-2
$\begingroup$

Was looking for materials for the answer to the question

Semiconductor foundries are a thing of the past. Rebuild the computer industry if you can

it turns out that it is seemingly possible to obtain a full stack of technologies for the chips from the 75-80's for a moderate amount of money. So Mr.X from the q decides that his shady moon deals may hit a golden opportunity and using that knowledge how to restore production, he may produce many many equipments for the production of many many processors like Motorola 68000 (specs: 8MHz, 1.5 Watt, 16/32bit, 16MB address space)

Sure it also needs RAM, SSD's, surface mounted components, and other dongles so as additional R&D to fit the purpose - resulting in a unit with 20W power consumption total.

Just vaguely multiplying numbers, to be able to get to some modern metric, to make a comparison performance of that unit is around 100'000 FLOPS, which is somewhat 1/1000'000 of a modern CPU performance.

  • really one can appreciate how much things have changed in 50 years, a minute of appreciation here

So a grand plan is

A many many production nodes on the moon produce and launch many many calculating nodes in space, Lagrange points, or as a ring around the planet.

  • Best kerbals work on orbital and launch mechanics, so there is absolutely no problems here, nothing, I repeat, nothing can go wrong - it all stable and sound, orbiting and being where all that should be.

The ring size is 40'000km in diameter, and a big hole in it of a diameter of 20'000km for the light to the planet to go through, placed in a Lagrange point with 10-second delay.

Data goes in and out thanks to his buddy Elon Tusk's satellite network.

System performance is around 1'922'654 PetaFLOPs, strong and steady, (which translates to 19e15 calculating nodes).

Clearly, it is not your general computer, but it relatively good at crunching numbers in parallel fashion, somewhat similar like GPU's do their things, so rendering CGI videos may be possible if one writes proper software and SDK's

It is not enough to overtake, you know which, cryptocurrency system, which does 40 times more calculations

  • it really should make you appreciate changes with chips production, as few times K1 energy is not enough, by far, to overdo what we do on a fraction of our computation power on modern hardware.

But the system easily trashes top 100 super computers from TOP500 list, combined - 1660 PetaFLOPs.

So on one side, it is not superior computing power, but it is not on the low side as well and can be put to some good use.

The question is:

Assuming the hardware is in place where it should be, the software is running, joules are consumed - what kinds of gains/profits we may expect to have?

Not sure which metric can be used to evaluate the potential of rewards, but is it that golden opportunity Mr.X expects it to be, or is it a bust?

Will Mr.X be able to pay to his buddy Mr. Elon Tusk, launching rockets to the Moon or it won't suffice even feed that Mr.X alone?

What kind of monetary value such system may have?

or if we spin it differently, as it seems to be correlated numbers:

How much money Mr.X earns, or what is his potential net worth?

  • it is not important which metric do you use for evaluation, as long as it is mentioned. An answer is a number in a currency of your choice.

Mr.X does not plan to sell it as a whole but may consider some IPO, and sure collect money for entities who are interested to crunch some numbers, work with big data, train neural networks, etc. So what are reasonable expectations in the case?

Clarifications

  • The world rules and setting etc - @JBH

Your planet, present-day, present economy, present prices, present rules, etc.

Forbes is proud of their Death Star price calculation article, (do not click, find the origin story, but it is a top link in google), but WB is heistant on the question which is not far from Space Based Solar Power, SBSP, on which NASA regularly updates us with studies, since 1978. Nor is it much different from usual, simple or complex, analysis for an application of or investment in, to have an answer - do we invest or not - which are done on regular basis in real world. Which is suitable, I guess, to be answered using science, geography and culture.

Assigning a potencial value is important part in many cases, in cases like this as an example, of worldbuilding process of hard scify. And as already present answers confirm, the problem isn't as simple as just multiply two numbers.

