Posts: 607
Threads: 66
Joined: Jun 2013
I stumbled upon this yesterday, a novel way to construct the logic elements of a computer.
10^17 gates operating at 10^10 hertz gives you 10^27 bits-per-second in a processor volume of 1 cubic centimeter (not counting the resonance cavity and the machinery to achieve the cryogenic temperature). Not bad, not bad at all.
Posts: 607
Threads: 66
Joined: Jun 2013
06-26-2013, 02:25 PM
(This post was last modified: 06-26-2013, 02:27 PM by JohnnyYesterday.)
At the energy dissipation rate of 10^-27 joules per logic operation, times 10^27 operations per second, gets you a total power consumption of 1 joule-per-second, or 1 watt. 1 watt! That's 1/20th of what our meat brains run on to achieve a paltry 10^18 operations per second.
10^27 divided by 10^18=1,000,000,000. So a cubic centimeter of helical logic-based computronium is 1 x 10^9 times faster, while simultaneously being 20 times more energy efficient, than an average human brain, which has a volume of 1000+ cm^3.
That's not counting the refrigeration requirement, of course, but with vacuum flask insulation and futuristic refrigeration technology, I'd conservatively guess a few tens of watts could take care of that in baseline-shirtsleeve-environments. Of course in deep space at 30 Kelvin, or so, little more than passive radiator panels would be required.
Posts: 194
Threads: 23
Joined: Mar 2013
Posts: 11,753
Threads: 454
Joined: Apr 2013
06-27-2013, 12:55 AM
(This post was last modified: 06-27-2013, 12:57 AM by stevebowers.)
Reversible computing can in theory operate with fantastically low energy requirements, but (if I understand it correctly) the type of computing that can be done reversibly is limited in nature. If you want a computing system that can perform an unlimited range of operations you have to use irreversible computing as well, or instead.
On the other hand, I see that Merkle and Drexler state that Charles Bennett has 'has proposed very general methods for converting arbitrary irreversible computations into time and space efficient reversible computations.' If this is true of the sorts of computations that archailects want to make, then maybe they would prefer to use reversible computing for most, or all processing.
Posts: 607
Threads: 66
Joined: Jun 2013
(06-26-2013, 04:11 PM)Tachyon Wrote: That sounds a bit high.
General physical limits gives about 10e22 bits/s-cm^3, see:
http://www.orionsarm.com/eg-article/48507b746e356
http://scorevoting.net/WarrenSmithPages/...works.html
Those appear to be talking about irreversible computation, so their figures aren't applicable here since reversible computation is thermodynamically reversible.
That's why it is so efficient, and doesn't turn into a hot plasma from its own waste heat.
I know it's an appeal to authority (in this case relevant authorities), but I tend to think Ralph Merkle and Eric Drexler get their stuff right.
Posts: 7,362
Threads: 297
Joined: Jan 2013
(06-26-2013, 02:25 PM)JohnnyYesterday Wrote: At the energy dissipation rate of 10^-27 joules per logic operation, times 10^27 operations per second, gets you a total power consumption of 1 joule-per-second, or 1 watt. 1 watt! That's 1/20th of what our meat brains run on to achieve a paltry 10^18 operations per second.
10^27 divided by 10^18=1,000,000,000. So a cubic centimeter of helical logic-based computronium is 1 x 10^9 times faster, while simultaneously being 20 times more energy efficient, than an average human brain, which has a volume of 1000+ cm^3.
That's not counting the refrigeration requirement, of course, but with vacuum flask insulation and futuristic refrigeration technology, I'd conservatively guess a few tens of watts could take care of that in baseline-shirtsleeve-environments. Of course in deep space at 30 Kelvin, or so, little more than passive radiator panels would be required.
