> We use a logic primitive called the adiabatic quantum-flux-parametron (AQFP), which has a switching energy of 1.4 zJ per JJ when driven by a four-phase 5-GHz sinusoidal ac-clockat 4.2 K.
The landauer limit at 4.2K is 4.019×10^-23 J (joules). So this is only a factor of 38x away from the landauer limit.
It's not about clock speed per se. It's about the lowest possible energy expenditure to erase one bit of information (or irreversibly destroy it by performing a logical operation). The principle comes about from reasoning about entropy loss in said situations. There's a hypothesized fundamental connection between information and entropy manifest in physical law. The idea is that if you destroy one possible state of a system, you have reduced the entropy of that system, so the 2nd law of thermodynamics implies that you must increase the entropy of the universe somewhere else by at least that amount. This can be used to say how much energy the process must take as soon as you choose a particular temperature.
This applies to any irreversible computation.
IMO, The fact that it's only 38x the minimum is MIND BLOWING.
> I wonder how much that speeding ticket would cost.
I once got one for going at about 3x the speed limit (very nice road in Brazil, broad daylight, nobody in sight, short run, and very unfortunate radar gun positioning). The policeman was impressed and joked that he would like to, but couldn't give me 3 speeding tickets instead of one.
There are some places where the cost of the ticket goes up the higher you are above the limit. Assuming it’s a school zone in one of those areas (pick whichever one you can answer for), what would the cost be then for going 28.4 mkm/hr?
I think they were alluding to the fact that it is impossible to measure the one-way speed of light, and even the definition is an assumption based on the two-way speed
Since the one-way speed is impossible to measure can we just define it to be half of the two-way speed? There's no way to tell if that's correct or not but it makes no difference in any measurements you could make. So maybe for convenience we can just choose the one-way speed to be half the two way speed to simplify the calculations.
Is there an idea of entropic potential energy/gradient/pressure? Could you differentiate encrypted data from noise by testing how much energy it requires to flip a bit?
No, because the energy of the system isn't related to the order of the underlying data, it's related to the changes that happen to the underlying data. If you have 5 bits and flip 3, it takes the same energy regardless of if the 5 bits have meaning or not. This is speaking in terms of physics. There obviously could be some sort of practical side channel attack based on error checking times if this was an actual processor.
Only if you were measuring a system with access to the key material and doing something with the plaintext, in which case this would be a side-channel attack (and an already-studied one). The whole point of encryption is that the data output is indistinguishable from noise without knowing the key.
Note that the gates themselves used here are reversible, so the limit shouldn't apply to them. But the circuits built from them them aren't reversible as far as I can see in the paper, so it would still apply to the overall computation.
The interesting thing about this is that if we get close to the Landauer limit we may have to seriously start thinking of using reversible computing[1] paradigms and languages to get optimal performance.
Not sure what you're trying to say. Room temperature is 293K, and at temperature the Laundauer limit is 2.8E-21 J. The chips in the article were cooled to 4.2 K, and at that temperature the Laundauer limit is 4.0E-23 J.
That number checks out to be accurate to just over 4 significant digits. Divide the 4.019×10^-23 J by 4.2 K and ln(2), and you get a result just a hair smaller than the Boltzmann constant.
Unless the formula at [0] is wrong, in which case my calculations would also be wrong.
Given the cooling requirements, I suppose it would create completely impassable rift between datacenter computing and other kinds. Imagine how programming and operating systems might look in a world where processing power is 80x cheaper.
Considering that "data centers alone consume 2% of world's enegy", I think it's worth it.
It seems likely that the more efficient our processors become, the larger share of the world's energy we'll devote to them [0]. Not that that's necessarily a bad thing, if we're getting more than proportionally more utility out of the processors, but I worry about that too [1].
>Not that that's necessarily a bad thing, if we're getting more than proportionally more utility out of the processors
The trend seems to be that we get only a little bit of extra utility out of a lot of extra hardware performance.
When the developer upgrades their PC it's easier for them to not notice performance issues. This creates the situation where every few years you need to buy a new PC to do the things you always did.
Is that the case for our highly centralized clouds? No one's putting a liquid nitrogen cooled desktop in their office so this type of hardware would be owned by companies who are financially incentivized to drive down the overhead costs of commoditized functionality like networking, data replication and storage, etc. leaving just inefficient developer logic which I assume is high value enough to justify it.
It's quite common in the cloud industry to trade hardware for shorter development cycle these days too. I think that's because there is still very high growth in the sector and companies all want to be the first offering feature X.
As cloud services become more mature I expect people will be more cost sensitive. Though when that will happen I cannot say.
I'm sure a lot of developers upgrade their PCs (they were called workstations at a time) because of material problems - keyboards getting mechanically worse, and laptops can't easily get keyboard fixed, screens getting dead pixels, sockets getting loose, hard-to-find batteries getting less charge, and maybe some electronics degradation.
