Hacker News Reader

> This is why technology is brilliant. It allows us to do much much more with the same resources. We should want more technological innovation.

But that's not what happens in practice. There is this paradox, the name of which I forget, which says that the more efficiently we use a resource, the more we is it in absolute terms. We never really freed up any leisure time when we introduced household appliances or when productivity went up. Cars became both more fuel-efficient but also heavier, so we burn more fuel. Smartphones have gotten really good at managing power, making them a viable choice for plenty of activities throughout the day, thus increasing their total energy consumption.

Personally, I just don't believe more efficient tech will help reduce our power consumption. It will require either a completely different tech that doesn't require electricity or fuel to run, or a change in societal habits.

> There is this paradox, the name of which I forget

It's called the Rebound Effect: https://en.wikipedia.org/wiki/Rebound_effect_(conservation)

Eh, I’m not so convinced that the relationship is direct - Usage is going up and power efficiency is also going up, but I’m not convinced the two are necessarily linked.

I spend roughly the same amount of time at the computer as I did 20 years ago, but my old desktop PC and CRT monitor was certainly not as power efficient as my (bigger) current monitor and Mac mini.

"I spend roughly the same amount of time..." the paradox works if you define usage as energy consumption, not time spent.

20 years ago, you were not streaming 1080p videos or downloading 50 GB video games. Even kids do that on a regular basis now and a bigger share of the population worldwide has access to similar services

No, but I was torrenting those movies to my seedbox which was on 24/7.

IMO I don’t think usage scales to power consumption - this sounds like correlation vs causation to me. It sounds like use has also increased while efficiency has increased, not one because of the other (ie I would personally still be streaming the same number of movies if my AppleTV used twice the power, and if they release a new version that doubles the energy efficiency that won’t cause me to watch more)

"In economics, the Jevons paradox occurs when technological progress or government policy increases the efficiency with which a resource is used (reducing the amount necessary for any one use), but the rate of consumption of that resource rises due to increasing demand."

https://en.wikipedia.org/wiki/Jevons_paradox

It is interesting they consider it a paradox. When the demand curves show it. As making something more efficient moves the supply curve around (and moves where MR=MC is). Or in econ 250 class speak 'shift the supply curve right'. I think the 'paradox' comes in where they do not consider there is more demand on the other side of lower prices? Not all goods goto infinity on cost vs demand curve. But some sure act like it.

I guess the "paradox" or counterintuitive part comes in situations when the price elasticities are such that a 10% efficiency gain (or price cut) would result in >10% increase in demand and hence in final spending.

Another part may be that non-economists (and even lots of students who succesfully passed econ 101) don't think about shifting demand or supply curves. In my experience, most people who remember the textbook supply-demand curves only think about moving along the curves (which makes the classic diagrams pretty crappy pedagogical devices IMO).

agreed. 1 to 1 would be an interesting result and probably a linear set of curves. But some curves look more like log scales so the effect depending on the slope of the curve could be wildly more or less.

I agree the classic diagrams are kind of crappy. As usually they are very simple just for demonstration. But to make both of those curves is usually more like a N dimensional curve. But remembering back on my econ classes I took they were excited about point slope formulas and maybe a mix of linear algebra in the advanced classes. Which was amusing from the CS/Math background which was my major.

In urban planning there's the 'law of induced demand' that has parallels to this and shows the paradoxical nature. When a road starts to see congestion, the reaction in the past has been to add lanes, widen the road, and/or increase the speed limit to get more cars through faster. This has its intended effect in the short term, but makes the road more attractive to drivers. More drivers start using the particular road, exceeding the planned capacity and in the average case causing worst congestion than existed before.

I'm not sure about your general point - people definitely use more things as they become cheaper, which is a good thing IMO, but you're right that this means power consumption doesn't necessarily go down with lowering costs.