OP is not a regular supercluster user, he does not know how things are rolling in this inustry, hence he seeks the help of WB community to assess described space construction on its certain parameters, instead of going elswhere (which OP will try as weĺl, failing on WB with this Q was sort of out of expectations). Assess in a setting which we are aware of - real world - and translating and incorporating things into a story is OP's work. For those who like handwave - worry not - behind the scene OP will handwave to his heart content, but he would like to have some healty roots for his handwavium.

$\endgroup$
7
  • 1
    $\begingroup$ Building a computer in space sounds like a terrible idea... lots of destructive radiation, extra hard to lose all that heat. Be better to build it on the moon, if you really didn't want to build it on Earth. But anyway... I'm having trouble seeing what you're trying to ask here. Are you asking what the value is of work done by a supercomputer? $\endgroup$ – Starfish Prime Mar 21 at 15:02
  • $\begingroup$ @StarfishPrime you can use price per hour for a supercomputer, not sure how good a match it is, it is up to you, whatever metric you choose - Mr.X is interested in how big a cash flow can be, in his pocket, not counting expenses. $\endgroup$ – MolbOrg Mar 21 at 15:22
  • 2
    $\begingroup$ So freed of all the fluff, you're just asking what a massively parallel supercomputer is worth per hour? $\endgroup$ – Starfish Prime Mar 21 at 15:27
  • $\begingroup$ @StarfishPrime man, you party killer, lol. In a sense, but I can suggest uses that won't be that simple. $\endgroup$ – MolbOrg Mar 21 at 15:29
  • 3
    $\begingroup$ VTC:Too Story Based. How can anybody know what the potential value of the resource can be without you telling us everything possible about both the resource and the world in which the resource exists? Governments, economies, law, religion, demographics... and that's just scratching the surface. In other words, it can have any value you, the author, wants so long as you write the story to accommodate it. I.E., You're not asking about a rule of your world, you're asking us to write your story for you. $\endgroup$ – JBH Mar 21 at 18:37
2
$\begingroup$

Sell Computer Power

As with any other computer cluster, I don't particularly care where the computer is situated as long as access is good and fast enough. You can just sell processing capacity just like most other super computers.

For monetary comparison Amazon cloud computing comes to mind (maybe they are already running the network on satellites, I have no way of knowing.) Finding prices for super computers is a bit harder but should be easy enough to find.

Ball Park Revenue Figure

Because of the question I will give a ballpark figure of what might be expected in revenue. I calculate revenue because profit is something completely different and all depends on your company. In real life it will be probably orders of magnitude off since no effect of increase of supply of computer power, laws and regulations and services have been taken into account

So the cloud computing of Amazone (AWS) is 240 teraflops on average. There revenue of 2020 was 45B dollar ($45x10^9, presuming the American number system). So since you have 8 million the computer power you would expect a revenue of 8 millions times that. Which would give you a revenue of 0.36x10^21 of which is roughly almost a third sextillion. For your information we have roughly 37 trillion dollars of money circulating in the economy.

So seeing that you would create a complete monopoly and a huge increase in supply you can't estimate the revenue based on any current worldly situation.

Note: See remark by Rob Watts for an error in the numbers.

$\endgroup$
10
  • $\begingroup$ "Finding prices for super computers is a bit harder but should be easy enough to find." - please, do, what number can be expected? $\endgroup$ – MolbOrg Mar 21 at 15:27
  • $\begingroup$ Harder but easy enough, eh? ;-) $\endgroup$ – Starfish Prime Mar 21 at 15:28
  • $\begingroup$ I think not, than it depends on the type of super computer (GPU vs CPU and distributed nodes vs single and the amount of memory pipelines) and the use (Scientific, fun, bitcoin mining). Thanks to the Dutch Research Organisation (NWO) I can even use the super computer for free if I have a good research plan. $\endgroup$ – D.J. Klomp Mar 21 at 15:35
  • $\begingroup$ Also with these things the amount of service and support would become important. Being able to run and optimize test jobs before using an orbital super computer would become very valuable. I would feel very bad if I ran an orbital super computer and ran into a null pointer exception at the end of my job in the last data export step. (Never happened to me off course :D) $\endgroup$ – D.J. Klomp Mar 21 at 15:41
  • 1
    $\begingroup$ You've misunderstood the article you linked to. The 240 teraflops the article mentions isn't AWS's total capacity, it's a supercomputer they made with a part of their compute cloud. The article headline might be a little better for real numbers - it cost the researchers \$33K to get a petaflop for 18 hours. If the in-orbit supercomputer used the same pricing that works out to about \$26.6 billion per year for its full capacity. $\endgroup$ – Rob Watts Mar 26 at 16:37
6
$\begingroup$