Another reason is upgrades to software, which maintain general bloat, and which is hard to control; new hardware is easier. That's however is very noticeable.
On top of that, just "better" hardware - say, in a decade one can have significantly better screen, more cores and memory, faster storage; makes easier for large software tasks (video transcoding, big rebuilds of whole toolchains and apps, compute-hungry apps like ML...)
> Not that that's necessarily a bad thing, if we're getting more than proportionally more utility out of the processors, but I worry about that too
I have two points to comment on this matter.
Point 1: The only reason I would worry or be concerned about it is if we are using terribly-inefficient programming languages. There are languages (that need not be named) which are either 3x, 4x, 5x, 10x, or even 40x more inefficient than a language that has a performant JIT, or that targets native code. (Even JIT languages like JavaScript as still a lot less efficient because of dynamic typing. Also, in some popular complied-to-native languages, programmers tend to less efficient data structures, which results in lower performance as well.)
Point 2: If the inefficiency arises out of more actual computation being done, that's a different story, and I AM TOTALLY A-OK with it. For instance, if Adobe Creative Suite uses a lot more CPU (and GPU) in general even though it's written in C++, that is likely because it's providing more functionality. I think even a 10% improvement in overall user experience and general functionality is worth increased computation. (For example, using ML to augment everything is wonderful, and we should be happy to expend more processing power for it.)
Both are important and 1 can be even more useful. Use 1 to easier build very complex systems. Once they are working and you are selling them - optimize. Without#1 you can't get #2
Nuclear fission has well-known drawbacks and risks that are fundemantal, they will never be engineered away (risk of catastrophic explosion, huge operating costs, risk of nuclear waste leakage, risk of nuclear weapon proliferation). Why do you think the future will be significantly different from the present in this regard?
These risks will most likely only reduce over time as our science and knowledge develops, barring a world-affecting mistake or war. If we can make a ton of energy, we can do many other things like launching the waste into outer space, and having a positive ROI by making more energy than it cost.
Considering that many modern interfaces are somehow less responsive than ones written over 20 years ago even when running those programs on period hardware, I feel certain that you are right.
This is a very complex topic and there's a bunch of reasons for that.
And no, software efficiency isn't even the main factor. Not even close.
Just a few pointers: polling on peripherals instead of interrupts (i.e. USB vs. PS/2 and DIN) introducing input lag, software no longer running in ring-0 while being the sole process that owns all the hardware, concurrent processes and context switches, portability (and the required layers of abstraction and indirection), etc.
It's a bit cheap to blame developers while at the same time taking for granted that you can even do what you can do with modern hard- and software.
Everything comes at a price and even MenuetOS [1] will have worse input lag and be less responsive than an Apple II, simply because you'll likely have USB keyboard and mouse and an LCD monitor connected to it.
Apple II? I'm talking about things like VS6 vs modern VS. Even when performing the same tasks, which means modern VS has no reason to be doing more work than VS6, the modern one is less responsive. There are several videos on youtube demonstrating the dramatic difference.
Well 20-ish years ago when I was working with VS6, it ran like dog shit on contemporary hardware.
Big projects would pretty much kill it and the occasional crash was to be expected.
Sure, firing up VS6 on more modern hardware let it fly by comparison, but then again its features paled in comparison to those available with modern VS.
On the other hand, I don't use VS anymore, since VSCode is all I need and runs faster than VS6 back then (even on my 5 year old mid-range laptop) so no complains there.
Some things have improved a lot. Remember how long it took to boot a 2009 PC. I suspect if hardware perfromance stagnates then software optimisation will develop again.
Swap your OS's and you'll still notice an improvement in the hardware. Windows has always been slow. SSDs have made even windows fast to boot on every recent laptop I've used though.
It takes things way beyond simply "emulating Javascript in Javascript", yet is presented so well that you barely notice the transition from current (2014) reality to a comically absurd future.
Solid state physics begets both cryogenic technology and cryocooling technology. I wouldn't write off the possibility of making an extremely small cryocooler quite yet. Maybe a pile of solid state heat pumps could do it.
Really. Vacuum casing is not even close to sufficient to set heat absorption to zero because of thermal radiation.
And you can't just make the walls reflective once the cold object gets smaller than the wavelength of the radiation. The colder the object, the longer that wavelength.
The way it works is that the entire assembly is in a vacuum. It kinda has to be as any gas which touches it will instantly condense to it or freeze to it. You then have a dual cryostat of liquid helium and liquid nitrogen cooling down the assembly (within the vacuum). The helium and nitrogen cryostat also have a vacuum shield. The nitrogen (liquid at 77K) is a sacraficial coolant which is far cheaper than liquid helium (liquid at 4K) that you need to get to these temperatures. Your're right that thermal radiation is an issue so you have to be careful with the placement of any windows or mirrors around the device.