But just to add another dimension to the discussion, there's an interesting book I've heard about, the thesis of which is basically that technology has made a lot of material goods much more efficient. E.g. they give the example of aluminum soda cans - nowadays, the cans themselves use much less material than 50 years ago - using modern production technology, using CAD systems to design them better, etc. has led to a lot less material wastage (I think it was like 25% or something like that, but I'm not sure).

And in general, from the point of view of the companies selling material goods, the less material and more efficiently produced materials the better, since that's a direct cost, meaning they are in general working to make everything a lot more efficient.

With a lot of these arguments (including for example the Unabomber manifesto), I wonder if we're not projecting our own behavior on technology, because we see the effects so easily when looking outward rather than inward.

Technology just is. How it's used makes the difference, and that stems from our way of living in the world. Any tool can be used for both "good" and "bad", however you define those.

I think we tend to prioritize our own desires above anything else, because "someone else" might do that and gain an advantage over "us". We're largely stuck in survival mode, rather than an abundance mindset where we take a step back and reflect on what the goal is, and what is needed to achieve that goal.

Significantly, shifting the image calculations from the server to the client means the work is duplicated for every client. While it saves energy for the server, it increases the overall energy consumption required to deliver the content.

Apparently the mix-blend-mode for images on the client is noticable performance-wise: https://github.com/lowtechmag/solar/issues/6

Can one client do the work and post back the final version to the server so the server sends that version to all the other clients?

Don't trust the client with that, or everybody is getting goatse'd.

A. Cryptographically verify the final version

B. Schedule the work to be performed on multiple clients in case some fail

> A. Cryptographically verify the final version

How?

False. Energetic costs of distributing duplicate information is significantly more expensive than local computation. Eg: measuring joules spent for decompression vs transmission is orders of magnitude over the n duplication.

But surely the orders of magnitude thing factors in when you can have orders of magnitude more clients than servers? I think you'd need to run more specific numbers to lean either way on this.

You seem to not realize that moving the calculation client-side means to send a JavaScript implementation, which for all sensible implementations would be significantly larger than the pre-computed result.

The way the image is colored on the client is just some CSS (mix-blend-mode). Check it in the inspector.

No useless JavaScript in this case.

Check out the upcoming ATX power consumption and compare it to what we're using currently...

What you said was true for a very long time, but Moore's law has slowed down a lot over the last 5yrs and power consumption has increased significantly, and is about to become unreasonable in my opinion.

Maximum power draw of 2.4 kw for about to 10% of the uptime is ... sadly going to be reality soon.

> power consumption has increased significantly, and is about to become unreasonable in my opinion

It is not the first time it has happened, and then went down again. It's just one of the usual "cycles".

I don't think we've went through a circle of more then doubling the power consumption before, but I've only been paying attention to it since shortly before 2010, as I was using prebuilt systems before that.

Maybe you're right and more then doubling the power consumption is a normal thing that just occasionally happens.

Precisely it was with Core that I think the previous "nuclear power CPUs" cycle ended.

I feel like it's a very useful metaphor for the work ahead in dealing with energy production, consumption, storage, efficiency, etc.

Layer 1: We must think of clever solutions to reduce our energy footprint where possible and shifting typical energy uses around is one way to do so

Layer 2: But we must also be conscious and aware of economies of scale, and shifting the burden of energy usage from one central server to N clients is ultimately not a good answer for NET energy consumption.

Layer 3: But for some constrained resources, that might be worthwhile, depending on the magnitude: The incremental cost of extra rendering on the client is a rounding error. But on a solar-powered low-energy server, it MIGHT be meaningful, and be the difference between something working and not.

This is a lens we must apply to everything:

* Logging * Battery technology * Buying locally vs economies of scale * International shipping * Container ships * etc.

I'd rather avoid all these layers and just build so much nuclear we get to a future where electricity is too cheap to meter.

Current world power consumption is about 15 terawatts. That's 15,000 gigawatts. [0] One gigawatt takes about 1 nuke. So, 15,000 nukes. Round that down to 10,000 nukes; ignore demand from new electric cars and cooling, and required downtime.

