn it, Slashdot too for the entry
They come in the dark, only in the darkest.
I would think that for AI one would prefer precision. If AI uses inexact data, doesn't that just introduce "artificial stupidity"? We already have enough of the natural kind...
With those new chips, will it be VAXorcist, the Sequel?
cpghost at Cordula's Web.
My bank has has that capability for several years now, along with the cable company, electric company, ....
This will be the perfect chip to process polls, political speeches, economic analyses and penis measurements!
This is an old idea in the semiconductor industry. It comes from the fact that, on a typical IC, a few transistors will be marginal or defective. This causes rejects (or binning as a slower part) during final inspection. There's an existing market for DRAMs with a few bad bits. They go into telephone answering machines, and (I suspect) low-end TV sets.
Other schemes for dealing with this problem are to have some extra units on chip, and switch the bad ones out during final test. This is routinely done for DRAM, and for the Cell processor chips used in the PS3. (Only 7 of the 8 auxiliary CPUs in the Cell are live.) It's also possible to use error correction to fix up marginal RAM.
At the CPU level, architecting around errors is quite feasible. The UNIVAC I had that. So did many IBM mainframes, where everything was done twice and checked. But this was to catch rare errors, not frequent ones. The reaction to an error was a "machine check" interrupt, which generally meant killing a program or at least backing up to the last checkpoint.
Recovering from errors is complex. In theory, you could have something like multiple unreliable FPUs with checking, followed by a retirement unit that handled the discrepancies by backing up and redoing the computation. That would probably require about half an acre of additional grey cubicles at Intel in Santa Clara to get right.
I can see it now: "Whoops! We accidentally sent the wrong chips for the manufacturing of medical and military equipment. But we did increase our profit by $150,000 last quarter".
So..basically what we have here is a computer chip that is designed to be lazy?
I envision the "less precise" CPUs being used in consumer laptops where people are just watching movies or listening to music.
It does not matter if the MPEG4 conversion is slightly off with the color, because the consumer's eye won't detect it. The selling point will be a laptop or tablet that lasts 10x longer on a battery charge.
Exactly that.
Prepare to see GPU which go into "fudged mode" when dealing with graphics (3D, Video, etc.), and which go into "high precision mode" when doing science (OpenCL, CUDA, etc...)
Then further down the line, be prepared to see the "high precision mode" to be a paid-for only option.
(Buy a GPU marketed as tablet/latptop/entry-level desktop: Only "fudged mode available",
Buy a GPU marketed as high-level desktop/workstation/cluster: "High precision mode" available too, costs 2x more, although it's exactly the same chip (only perhaps with a different number of disabled/enabled core) )
That's already the case with other pro features:
- ECC mode is only availble on cluster OpenCL/CUDA cards (although they don't use ECC DRAM chips. Instead, they reserve a small portion of the memory to do checksumming in firmware/software). They are identic. Or in fact even cheaper (the graphic output is disabled or not even soldered-on).
- Quad-buffer stereo OpenGL is only available on "workstation"-grade cards, although there's no peculiar hardware requirement (and a subset of the same capability is available as proprietary gaming 3D-Stereo DX3D/OpenGL on some mid- and high-range models).
So, yeah, one more caracteristics that will be artificially price-tired through a pure software setting!
And one more opportunity for the open-source drivers to shine...
Well, except maybe they will lack the necessary man-power, due to the required additional reverse engineering, or due to the seldom needed feature.
(Although, we maight see a better chance with AMD hardware:
AMD supports the development of open-source drivers by providing documentation for almost everything (except Video DRM), and the computing part is recent enough (OpenCL was recently developped and is only on version 1.2) and relies on less quirks and optimisation than graphics: so performance shouldn't be lagging that much behind the closed source drivers.
When you also take into account that being open-source these drivers are easily packaged-with and maintained by distributions, thus making them a little bit easier to deploy (no need to add a manufacturer's 3rd party repository, no need to recompile a separate kernel module, etc. always compatible with up-to-date Xorg/Wayland API & ABI), we can expect the AMD hardware to see more open-source usage for computing, and thus the computing feature being more sought after and also developed by the opensource drivers).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
They could be useful in a few small circumstances, but for the vast majority of cases, I'd be interested in how a speed payoff is going to be beneficial given you don't know whether you got the correct answer.
You are not getting the "correct" answer from your current CPU. Floating point math is an approximation. You *may* get 15 or so digits of precision. For certain sequences of numbers and operations you may get far less. Plus there may be rounding errors as numbers are converted between binary (what the hardware floating point unit, FPU, uses) and decimal (what people normally use).
Here is an example. Try 0.5 - 0.4 - 0.1 in your favorite calculator app. You may not get zero, especially if the app naively uses the hardware FPU. This is why some calculator apps use decimal arithmetic internally. Doing so can also let the app be compatible with 64-bit math. The FPU in mobile devices usually is not.
The only thing interesting (to me) about this development is if the processors give the same output for a given input. If they do, then it's basically using the same principles as overclocking. What would be a much more interesting development would be to see a modular processor that streamlined itself to save power on frequently used processes and then distributed the remaining power to "harder" work.