Souce. I have a PhD in physics where I used equipment cooled to 4K.
Great, then we both have physics PhDs, and you'll know that none of that equipment has, or easily could be, sufficiently miniaturized, which is the topic of discussion ("extremely small cryocooler"). You can't put nested closed dewers of liquid nitrogen and helium on a O(1 mm^2) microchip, and the reason is exactly what I said: it will warm up too fast.
The topic is cooling small objects so that personal electronics (e.g., your phone) can compete with datacenters. Cold at scale (i.e., in datacenters) is comparatively easy.
It also doesn't change the fact that smaller devices are harder to put wires on - but they're both polynomial scaling factors that other polynomial scaling paradigms could cancel out.
The topic of discussion is datacenter vs. an extremely small cryocooler. What is the other polynomial scaling paradigm that would cancel out the datacenter's advantage?
The effectiveness of thermal insulators and heat pumps. If they make the cost difference negligible then other costs will dominate. We're talking about future decades of material science advancements here, a lot is possible.
I don't see an impassible rift. Probably at first, but supercooling something very small is something that could certainly be productized if there is demand for it.
I can see demand in areas like graphics. Imagine real-time raytracing at 8K at 100+ FPS with <10ms latency.
Moore's Law was originally about the number of transistors per unit area and that has absolutely stalled. They are comparing 72 and 64 core processors to single core ones to try to make the claim that "transistors per chip" are doubling in accordance with Moore's Law (which they still aren't--you can plot a line through and see that we are falling short of linear on a log scale). Are you really arguing that simply making microprocessors larger, ignoring cost and density, is an improvement in processing power (especially for the average end user, who is on a machine with only a handful of cores)?
Look at the actual transistor sizes. In 2009 we were at 32 nm. We're now in 2021, so if transistor sizes had kept halving every two years we would be at 0.5 nm. Clearly, we are not anywhere close to that--we're off by a factor of 10, and that's only with the very latest and greatest manufacturing processes that almost no consumer chips use (not to mention that the 5nm process used by AMD is not the same as a 5nm process used by Intel). As the article itself notes:
> Microprocessor architects report that semiconductor advancement has slowed industry-wide since around 2010, below the pace predicted by Moore's law.
Of course transistor companies are happy to claim that they are secretly keeping pace, but in terms of commercially available microprocessors it is unquestionably false. Anyone using the "doubling every two years" approximation to decide how much more computing power is available now than 10 years ago, or how much more will be available 10 years in the future, is not going to arrive at correct figures.
Moore's law was stated originally as the doublinkd of the number of transistors per integrated circuit per year, later modified by him to a doublingevery two years. Reading the article where he first proposed the rule (cited in the wikipedia article linked in parent) this confirms that definition, as opposed to transistors per unit area or other measures.
No, it was per area per economic unit originally, and the definition was changed in ways that don't make a lot of sense. Clearly the performance of a 72-core machine has little relevance to personal computing.
Not commenting on Moore's law, but I don't see how the last part is true when GPU have become a main component of personal computing. They are in fact becoming more and more important with the need for AI inference and always more demanding rendering tasks (both in resolution and frame rate).
The vast majority of software isn't written for the GPU. That's still true today, and it will likely be true in ten years as well. Unless we reach a point where that changes (and most software can fit its restricted paradigm), we should compare apples to apples. If we were talking solely about GPU performance, I would be more inclined to agree with claims about "80x more processing power than 10 years ago" etc, though.
Apple GUI frameworks enable them to rely heavily on the GPU, and the smooth and slick GUI are definitely a big part of their value proposition. While I agree that the business logic of most apps aren't using the GPU for compute, I think most app will have a rendering side where it is important.
Moore's laws has nothing to do with how fast a chip is. It deals with how many transistors you can fit in a given area.
This can equate to a faster chip because you now can do more at once. However, we hit the frequency limits a while ago for silicon. Particularly, parasitic capacitance is a huge limiting factor. A capacitor will start to act like a short circuit the faster your clock is.
Moore's law has a little more life, although the rate seems to have slowed. However, at the end of the day it can't go one forever you can only make something so small. One gets to a point they have so few atoms to constructing something useful becomes impossible. Like current transistors are finFETs because the third dimension gives them more atoms to reduce leakage current, compared to the relatively planar designs on older process nodes. However, these finFets still take up less area on a die.
Since it can't go forever - it is time to update its definition to reflect what we are doing with computing now - adding cores and improving energy efficiency. No?
adding more cores requires moores law. When we hit the end the only to get more cores will be either larger dies or more dies.
Moore's law does help with efficiency to some degree, small transistors generally require less power to switch. However, most the power is last due to the miles of wiring in a modern chip when running at such high clock cycles. Again, it's the parasitic capacitance.