Short answer: no.

This is based on two things, one apparent from this question and the other inferred from your previous, linked question.

The total processing power of your cloud of computers is pretty good, but the individual power of a given node is pretty gosh darn low by the standards of the last 30 years. Moreover, the nodes are out in space with a small but non-trivial round-trip time limited part by lightspeed, partially by bandwidth limitations induced by your lack of fancy high performance signal processing wizardry and partially by the sheer amount of hassle of distributing work between your hojillion compute nodes, orchestrating the tasks and collating the results.

What this means, then, is that your system is of most use for problems which can be broken down into a stupendous number of relatively simple operations. Individual operations must not form critical parts of the processing pipeline, because being in space means that cosmic radiation is going to scramble all the things in unpredictable ways, either meaning you have to be prepared to have dubious results in your dataset or to dedicate yet more processing power to verifying quality of results and rescheduling work. Between that and the communication delays, you have to be prepared for your work to take quite a while to complete.

What you end up with is something that looks a little bit like old-school batch-mode mainframes of the sort that might be owned and operated by government labs, universities and well-off commercial operations, and not anything that looks like the bulk of stuff that is run "in the cloud" nowadays. Your supercomputer is a very specialist bit of kit.

Now, from your previous question:

The military of each country took matters into its own hands, producing chips in secret, hardened underground bunkers

Clearly, military operations requiring supercomputers will not be requiring your services. Similarly, no big non-military batch-mode tasks run by governments and considered even the slightest bit sensitive will be put in your hands. Whether or not universities would use government computers is a bit of an open question, but if they do any kind of sensitive research (and you can assume that the biggest and richest and most famous universities do) then they might be required to use government approved computers for that, too.

What you're left with, then, are commercial organisations who need a lot of compute but don't do anything that's super sensitive by the standards of their local government, and small universities who aren't important enough to do sensitive work.

The latter can't afford to pay much for your services. The same will be true for many small commercial outfits and amateurs. You're left with a relatively small core of big-spending corporations who need to do a lot of fancy simulation work.

And they'll be asking why they need to use your slightly flaky deep space computer with huge round trip times instead of something hosted a bit closer to home.

How much money Mr.X earns, or what is his potential net worth?

Does he have a monopoly on compute power?

Then he'll have the petrochemical and aerospace industries on a short leash.

If there are alternatives, no-one will be that interested in his offering, or will at least have plenty of room to negotiate fees. He might get rich, but he won't be Bezos rich.