Well, you can expect that the amount of fudging will be different for different type of video frame, or different part of the computation pipeline.
And that the error will be created in such a way as to be a small relative error (the wrong part being mostly in the less-significant bits) instead of a completly random error (any bit could be flipped, including the most significant part, or even the exponents).
Thus you'll get video noise (similar to the kind of picture degradation you could get by disabling post-processing or using fixed-point implementation), instead of random splashes of colour (similar to the king of degradation you could get with a packet error in the stream).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
In a way, yes, it is used in graphics hardware, though you generally get FULL precision for basic ops like +, -, *.
In the first generation of CUDA devices, even those where tiered. You got either fast 24bit-only integer ops, or full 32bits integer ops.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
As anyone who does numerical analysis knows, our chips now are accurate only to a certain precision.
Coulda sworn that HFT was going to make us millions.
Actually, if the chips aren't biased to make errors in some strange way, I guess it would balance out. OTOH, if you have some trading algo that looks for something like option pinning, and there is systematic bias around integers... oopsie!
I guess the HFTs will stick with numbers they can count on... unless they're greedy enough to think the power savings will be a good thing, and careless enough not to understand the nature of the error in the calculations. We could see a subcategory of HFTs that work well under these conditions... or we could just ban it because it's of no socially redeeming value.
How is this any different than the "fuzzy logic" concept that rears it's ugly head every 10 years or so?
Wasn't there an article somewhere mentioned on Slashdot about the brain, how it works and this kid from Africa who wants to create computers using techniques which animal brains use? Wish I could find it now. A good read and very thought provoking. This sounds like an application of the same ideas... ideas which come from our own animal brains.
Except that tomorrow that little hobby application is suddenly used in an engineering project
In which case it will be ran on a workstation instead of the smart phone it was developed on, which will either not have the "low-power fudged mode", or could atleast switch between "low-power" (say for desktop eye candy. Where TFA's 7.5% relative error is acceptable) and "high precision mode" (for any general purpose calculation done)
No bridge will fall.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
I can imagine a few cases when it could be allowed, based on mathematical proof in advance that error level would be acceptable.
Audio/Video playback in a noisy environment
Processing similar to PageRank and the recently announced NetRank for biochemical analysis might be able to produce better results for a given cost in electricity. In other words, deeper graph analysis traded for less significant digits
CPU-controlled activities that depend on statistics and sensors, for example street light control, voice/gesture based activation of lighting
Applications in which low power is the most important thing, especially if it is output meant for a human brain which already operates on a lossy basis. A wristwatch might be lower power if it is allowed to be correct within plus or minus 15 seconds.
My question is whether they have controlled for where the error occurs. The nice thing about approximations is that you know what the error is.
According to the article, the low power increase the relative error to 7.5% (quite huge) but reduce the power requirement 15x (massive benefits).
A possible explanation:
Some mathematical computation (like trigonometry) is done with lookup table and interpolation.
By using as simpler (like linear instead of polynomial)- or even doing away with- the interpolation step, you can quite speed up and lower the power requirement for corresponding ops.
By doing this you only increase the expected relative error. Not occasionnaly producing garbage.
Thus only get more approximative DCT step in you video decoding, and the output is more "blocky" (see the attached JPEG in the article).
Another explanation:
TFA speaks about reduced precision multiplication and addition.
So you could also use a simpler (but more error prone) circuitry for handling the least significant bits (TFA mention lower voltage).
If you can have bit errors anywhere including the MSB then you're going to be limited to situations where you don't actually care about the answer
Or situation where you don't actually need exactly 1 answer pro input, but where you somewhat statistically combine ("reduce") the output. (example: you only need an average of all results) and the b0rked-bit-flipped-results would be dropped with most of the other outliers.
You trade a loss of precision (the final mean will be done on less sample - you loose p.pp% of them as outliers) against a massive power requirement decrease (15x less power).
Again, that's not how the chip works.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
I've been doing a half-assed job for years.
I cannot wait until these chips start doing high frequency trading in the financial markets....
If telephones are outlawed, then only outlaws will have telephones.
Can't wait for someone to start using these chips in Point of Sale (you thought I meant something else, huh!) devices...
"Okay, you had a bowl of soup, $2.99, salad, $2.38, chicken fettuccine Alfredo, with sausage, $11.99, + $1.55, glass of white wine $3.99, with tax... comes to $48.50. Will that be cash, check, or charge?"
That isn't where you want to implement decimal math. For languages, decimal representation and math should be provided by libraries, simply because anything that can be shipped out into a library without significantly reducing efficiency or code readability should be (to reduce unnecessary language complexity).
As for platforms, I'm not sure what you mean. That word has many meanings in computing, but IMHO none of them should care about decimal math.
I am the other way round. What bothers me is architectures that DO provide decimal functionality. It is a total waste of silicon and/or ASM instruction bits to provide something that can be done far faster in binary with no loss of accuracy (compared to native decimal, if done right). Any decent decimal library will internally be binary anyway, not BCD or similar.