It’ll all be wasted. When gasoline prices plummet, everyone buys 8mpg SUVs. If power & performance gets cheaper, it’ll be wasted. Blockchain in your refrigerator.
As processing power cheapens, programmers will settle for lazy and inefficient code for everyday consumer applications. It will be easier to be a programmer, because you can get away with writing shitty code. So wages will fall and the prestige of being a software developer wanes. The jobs requiring truly elite skill and understanding will dwindle and face fierce competition for their high pay.
Before this happens, I recommend having your exit strategy for the industry, living off whatever profits you made working as a developer during the early 21st century.
I’m less pessimistic. Even if CPUs are 10x faster than they are, that still opens up more opportunities than what can be “absorbed” by less efficient coding/masses. There will always be demands for eking out more of the available processing/compute power and doing so will always be a difficult task. For example, today you can edit large scale videos in real-time and view various movie-quality SFX applied real-time on a consumer desktop. More computing power = more ability to do things cheaply that were in feasible/impossible before. You’re limited by your imagination more than anything.
What’s truly more of a threat is AI-aided programming if that ever becomes a thing. Again, I’m not worried. The gap between telling an AI “do something that makes me $1 billion dollars” and “write a function that has properties x/y/s” or “do this large refactor for me and we’ll work together on any ambiguous cases”, is enormous. So you’ll always have a job - you’ll just be able to do things you couldn’t in the past (it’s questionable whether an AI can be built that generates programs from vague/poorly defined specs from product or even that generates those specs in the first place.
As an obvious counter example to your theory, we have CPUs that are probably 10000x more powerful than in 1980 (actually more if you consider they have processing technologies that didn’t even exist back then like GPUs and SIMD). The software industry is far larger and devs make more individually.
Technically SIMDs and GPUs existed back then but in a much more immature form, being more powerful, cheaper and widespread today than what was available in the 80s.
Processing power has cheapened exponentially for the last 50 years (Moore's law). I am skeptical that a continuation of this trend will drive a fall in wages.
In my own experience performance optimization is an important but infrequent part of my job. There are other skills an elite programmer brings to the table like the ability to build a mental model of a complex system and reason about it. If downward pressure on wages occurs I think it will be for another reason.
I think in general you are right, however, there will certainly be sectors where it will be a lot easier to just throw it at a wall of computation than pay someone to think about it.
But architecting complex systems so that they are maintainable, scalable, and adaptable... there's not gonna be enough cheap computation to solve that problem and omit top talent for a long time.
Energy cost and thermal cooling place restrictions on the computations. C++, with all it flaws, stays in use because it does give control over performance trade-offs.
Hardware efficiency is only relevant for programming effort today in a few industries, and even there, only in a few specific applications.
If you gave me a processor that could run instructions instantly, my product's release would at best be brought forward 1-2 weeks.
The largest efforts in programming are to do with translating requirements into code. Efficiency is a part of that, and there are obviously problems where it dominates, but there are many other difficulties even when it isn't.
Nice, but requires 10 K temperature - not very practical.
Once this can be done at the temperature of liquid nitrogen, that will be a true revolution. The difference in cost of producing liquid nitrogen and liquid helium is enormous.
Alternatively, such servers could be theoretically stored in the permanently shaded craters of the lunar South Pole, but at the cost of massive ping.
If the throughput is fast enough 3+3=6 seconds latency doesn't really sound that bad. There are websites with that kind of lag. You can't use to build a chat app, but you can use it as a cloud for general computing.
Fun aside I learned about recently: we don't actually know if the speed of light is the same in all directions. So it could be 5+1=6 seconds or some other split.
Yes! In uni I had this idea for a program that would triangulate where lightning struck by having multiple computers across campus listening for thunder, then measuring the difference in arrival times of the thunder. Like GPS in reverse.
Soon I realized I’d need the computers’ clocks to be as in-sync as possible, which in turn requires one-way latency calculations. I spent an hour diagramming with pencil and paper until I convinced myself it was mathematically impossible or I wasn’t clever enough to find the solution.
Looking back, I was more worried than I should’ve been about this (wired latencies are usually far less than 5ms, which for sound is ~1.5m) and less worried than I should’ve been about the fact that thunder isn’t anything close to a point source of sound.
Wired latencies are more limiting than you would expect! Light in glass in 2/3rds of that in a vacuum. At 2*10^8 the longest straight arc on earth would take 100ms to traverse in fibre optic cable.
Latency caused by the speed of light is a tangible and impactful factor in our lives!
It's amazing that something so incomprehensible (3x10^8 m/s) can actually be experienced.
That price is more than a decade out of date. Helium has been about 10x that the past half decade. I used to pay about $3,000 per 100L dewar a few years ago. Sounds like that price was still common in 2020: https://physicstoday.scitation.org/do/10.1063/PT.6.2.2020060...
