Hacker News Reader

Comment by gamepsys | Hacker News Reader

> (writing) an energy per bit of information of 17 μJ [0]

It's interesting to think about data as μJ per bit. The article seems to hint that this is too high for practical use, but I'm honestly clueless how this would compare to other storage mediums. It also makes me wonder what amount of energy premium are we paying for remote data and what amount of energy premium we are paying for wireless transmission.

[0] From their Nature article

View on HN

Not exactly. "dual-beam volumetric nanoscale writing [...] contained two laser beams [...] writing laser beam [...] energy per bit of information of 17 μJ [...] and a [...] deactivating laser beam [...] energy per bit of information of 3.64 mJ".

That brings us to $800/TB to write.

Also, if pixel == bit (I doubt this, due to error coding there should be more than one pixel per bit), the following is also very relevant: "The pixel dwell time (that is, the light exposure time) was 100 ms, which involved 4.2 × 10^6 pulses [...]. [...] took 0.35 ms to move from one pixel to the next [...]"

This would require 25440 years to write 1TB.

Scientists will need to find a way to succeed in writing a bit in less than 4.2 million "attempts" until this technology is commercially interesting. Still, this is an exciting development because it may produce a commercially viable medium if the density ambition is reduced from 200TB/disk to something like 20TB/disk or 2TB/disk.

https://www.wolframalpha.com/input?i=3657+%CE%BCJ+%2F+bit+*+...

https://www.wolframalpha.com/input?i=100.35ms+%2F+bit+*+1+te...

Writing a terabyte of information at 10c/kwh would cost $3.78 [1]. Not nothing, but doesn't seem entirely impractical for all cases.

Heat dissipation might be an issue?

[1] https://www.wolframalpha.com/input?i=17+%CE%BCJ+%2F+bit+*+1+...

Interesting. I think AWS Athena is $5/TB for queries.

I tried some numbers for writing to hard drives (it seems like SSDs are hard drives are not super far-apart on energy per byte written)

And it seems like writing to HDDs is between $.15 to $.56 per TB depending on write speed and watts/time assumptions. It gets tricky to think about because there the sunk cost of powering the drive when idle, and the delta between idle and write power needs, but I think that range is representative for $/TB on a HDD.

Wolfram link: https://www.wolframalpha.com/input?i=20w+*+second%2F100MB+*+...

> writing to HDDs is between $.15 to $.56 per TB

That seems obviously wrong (writing a TB will consume less than 1 kWh), and indeed your link shows 0.56 cents (i.e. $0.0056) per TB.

It is definitely wrong, there’s no way a hard drive that can write a TB in, say, 4 hours would consume anything close to 250 watts.

37.8 kWh per TB, so about 10 kW at a write speed of ~75 MB/s (~3.7 hours for the TB).

That means you'd not only have to deal with the heat dissipation of ~five space heaters at full power, but you'd also need three-phase power to supply it.

Or write slower I guess?

At 75 MB/s, this thing takes 30 days to fill. How much slower would you like?

Who says we need to fill it up in a reasonable time frame? If each disk costs a comparable amount to a dvd (cents) then the expensive part is the electricity (and maybe the equipment to write to them) and it's ok if most of the theoretical storage capacity goes unused. The important metric here for something like data backup is cost/byte to write and cost/byte/month to maintain.

> Who says we need to fill it up in a reasonable time frame?

So you yourself are saying anything longer than a month is unreasonable?

In any case, this lab experiment will need much improvement in speed and energy requirements.

No, I'm suggesting that it could take a decade and still given the numbers in the thread I replied to (which I agree appear not to reflect reality) be a viable product. It never needs to be 'full'.

A few years maybe? The point of large storage mediums is rarely to fill them up as quickly as possible; rather the opposite.

BRB getting a new power supply.

Just do all your disc writes in winter, problem solved.

It seems doubly weird, as blu-rays and other such things they are comparing to aren't written, but pressed as is when sold in large copies.

I get that a lot of uses may be as a hard drive, but imagine ginormous AI model files or some such, pressed and shipped as updates. The energy cost would be far different.

Or maybe Ultra Mega Super HD 4D video.

Or... or... blockchain history for bitcoin in a few years.

Petabyte-sized models will be impractical for some time still. OTOH, 30 years should give us a million fold improvement in capacity, and, with in-memory computing (you really don’t want to move a petabyte) I’d say there is hope.

Also :

> The researchers note that the entire procedure used to create blank discs made using AIE-DDPR films is compatible with conventional DVD mass production and can be completed within 6 minutes. Gu says these new discs may therefore prove to be manufacturable at commercial scales.

So I'm triple-confused ??

> It's interesting to think about data as μJ per bit.

There are physical limits to computing, specifically for Processing and memory density, Processing speed, Communication delays, and Energy supply:

* https://en.wikipedia.org/wiki/Limits_of_computation

Reply on news.ycombinator.com