Be very cautious about estimates for the human brain's capabilities in terms of computer science nomenclature. Not enough is understood about how the brain works to make a reasonable assertion at this time (and I am aware that you can find all sorts of estimates everywhere based on anything from action potential rates to extrapolated neural circuitry in the eye)
Posts: 16,264
Threads: 739
Joined: Sep 2012
Some thoughts here:
First, the 10^27 bps number is impressive but I don't think it violates anything physics wise because basically what is being described is parallel processing. You could take a 1 bit/sec processor and if you ran 10^27 of them you'd have 10^27 bit/sec processing going on. In practice it appears that in terms of actual bit flipping or 'speed' its only operating at about 10 gigahertz. With the right programming you can take advantage of that 10e27 bps, true, but I don't think you're dealing with physics issues in the same way as you do with an Ultimate Chip (hope that makes sense).
Second, Johnny, you start out talking in terms of bits/sec, but then switch to operations/sec. Isn't this comparing apples and oranges? I would assume it takes more than one bit to perform even the simplest operation. So that's going to reduce that impressive number a good bit (ha! a pun). Still impressive, but let's get our terms straight. We've run into this issue in past discussions.
It seems to me that comparing the Ultimate Chip and this device is also somewhat like comparing apples and oranges. The UC is intended to operate at 'room temperature' pretty much while this tech is described as operating at cryogenic temps from what I can gather (haven't had time to read the whole article, sorry). And I also recall Adam stating when the UC was first introduced that it is the best possible within the temp range described and using phonons since they are most efficient in these conditions. That certainly leaves the door open to many other devices that might operate under very different conditions. A super-cooled or actively cooled UC might be much more powerful, although such hasn't been described yet.
To date, we do NOT have reversible computing really addressed in the setting, mainly because we haven't been able to find much solid info on it. If it can be made to work it may have a major impact on computation. That said, one thing I seem to recall is that it is overall somewhat slowed down by the need to perform all its calculations in reverse in order to 'cancel out' the entropy or energy dissipation from doing them initially. So this may also slow things down overall a bit. Although maybe this tech gets around that somehow.
Re brain - current OA convention is that a human brain does about 1e18 bps in processing, which puts it about 1000x above the number that was mainly being thrown around for this when we created the brain size articles and such. Given how little we know about the brain, including how optimized it is to do whatever its doing, this number seems as reasonable as any other to me. I see no reason to change it.
Ok, all I've got for now. Need to get to bed.
Thanks!
Todd
Posts: 607
Threads: 66
Joined: Jun 2013
(06-27-2013, 01:41 PM)Drashner1 Wrote: Second, Johnny, you start out talking in terms of bits/sec, but then switch to operations/sec. Isn't this comparing apples and oranges? I would assume it takes more than one bit to perform even the simplest operation. So that's going to reduce that impressive number a good bit (ha! a pun). Still impressive, but let's get our terms straight. We've run into this issue in past discussions.
The paper says logic operations. I assumed, perhaps incorrectly, that by operations they were referring to the switching of binary logic gates, which logically have to be in one of two states.
The operating temperature is most certainly cryogenic: 1 Kelvin.
Can't get much colder than that.
Posts: 7,362
Threads: 297
Joined: Jan 2013
Regarding the cryogenic requirement; how early in the setting would it be possible to have a domestic device capable of cooling a Helical Logic Cube to 1 degree? Not that everyone would require one but I'm wondering how long after a successful lab prototype could migrate into the public sphere.
Posts: 620
Threads: 23
Joined: Mar 2013
(06-28-2013, 07:00 AM)Rynn Wrote: Regarding the cryogenic requirement; how early in the setting would it be possible to have a domestic device capable of cooling a Helical Logic Cube to 1 degree? Not that everyone would require one but I'm wondering how long after a successful lab prototype could migrate into the public sphere.
A device for cooling small (5-atom) molecules using electric fields is described in http://physicsworld.com/cws/article/news...is-no-myth so such a device is certainly do-able and seems amenable to mass production. However, there is a question of marketability. What use would the average zar have for such extreme cooling, particularly those who could simply place whatever needed cooling into a shady area on the outside of their habitat? If the technology has only a very limited market appeal, it is likely to remain a niche product at best and so remain relatively rare and expensive. I find it difficult to imagine the need for a domestic cooler that can refrigerate items down to one Kelvin.
Radtech497
"I'd much rather see you on my side, than scattered into... atoms." Ming the Merciless, Ruler of the Universe
|