Plus, liquid helium is produced as a byproduct of some natural gas extraction. If you needed volumes beyond that production, which seems likely if you wanted to switch the world's data centers to it, you'd be stuck condensing it from the atmosphere, which is far more expensive than collecting it from natural gas. I haven't done the math. I'm curious if someone else has.
That's the cost for one-time use of the helium. If you're running a liquefier the cost is much lower, since you're recycling the helium, but it still "costs" ~400W to cool 1W at 4K.
In other words, the helium is used to move heat around, and isn’t lost. It’s like the refrigerant in your air conditioner, except that it needs several stages of pre-cooling.
Initial cost is high but running it should mostly just incur energy costs if it’s well designed.
Edit: Didn't read the OP carefully... Am idiot. Anyway, maybe someone reads something new to them.
Nitrogen won't get you below 10°K, tho. It's solid below 63°K (-210°C).
You know things are getting expensive, when superconductors are rated "high temperature", when they can be cooled with LN2...
Helium (He2) is _practically finite, as we can't get it from the atmosphere in significant amounts (I think fusion reactor may be a source in the future), and it's critically important for medical imaging (each MRI 35k$/year) and research. You also can really store it long term, which means there are limits to retrieval/recycling, too. I sincerely hope we won't start blowing it away for porn and Instagram.
I'm no physicist, but wouldn't you need some kind of medium to efficiently transfer the heat away?
On the moon you have no atmosphere to do it with radiators with fans, so I gues you would have to make huge radiators which simply emit the heat away as infrared radiation?
> On the moon you have no atmosphere to do it with radiators with fans, so I gues you would have to make huge radiators which simply emit the heat away as infrared radiation?
Exactly. You can still transport the heat efficiently away from the computer using heat exchangers with some medium, but in the end radiators with a large enough surface area will be required.
Works well enough on the ISS, so I imagine it'd work just as well on the Moon.
I'm also not a physicist, but for the fun of the discussion...
Radiative heat loss scales with the fourth power of temperature. I don't know what temperature the ISS radiators are but suppose they are around 300K. Then I think the radiative surface to keep something cool at 10K would need to be 30^4, or 810000 times larger per unit heat loss. So realistically I think you would need some kind of wacky very low temperature refrigeration to raise the temperature at the radiator, and then maybe radiate the heat into the lunar surface.
Doubt if 80x difference would make it attractive. If it were 8000x then maybe.
And that only if you use the soil for cooling, which is non-renewable resource. If you use radiators, then you can put them on a satellite instead with much lower ping.
"Oh no, we bricked a lunar computer! Go grab your pressure suit, Mike! Back in a week, darling... Tell your mother I won't be attending her birthday party."
It'll be interesting to see if the cryptocurrency mining industry will help subsidize this work, since their primary edge is power/performance.
During stable price periods, the power/performance of cryptocurrency miners runs right up to the edge of profitability, so someone who can come in at 20% under that would have a SIGNIFICANT advantage.
In this paper, we study the use of superconducting technology to build an accelerator for SHA-256 engines commonly used in Bitcoin mining applications. We show that merely porting existing CMOS-based accelerator to superconducting technology provides 10.6X improvement in energy efficiency.https://arxiv.org/abs/1902.04641
Looks like the admit it’s not scalable and only applies to workloads that are compute heavy, but a 46x increase over cmos when redesigning with an eye for superconducting env optimizations
Even if power is free, you still get only a limited amount of it to turn into hashes. Power sources like plants or dams only produce so much. Even if you have bribed a corrupt grid operator to give you 100% of the output from a X megawatt coal plant for free, you're still limited to X megawatts of hashing. If you can turn that X megawatts into 10*X megawatts' worth of your competitors' hashing, well, you just 10xed your profit.
If something like that happens it will have far reaching consequences IMO. I'm not pro blockchain.. but the energy cost is important and it goes away significantly people will just pile 10x harder on it.
Not a physicist so I'm probably getting different concepts mixed up, but maybe someone could explain:
> in principle, energy is not gained or lost from the system during the computing process
Landauer's principle (from Wikipedia):
> any logically irreversible manipulation of information, such as the erasure of a bit or the merging of two computation paths, must be accompanied by a corresponding entropy increase in non-information-bearing degrees of freedom of the information-processing apparatus or its environment
Where is this information going, inside of the processor, if it's not turned into heat?
If your computational gates are reversible [1], then in principle, energy is not converted to heat during the computational process, only interconverted between other forms. So, in principle, when you reverse the computation, you recover the entire energy you input into the system.
However, in order to read out the output of computation, or to clear your register to prepare for new computation, you do generate heat energy and that is Landauer's principle.
In other words, you can run a reversible computer back and forth and do as many computations as you want (imagine a perfect ball bouncing in a frictionless environment), as long as you don't read out the results of your computation.
[1] NOT gate is reversible, and you can create reversible versions of AND and OR by adding some wires to store the input.
