What can you do with galaxy sized computers?

[wumpus]

3 hours ago, kerbiloid said:

Almost nothing due to the 100000-years lightspeed delay between its memory registers.

Except that this can be easily falsified by asking "how big are the computers on the top 500 list"?  Answer: long enough that they are have "roughly 1us latency between nodes", but you better believe that they are running faster than 1MHz (note, in case I am wrong, I'm pretty sure it is impossible to build a top 500 compute with less than 10ns latency (a meter or two) which would give you 100MHz.  They run well over 1GHz and have some heft latency).

This means that our galaxy-sized "computer" is a cluster of much smaller nodes.  For the last decade or so, the basic Intel i3/5/7 core has been roughly the fastest possible single core.  I'd assume that there would be some similar value (but wildly different if you had a 3d computer, which would presumably involve reversible logic (for efficiency sake) and who knows whatever else.  But it would still be vanishingly small and talk to all other nodes (and presumably the nearest more often).

Throwing in some numbers:

Size of computer: 100,000 light years (size of milky way):

Number of nodes 10^100 (actually its probably a bit under 10^90, but keep the numbers round)

Node to node communication: ~50,000 year latency

Break the thing down into 10^50 sub-networks, each containing 10^50 computers:

Node to node communications latency: millisecond?  Not sure I have the calculations right.  But the jump is huge.  Note that millisecond latency is typically only good for "embarrassingly parallel" algorithms on modern Earth (I suspect that most of the machines with high protein folding scores have [tens of?] millisecond latency), and probably more so on a galaxy-sized computer.

break each subnode down to 10^25 computers each containing 10^25 computers:

Node to node communications latency is effectively zero.  Probably not good  enough for register to register communications, but separate threads don't notice the latency.  Note that there exist calculations based on energy levels and quantum basics that will compute the bounds of power of these nodes, but they are unimaginably powerful compared to modern computers and are presumably have similar cross-sectional area (although our nodes are 3d).  Note that my original 10^100 becomes an obvious problem here, I'm pretty sure that the cores become smaller than protons...  It really doesn't change the overall concept that much.

So the actual answer is "a galaxy sized computer can calculate anything that can be calculated in hundreds [that's roughly all the time the universe is good for] of steps via (nearly arbitrarily powerful computers) that can each calculate each step in infinitesimally small pieces (and then send arbitrarily large messages at unbearable latency between each other)".  This won't be many problems, and you can expect the rest of the network to be busy serving cat videos and porn.

[kerbiloid]

8 minutes ago, wumpus said:

This means that our galaxy-sized "computer" is a cluster of much smaller nodes.

Yes. Smaller nodes, connected with dial-up (let's be optimistic).

All you get as a result - a zillion of hanged nodes awaiting response from other nodes.

 

 

[YNM]

The slowest computer ever. I mean, what, the data takes a few years just to go from ROM to RAM ?

Still the same if it were a bunch of separate computers - the ones soo far away would need to transmit that data to you (or, the reverse, input from you to the computers) in the scale of thousands of years, even more.

Edited April 22, 2016 by YNM

[KAL 9000]

Ok, the previous solar-system sized computer, with several hours of delay between nodes, slowly figured out how it actually IS possible to use quantum entanglement for instantaneous communication, something our feeble human brains got completely stumped on. 

P.S. KSP with ALL the mods? Sign me up! But how would mods like Alternis and New Horizons work together?

[wumpus]

1 hour ago, kerbiloid said:

Yes. Smaller nodes, connected with dial-up (let's be optimistic).

All you get as a result - a zillion of hanged nodes awaiting response from other nodes.

Not dialup: talking to arbitrary nodes in a galaxy sized computer means 50k year latency.  You only get hundreds of messages back and forth before the universe ends.

Local messages (presumably smaller than solar systems) might still have dialup latencies but bandwidth could well be in the "FedEx 737 full of flash memory" bandwidth, with sub-sub-systems being limited by technology and quantum limits, not speed of light.

Amdahl's law comes down pretty hard.  Since the compute/latency is essentially infinite, *any* serial parts stop you dead.  Note that even in the sub-sub-networks, you have essentially infinite compute/latency&bandwidth and need almost the same level of parallelism that the "100 messages: *ever*" parts need.

[Nuke]

compute things very slowly. seriously the latency in the cpu would be atrocious.