You agree with me.
The chips are 15x as efficient, they calculated it themselves.
Intel developed inexact math calculation technology in 1995 for Pentium. Now that the patent has expired, everyone wants to copy it.
What kind of crap is this? This is not new, way too short floats and/or fixpoint math are core technology for anyone who programmed in the previous century.
Every GPU on earth already does this while rendering. Nothing new here, move along.
Weather prediction
Political polling
E-voting
Advertising science-y parts
Sports statistics
Boy, the list is endless!
For example, use analog circuitry for FP operations and do an ADC at the end only.Sometimes just a few transistors in analog mode can replace thousands of digital ones.
I know that the 8087 was culturally more significant because of the relative penatration of the 8086 and the 68000, as well as the perception gap between the two. Also, the 68882 was not the only other math coprocessor, just like the 68K was not the only option to the 8086.
(Hmm. The current wikipedia article on math coprocessors has some problems, as can be seen in the fact that "math coprocessors" re-directs to "Floating-point Unit".)
I would hate to see the younger generation losing opportuniites to understand that there were plenty of valid options to the 8086.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
So medical applications on tablets and phones give wrong answers. Okay?
And how about some idiot thinking a phone number will fit nicely into a double float?
If I want low power at the sacrifice of accuracy, I'll use something like an S08 or a FORTH processor, and fixed point math.
Fixed point math, you can at least control.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
Games, well, games are always okay with inaccuracy. Vocal codecs? I suppose you could force it to work.
But other solutions already exist, are cheaper, use less power, and behave more predictably.
Speaking of intel, they tried this same path about fifteen or so years ago, and I'm not talking about the FPU errors that have already been mentioned.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
I have dim memories of intel touting research in this area about fifteen years ago (plus or minus a couple). And of thinking it wasn't new back then. And it didn't get very far then (in spite of people saying, "Monte Carlo!").
Now we have even more better solutions for when accuracy doesn't count.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
We already have plenty of cheap, way-low-power solutions for all of these.
Fifteen years ago, in fact, this idea was not new, and the options available then were way better on price, power, speed, debuggability, everything that counted.
This idea was innovative about once, back in the mid-eighties, when TI and Motorola (among others) first started producing signal processors.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
But, yes, since back before the '60s. And microcontrollers doing DSP since the '70s.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
Getting all sorts of details wrong and ending with the right conclusion.
If you want low power DSP, we have low power DSP. If you want a graphing calculator with an OS, we can do it now with a cold fire or ARM integrated CPU running on ordinary batteries, if we are willing to do the work.
While the power use analysis work is definitely going to be useful, the CPU they are trying to sell is a boondoggle.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
Analyzing the power use in digital ICs is very useful.
The CPU they are trying to sell is a boondoggle. Shame to see the Indian government treating it seriously.
Case of engineers not understanding what they are working on, getting all focused on the initial target, not recognizing the far greater value in the tools they built trying to get to a wrong target.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
Try to fit Apple's budget in a C double float.
Or Microsoft, Facewhatever, Google, etc.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
I'm not a fan of fixed-width floating-point, but the current floating point stuff is fairly well characterized.
This is going to introduce a new kind of bias to the errors, and the bias always adds up. Sure, we don't need more than 6 digits for a lot of stuff, but the way we generally do fractional math, we use extra accuracy to buffer the result from the cumulative bias.
These chips are going to bring new bias profiles, and it will likely take a while to get a clear picture of the profiles, and then you have to start programming against (yet another) profile.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
I don't think they are talking about variable fudging.
They are talking about altering the geometries of the circuits -- transistors, gates, and some whole logic sections removed. You're not going to be putting those back in at run-time.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
You go to the friendly article, and one of the places they are trying to sell this is India, where the OLPC is too expensive, takes too much power.
You also get a good look at how far off the "MPEG-4" conversion is going to be. It's not pretty.
The power-reduction techniques are interesting, I'll grant that. But I think everyone trying to do a one-size-fits-all capture of India should just take a step back and give it a second thought. A graphing calculator running netBSD would go a long way as in intermediate step towards the goal without having to subject an entire country's lower classes to enforced inaccuracy.
And I have to imagine the floating point errors would impinge the encryption.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
Where not only do we dumb-down the people, we dumb-down the computers!
I must have missed that scene in Idiocracy...
òÓ,)_\,,/
Ha! All you Gentoo haters mocked my use of ffast-math in everything as being dangerous and unsafe but now they're doing it in HARDWARE!
HA! I say! Who's right now bi-SR$%T%&YENHLKFGLDHIH
you have problems
Is it 15 times more efficient, or about 15 times more efficient?
120 characters ought to be enough for anyone
Was initially responding to my memories of past attempts in the direction of non-exact processing.
Having read the friendly article now, the research on power reduction in general may be useful. From the summary, I can't be sure, but it looks possible.
But, if I'm going to screw up the floating point for power, I'd just as soon go to fixed point or mixed fractions on a real low power CPU with known behavior.
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.