I was curious about this too. This chip is using adiabatic computing, which means your computations are reversible and therefore don't necessarily generate heat.
I'm having trouble interpreting what exactly that means though.
After reading the output, you run the entire computation backwards (it's reversible after all) until the only state in the system is the original input and initial state.
Then you can change the input to do another calculation.
If implemented "perfectly", heat is produced when changing the input and to read the output.
Reading the output will disturb it so the computation won't perfectly reverse in practice, however if the disturbance is small, most of the energy temporarily held in circuit is supposed to bounce back to the initial state, which can then be reset properly at an appropriately small cost.
Quantum computation is very similar to all of this. It too is reversible, and costs energy to set inputs, read outputs and clear accumulated noise/errors. The connection between reversible computation and quantum computation is quite deep.
It's still getting turned into heat, just much less of it. The theoretical entropy increase required to run a computer is WAY less than current computers (and probably even the one in the article) generate so there is a lot of room to improve.
If it ever gets to home computing, it will get to data center computing far sooner. What does a world look like where data center computing is roughly 100x cheaper than home computing?
Flexible dumb terminals everywhere. But we already have this with things like google stadia. Fast internet becomes more important. Tricks like vs code remote extensions to do realtime rendering locally but bulk compute (compiling in the case) on the server become more common. I don't think any of this results in radical changes from current technology.
Dumb terminals everywhere. A huge upgrade of high-speed infrastructure across the US since everyone will need high throughput and low latency. Subscriptions will arise first, as people fucking love predictable monthly revenue - and by people I mean vulture capitalists, and to a lesser degree, risk-averse entrepreneurs (which is almost an oxymoron...), both of whom you can see I hold in low regard. Get ready for a "$39.99 mo. Office Productivity / Streaming / Web browsing" package", a "$59.99 PrO gAmEr package", and God knows what other kinds of disgusting segmentation.
Someone, somewhere, will adopt a Ting-type model where you pay for your compute per cycle, or per trillion cycles or whatever, with a small connection fee per month. It'll be broken down into some kind of easy-to-understand gibberish bullshit for the normies.
In short, it'll create another circle of Hell for everyone - at least initially.
The great thing about capitalism is that it replaces bad services with good ones. Only exceptions are natural monopolies that the state fails to regulate, which datacenters are not.
The problem is the capital cost of the cryocooler.
The upfront costs of a cryocooler, spread out over the usable lifetime of the cryocooler (they're mechanical, they wear out), vastly exceeds the cost of electricity you save by switching from CMOS to JJs. Yes, I did the math on this. And cryocoolers are not following Moore's Law. Incredibly, they're actually becoming slightly more expensive over time after accounting for inflation. There was a LANL report about this which I'm trying to find, will edit when I find it. The report speculated that it had to do with raw materials depletion.
All of the above I'm quite certain of. I suspect (but am in no way certain) that the energy expended to manufacture a cryocooler also vastly exceeds the energy saved over its expected lifetime as a result of its use. That's just conjecture however, but nobody ever seems to address that point.
I'm one of the authors of the published paper that IEEE Spectrum referred to in the post. First off, thanks for posting! We're so delighted to see our work garner general interest! A few friends and relatives of mine mentioned that they came across my work by chance on Hacker News. I already noticed the excellent questions and excellent responses already provided by the community.
This comment might get buried but I'd just like to mention a few things:
- Indeed, we took into account the additional energy cost of cooling in the "80x" advantage quoted in the article. This is based on a cryocooling efficiency of 1000 W at room temperature per Watt dissipated at cryotemps (4.2 Kelvin). This 1000W/W coefficient is commonly used in the superconductor electronics field. The switching energy of 1.4 zJ per device is quite close to the Landauer limit as mentioned in the comments but this assumes a 4.2 K environment. With cryocooling, the 1000x factor brings it to 1.4 aJ per device. Still not bad compared to SOTA FinFETs (~80x advantage) and we believe we can go even lower with improvement in our technology as well as cryocooling technology. The tables in Section VI of the published paper (open-access btw) goes on to estimate what a supercomputer using our devices might look like using helium referigeration systems commercially available today (which have an even more efficient ~400W/W cooling efficiency). The conclusion: we may easily surpass the US Department of Energy's exascale computing initiative goal of 1 exaFLOPS within a 20-MW power budget, some thing that's been difficult using current tech (although HP/AMD's El Capitan may finally get there, we may be 1-2 orders of magnitude better assuming a similar architecture).
- Quantum computers require very very low temps (0.015 K for IBM vs the 9.3 K for niobium in our devices). With the surge in superconductor-based quantum computing research, we expect new developments in cryocooling tech which would be very helpful for us to reduce the "plug-in" power.