On a very good day, a new nuke rounds down to $10 billion. 10^4 x 10^10 = $ 100 trillion. Not to mention mining, operating, disposal, and environmental costs. Or tsunamis, or profits, or interest.

In other units (Q = 10^18 joules) Total world energy consumption per year: 400 Q. Total solar power striking the Earth each year: more than 4 x 10^6 Q.[1] Catch 1/10,000 of that.

[0] https://science.howstuffworks.com/environmental/green-scienc...

[1] http://sites.science.oregonstate.edu/~hetheriw/energy/topics...

But isn't one of the problems of global warming (ie where humans have a "future") the amount of waste heat we are producing?

https://environmentalsystemsresearch.springeropen.com/articl...

> This essentially shifts some of the burden of computation from the server (PNG compression) to the browser (dithering interpretation). This may save some energy, or it may increase energy as most personal computer processors are much less efficient than server processors, and don't benefit from energy savings from caching.

I believe doing image color setting is already eaten up inside the cost of doing png decompression: the client is already doing image manipulation, this is just doing an additional multiplication, so i believe it is minimal additional cost on the client (i may be wrong tho, never wrote a png decoder).

> I think the implication is electricity consumption will also double roughly every 2 years, but Moore's law actually operates on approx. same timeline, so traffic can continue doubling at this rate without an increase in energy consumption.

We're not doing more with the same amount of energy, we're doing much more with more energy. This is always the same energy efficiency fallacy: being more efficient (in percent "out/in") usually requires having bigger "in" (bigger scale). And you know, the thing that capitalistic companies always seek to get bigger no matter what. There's no incentive to maintain you current consumption. See https://solar.lowtechmagazine.com/2018/01/bedazzled-by-energ... for more details by same source.

I thought Moore’s Law is about the density of components on an IC—what about it suggests a constant energy usage of IC as it gets denser?

I believe ICs are limited by how fast you can cool them. All that energy goes into heat eventually, so temperature is going to depend only on total energy / area

Digression: Is it true that every single watt my CPU/server/electronics consume is ultimately turned into heat?

To put it another way, does a server under a constant 500W produce the exact same amount of heat as a dumb 500W electric space heater? Excluding things like the tiny amount of energy that gets sent out on CAT5 cables, which I believe still gets turned into heat just elsewhere.

I think this is true, but it’s still counterintuitive. Makes me think there’s something more useful we could make electric space heaters do with their energy and still produce the same amount of heat.

The only other places the energy could go are small - LED light, radio waves, sound. Almost all of it will go to heat.

And all of those things will go to heat eventually too, as will all other energy!

> something more useful...

Ha! A space heater with bitcoin mining coils. Call them "bitcoils".

I'm sure you've heard the old joke about gamer college students using their fancy graphics cards to heat their rooms. :)

I did this last winter. Mining 24/7 on my personal rig noticeably took a bit of load off the heater and made a modest amount of ethereum. I stopped once the weather warmed as it seemed like a waste of energy.

I did the same, then my graphics card died and it was a net loss :-(

How were your electric bills?

Not appreciably different, that was the whole logic behind my thinking. In my case I have electric heaters, so doesn't make a difference compared to a gpu -- they're both effectively 100% thermally efficient.

In short, yes. Exactly (well minus a few used to spin fans and hard drives). Here's one source from a quick search:

https://www.pugetsystems.com/labs/articles/Gaming-PC-vs-Spac...

I remember seeing a paper referenced here once explaining why that energy loss is actually necessary, but I don't seem to have it saved anymore.

By a paper, do you mean the laws of thermodynamics?

The fans also eventually also dissipate their energy as heat when they push air around.

There is an alternative heater design though for the same power, which is a heat pump (the same as an AC, run in reverse), which can be much more efficient at heating (down to -5F outside, or so, when they may freeze over and stop working entirely)

> I remember seeing a paper referenced here once explaining why that energy loss is actually necessary, but I don't seem to have it saved anymore.