$\endgroup$
6
  • $\begingroup$ Good answer. See, much better than just multiply by hour rate. $\endgroup$ – MolbOrg Mar 21 at 21:18
  • $\begingroup$ This answer is closest to correct IMO but I'll point out that raw FLOPS does not a supercomputer make. Aerospace / petrochem / etc. simulations require high bandwidth, low latency communication of data between nodes to work or processing stalls and this orbiting cloud of processors falls utterly flat in that regard thanks to the speed of light delay. $\endgroup$ – GrumpyYoungMan Mar 21 at 21:42
  • $\begingroup$ @GrumpyYoungMan things aren't that simple, there are enough tasks which can be compartmentalized well enough. But overall the system has properties of a GPU, FPGA, PC - which is quite interesting; when u talk about lag, u forget how slow those things are, worst lag within the system on pair of 200ms, a modern system can do a lot in this time, but they are not. So if we scale it to the performande it may look like delay of 200ns in a modern system, which isn't a small number, but nor is it big. Sure task optimization is required. $\endgroup$ – MolbOrg Mar 21 at 22:15
  • $\begingroup$ The weakness of such a computer is in failure rates. Amazon S3 plans on everything failing. (See the ACM article.) Failure rates are higher in space - both in transient bit changes and in hardware failures. The question is how much will that effect the business plan. What uptime can you plan on? What customers will accept that much down time and errors in the communication of the problems to that computer and in the communications of the results back? $\endgroup$ – David R Mar 22 at 15:54
  • $\begingroup$ @DavidR sure there is all sort of challenges to make the thing work, so as to have errors detection corrections, but nothing new, we have quite developed software approaches. SLA and other things are meaningful only after we recognize use cases, which by the answer is not that many. So there are almost zero real-time services, an hour delay in delivering a result which otherwise they would mulch for a month does not matter. Amazon is more in realtime, I guess. may u give a doi? in general I would not expect the system to have downtimes, it a parallel system, connection including $\endgroup$ – MolbOrg Mar 22 at 18:29
2
$\begingroup$

The market rate for computing power can be found e.g. here (linking to AWS is simply an example and not intended as an endorsement of any specific cloud vendor). Having it in orbit will probably affect the prices, but how?

  • Legitimate customers might be worried about the lack of a jurisdiction. They might not even be able to use it.
  • Criminal customers might see it exactly the other way around.

So the business model is bulletproof hosting on a large scale? Until downlink sites are pressured to cut the net connection. Unless the country where the original launch took place claims jurisdiction and you have only the drawbacks of being in space ...

$\endgroup$
5
  • $\begingroup$ yes, amazon can be a gauge, in some sense, so as it quite complex to get some number estimate from that calculator link, as one has to connect amazon hardware capabilities, which specs they aren't that clear about on their own, if I recall correctly, so as there is too many services they have. So the link in no way an easy number generator which we are looking for. The system nodes aren't that flexible, they are more like a GPU than pc. The convenience of service depends on sat network, so the owner of it guards the key. Bulletproof hosting so as own sat network coming next. $\endgroup$ – MolbOrg Mar 21 at 16:26
  • $\begingroup$ @MolbOrg, at some point this non-virtual private network has to connect to the internet backbones. Those are on earth. You're only delaying the problem with private commo sats. $\endgroup$ – o.m. Mar 21 at 16:28
  • $\begingroup$ "has to connect to the internet backbones" sure, StarLink already has coverage over most of the planet. But to be able to provide hosting one needs to have more or less modern hardware which r&d costs money, this or another way, and supercluster from q may be a tool to leverage some pocket money for that r&d. Connection of hosting cluster with the planet also can't be outsourced to 3rd parties or they hold your business by the neck and forget to be bulletproof. But yeah, sooner the better $\endgroup$ – MolbOrg Mar 21 at 17:02
  • 1
    $\begingroup$ @MolbOrg, do you have a scenario for your game or novel in mind or do you seek fresh answers? Sounds like the former to me. $\endgroup$ – o.m. Mar 21 at 18:30
  • $\begingroup$ yes, I do have a certain development plan in mind, it less game novel related, but I hope it does not matter. I seek for problems, things I missed, some reality verification/checking. So as I'm happy to read some fresh take on a problem, D.J. Klomp answer was initially too shallow, but then it developed into quite a useful one, forcing me to look at specific nuances from a different angle. Use cases u suggested - I thought them all, they are good, in my opinion, what it lacks for me is your personal take on things. $\endgroup$ – MolbOrg Mar 21 at 18:54

Not the answer you're looking for? Browse other questions tagged or ask your own question.