- Our circuits are adiabatic but they're not ideal devices hence we still dissipate a tiny bit of energy. We have ideas to reduce the energy even further through logically and physically reversible computation. The trade-off is more circuit area overhead and generation of "garbage" bits that we have to deal with.
- The study featured only a prototype microprocessor and the main goal was to demonstrate that these AQFP devices can indeed do computation (processing and storage). Through the experience of developing this chip, it helped revealed the practical challenges in scaling up, and our new research directions are aggressively targetting them.
- The circuits are also suitable for the "classical" portion of quantum computing as the controller electronics. The advantage here is we can do classical processing close to the quantum computer chip which can help reduce the cable clutter going in/out of the cryocooling system. The very low-energy dissipation makes it less likely to disturb the qubits as well.
- We also have ideas on how to use the devices to build artificial neurons for AI hardware, and how we can implement hashing accelerators for cryptoprocessing/blockchain. (all in the very early stages)
- Other superconductor electronics showed super fast 700+ GHz gates but the power consumption is through the roof even before taking into account cooling. There are other "SOTA" superconductor chips showing more Josephson junction devices on a chip... many of those are just really long shift-registers that don't do any meaningful computation (useful for yield evaluation though) and don't have the labyrinth of interconnects that a microprocessor has.
- There are many pieces to think about: physics, IC fabrication, analog/digital design, architecture, etc. to make this commercially viable. At the end of the day, we're still working on the tech and trying to improve it, and we hope this study is just the beginning of some thing exciting.
This is an area that's still in its early stage, but yes, it would seem so.
Modules that are a few kilobytes in size have already been tested.
Even taking cryogenic operation into account, memory of this type consumes 10-100x less power than CMOS technology at roughly the same clock speed [1]
This is a very active field of research and there's a plethora of different approaches.
My guess is that 20 years from now there could be three types of computing:
• cryogenic quantum-computing for specialised tasks
• cryogenic ultra-high-performance computing
• high temperature computing (traditional CMOS-based)
with the first two not being available to consumers or small companies. Maybe it's going to be like in the 1970s and early 1980s when mainframes ruled supreme and you would rent these machines or compute time thereon.
People are accustomed to "the cloud" already, so it's not really a regression going back to centralised computing for a bit.
If you compare cpu power usage with ram power usage, you'll see ram is already quite efficient, so even if traditional ram connected to the said microprocessor cannot be brought under the magic of this method, it might work.
(Haven't read the article, or have any expertise in this field, so I might be wrong)
Huh, this seemed a bit too good to be true on first reading. But given that the limits on computing power tend to thermal, and that a superconducting computer presumably wouldn't produce any heat at all, it does kind of make sense.
True, but usually the problem is removing the heat from the chip, not the total amount of heat produced. If the heat is mostly produced by the cooling system then that problem all but goes away.
Not really at data centre scale. Heat directly in the CPU is a limiting factor on how fast an individual chip can go, and at the board level is an issue of getting heat away from the CPU somehow.
But that heat has to go somewhere. When you have rooms full of them the power and cooling issues become key in a way that doesn't matter when it's just a PC in your room.
Sure. But how efficient are they once you include the power used to keep them cold enough to superconduct? I doubt that they're even as efficient as a normal microprocessor would be.
“But even when taking this cooling overhead into account,” says Ayala, “The AQFP is still about 80 times more energy-efficient when compared to the state-of-the-art semiconductor electronic device, [such as] 7-nm FinFET, available today.”
AQFP logic operates adiabatically which limits the clock rate to around 10 GHz in order to remain in the adiabatic regime. The SFQ logic families are non-adiabatic, which means they are capable of running at extremely fast clock rates as high as 770 GHz at the cost of much higher switching energy.
Can't imagine why it would, but the lack of heat makes a 3D cpu much more feasible. So you could take 20 die, make 20 layers, and get radically more transistors per volume.
Maybe I'm misunderstanding but since this 3D CPU would be superconducting, it would conduct electricity without resistance and therefore not generate any heat while in use.
The adiabatic (reversible) computations themselves would be zero-loss, but in order to actually read the result of your computation you need to waste heat.
Maybe there are parts which are not superconducting? Like impurities in the material. So even though the generated heat is like 0.01% of the original, some heat is still generated.
If you went a bit colder you could use superfluid helium and laser cut microchannels through the whole die. You would need to cool it to 2K and every degree at that temp is a fight but once past the transition point it becomes a remarkable thermal conductor and loses all viscosity. It seeps through tiny pores in porcelain like it's a sponge.
Solid idea, but the same thing that makes it work is what makes it impractical. Helium is almost impossible to contain long term, and would have to be replenished regularly.
Presumably 80x less power including cooling means more than 80x less power not including cooling.
I'd think that should be enough to get quite a few layers, sure maybe some minimal space between layers for cooling, but radically less than the non-superconducting designs.