Taking a stab at explaining the energy loss…

When computing you are reducing local entropy (e.g. achieving a certain pattern of electric charges in chips and a certain pattern of pixels on the screen) while still increasing global entropy (the second law of thermodynamic). Heat dissipation from the local system to its surroundings accounts for the difference.

This is analogous to Schrodinger’s description of life where life forms increase the entropy of their surroundings to maintain their own low entropy.

The spinning fans use energy to move air, a (slightly) viscous fluid, which eventually encounters turbulence and friction and turns its kinetic energy into heat. Similar to discs that use energy to overcome friction in bearings and generate noise, which is a kinetic energy in the form of vibrations that also eventually dissipates into heat.

> I think this is true, but it’s still counterintuitive. Makes me think there’s something more useful we could make electric space heaters do with their energy and still produce the same amount of heat.

Indeed: people have experimented with mining during the winter to recoup electric heating costs, which makes sense but only if you ignore the (expensive and getting more expensive by the day) cost of procuring the original mining hardware (aka GPU).

Hardware is a high cost if you're using current gen mining hardware to be competitive. If power is essentially "free" (since you'd use it anyway for heat), then even very old mining hardware would have a net positive benefit for you. It might only be worth a cup of coffee a week, but it is "free" money you can collect...

It still depends on what your alternative is. Much of the world does not use electricity for heat; in the United States, electricity is rarely used for heat in in the parts of the country that traditionally get brutal winters. It's cheaper for me to heat my home via the forced-air natural gas furnace (even though it is a less efficient fuel-to-heat process as compared to resistance heaters) given the disparity in pricing between residential electric and natural gas costs. I tried mining ethereum for a month last winter and I don't think that I broke even (despite using a relatively recent RTX 2080).

This property is called Dennard scaling. It mostly does not apply to the latest few generations due to increasing static power (leakage current).

https://en.m.wikipedia.org/wiki/Dennard_scaling

Fascinating, thank you

The state of Missouri pushes SSN sorting to the browser. I’ll allow it.

Don't forget the reduced bandwidth usage- the energy used to transport the data across the entire internet.

It seems like we have tons of room to make the internet more efficient even without Moore’s law - how many backends and clients are written in scripting languages (PhP, Node, Python, Ruby…) without a whole lot of thought put into energy use (even if they are carefully designed to be fast or “efficient” by some metrics)?

And how much of internet energy use is made of serving ads? It seems to be a growing share, and ads increasingly want to stream video, which puts things like PNG decoding & dithering in the margins of overall energy use.

> but Moore's law actually operates on approx. same timeline, so traffic can continue doubling at this rate without an increase in energy consumption

Except that data bloat grows at a steady rate, and you only take advantage of Moore once you discard your old equipment and replace it with newer silicon. That generates ewaste, which is counterproductive to ecologically aware projects.

While I agree with most of your points, I think that these dithering images use less client-side power than a regular image compression (e.g. JPEG or PNG). The dithered images are smaller (according to them), and I don't see why they would use more CPU per byte to decode than JPEG.

> The dithered images are smaller (according to them), and I don't see why they would use more CPU per byte to decode than JPEG.

Just try substituting something else in that statement and see if it makes sense. “The lzma-compressed files are smaller than the raw test files, so opening them in an editor shouldn’t use more cpu” or “HEVC-compressed videos are smaller than DivX videos, so shouldn’t they be more efficient to decode and play (sans hardware acceleration)?”

The examples you give compare new/good compression (lzma, HVEC) with old/bad compression (uncompressed, DivX). Yes the new/good compression uses more CPU per byte.

I see no reason to believe their website uses a compression scheme that uses more CPU per byte than PNG or JPEG. I don't think they're using anything advanced. Actually I just checked, and their website is using PNG.

That is entirely forgetting that you throw away and replace hardware in that timeline - what happens to the old hardware?