I'm making a naive guess here. No, superconducting transistors are probably harder to create than non super conducting transistors so the limits on die size from defects are even more pronounced and superconducting doesn't change the speed of light for electrons on the chip so it doesn't change the timing issues arising from large dies.
The limits on die size for the competitive consumer chip market are nothing like that of the B2B market. Largest chip ever made was over 40,000 mm^2 [1] compared to Intel's 10900K at ~205 mm^2. In production mainframe chips like IBM's Z15s are on the order of 700mm^2. The fab process has a lot of levers so very low defect rates are possible but not at the scale of a consumer CPU.
Edit: I assume a supercoducting microprocessor would use a strategy similar to the AI monolith in [1]. Just fuse off and route around errors on a contiguous wafer and distribute the computation to exploit the dark zones for heat dissipation.
If implemented using switches based on the Josephson effect like here, then no.
The thickness of the required insulating barrier presents a hard lower limit to the structure size.
The actual value of that limit depends on the material used and the particular implementation of the Josephson junction, of which there seems to be quite a few.
So the limit depends on how thin the barrier can be made.
> Since the MANA microprocessor requires liquid helium-level temperatures, it’s better suited for large-scale computing infrastructures like data centers and supercomputers, where cryogenic cooling systems could be used.
I'd be curious what the energy savings are like at smaller scales -- 80x at data center scales, but how about for a smaller machine, like a PC with phase cooling?
The authors' power comparison is outlined in Section VI of their paper [0] (page 11-12). You might be able to figure out some intermediate scalings from that!
I would expect the cooling to have a scaling advantage-- heat gain is proportional to surface area, but the mount of superconducting mojo you can use is proportional to volume so it should be more energy efficient to build larger devices.
Significantly less the the energy it takes to cool current processors. Nearly five years ago I was talking to a professor who was at a conference on the future of computing and there were folks there from the big 5 and he said that all of them had done the math on how much it cost to run their data centers vs run the cooling for cryogenic super conducting computers, and they all concluded that it was vastly cheaper to run at cryogenic temperatures, at least in part because they would drop the cost of running the cpus essentially to zero because they would all be superconducting.
I have heard this elsewhere, but when I looked into it, it seems like it’s not that big of a challenge compared to the other difficulties of space. More of just a consideration that has to be part of the design. See: https://en.m.wikipedia.org/wiki/Spacecraft_thermal_control
Can anyone give an expert opinion on the difficulty of cooling in space?
The problem is that the act of computation is generating heat, so you can't just insulate your superconducting CPU, you need a way to dissipate the heat it must generate (this applies to all irreversible computations). This is difficult, because the only way to dissipate heat in space is radiation (with conduction usually acting as a middleman), which is constrained by the surface area of your radiator.
So no, it probably wouldn't be any easier in space.
Ideally this server farm would attach to a giant rock somewhere in a stable orbit and drill heat exchange pipes deep into the rock, like how we do for geothermal energy here on earth but in reverse. This whole exercise would require a nuclear reactor to be feasible both economically and engineering wise.
In theory, I think it is possible. The laser light just needs to carry away entropy, which can be achieved by modulating the beam so that it contains information. Of course, you'll deplete your batteries to power the beam, and the laser itself will have inefficiencies. So with current technology, the system would likely generate more heat than it removed. A civilization with really good engineering could probably make it work, though even in that case, it would still use many joules of free (i.e. usable) energy to remove 1 joule of thermal (waste) energy.
A much more efficient method of heat removal would just be to have radiating fins. So you're still converting heat into photons, but you lose much less energy in the process, because the photons you're radiating have a higher entropy.
Definitely possible to put some of the heat into a laser. It's simple to turn a temperature gradient into an electrical potential [0], and if you use that electricity to power a laser it will convert away some of the hat.
I can’t think of a reason it would violate any basic physical laws. Use a peltier cooler in reverse as the transducer from heat to electricity, apply to an appropriately spec’d solid state laser. Surely the devil is in the details somewhere..
This could explain the extreme push to the Cloud, for example with Atlassian who discontinues its Server products entirely, and only keeps Data Center or Cloud versions. It behaves as if personal computing or server rooms in companies won’t be a thing in 2025.
The other strategy that is ultra-efficient is to stop using the net to sell hoards of useless crap that will break the day after the warranty expires and cannot be repaired.
That would save money on the computing power as well as the mining, transportation of raw materials, refining, transportation of refined materials, manufacturing, transportation of finished goods and the whole retail chain.
Unless we achieve room-temperature semiconducting processors, this will only benefit data centres, most of whose power is used to sell stuff. Does anyone actually think that the savings will be passed on to the consumer or that business won't immediately eat up the savings by using eighty times more processing power?
Hey, now we can do eighty times more marketing for the same price!
The landauer limit at 4.2K is 4.019×10^-23 J (joules). So this is only a factor of 38x away from the landauer limit.
reply