Technical questions re military hardware.

Tenshinai wrote:

Yeah, and you'll never get a computer program that can defeat a chess grandmaster.

Noone has managed it yet without cheating to the point where playing is utterly pointless.

When the computer relies on a library with hundreds of TB worth of data about previously played games and variations on them rather than processing power, what´s the point of having the computer there for anything beyond searching the database anyway?

You´re not playing against a computer any longer by that point.

That's most likely the best we can do, just with increasing speed.

Oh yes, never said anything else. Problem is they suck at coming up with good alternatives to look at in the first place.

Garbage in, garbage out.

It does require creativity to create an alternative - which computers are probably never going to develop.

And textev already makes it blatantly clear that processing power alone isn´t good enough anyway.

Why don´t you try a "little" experiment?
Get 2 identical sailboats, have an expert crew on the one, put a computer in charge of the other, then see which one can get through a course faster.

Experiments almost like that has been done. Rarely does the computers fare well.

The one I have in mind right now was on a race track, top race driver vs. computer. Yes, it was a lot simpler due to their being fewer variables.

The human did win, but mostly by being able to corner faster, using tricks that had never been programmed in to the computer due to them being complex.

And still, a 10 year old with a map, pen and ruler can usually solve it better in a tiny fraction of the time.

Someone highly spatially skilled with better tools, well computers simply doesn´t beat them except with absurd once in a million luck.

And once more, 2 dimensional problem vs many dimensional with a lot of extra complexity added, which means computer power required to even reach the current level of "incompetence" goes up literally astronomically higher.

There's the rub though - to beat the computer, you need to have that innate ability to start with. Against a couple of hundred people selected at random? I'd put my money on the average computer time beating the average human time.

As for a more complex navigational system, I'll point you in the direction of Dwarf Fortress. Sure, eventually it will slow down and almost freeze due to the number of routes it needs to calculate, but for something that can run off pretty much any modern system, I think its a good attempt.

I guess a lot depends on what will happen next. I've been investigating memristors for computer use. It's still too early to say for sure, but if they work as predicted, they will allow us to move away from binary computers and thus have a lot more computing power in a smaller space.

Tenshaii wrote:

Why don´t you try a "little" experiment?
Get 2 identical sailboats, have an expert crew on the one, put a computer in charge of the other, then see which one can get through a course faster.

Experiments almost like that has been done. Rarely does the computers fare well.

Did a quick search. Did not find a result. Anybody have it?

With advent of SAR(synthetic aperture radar) able to read wind speed at distance, able to read wave height/direction at distance, the computer will smoke a human race crew in today's world. Especially true on a hydrofoil boat.

Tenshinai wrote:And still, a 10 year old with a map, pen and ruler can usually solve it better in a tiny fraction of the time.

Someone highly spatially skilled with better tools, well computers simply doesn´t beat them except with absurd once in a million luck.

And once more, 2 dimensional problem vs many dimensional with a lot of extra complexity added, which means computer power required to even reach the current level of "incompetence" goes up literally astronomically higher.

There are 24,978 towns in Sweden. How long will your 10 year old take to provide the optimal solution for the traveling salesmen problem for this data set and the proof that it is the correct solution? It took less then 92 CPU (single core) years to solve this in 2004, which means with a single rack you could solve it in under 10 days today assuming no algorithmic improvements and that the 1 core = 1 core. Current trends seem to be using GPUs to attack this, which appear to be considerable faster, one paper suggests that a single GPU is equivalent to 256 CPU cores in this role.

It kind of seems unlikely that Honorverse computers are slower then modern computers and seems a lot more likely they are 2-5+ orders of magnitude faster. So the same scale system would solve this in between 2.4 hours and 1.4 minutes. This is producing a proven best solution, it takes significantly less time to get a close to optimal solution.

So no, you won't be doing hyperspace path determination without using a computer except to prove that it can be done in an emergency, and the path you would produce in reasonable period of time is almost certainly going to be suboptimal.

Only task that a computer will ever be worse at than a human being is a task requiring intuition, or breaks in logic.

lets all break out the difference between software written by idiots and what a computer actually is and what it will never be.

kzt wrote:

Tenshinai wrote:And still, a 10 year old with a map, pen and ruler can usually solve it better in a tiny fraction of the time.

Someone highly spatially skilled with better tools, well computers simply doesn´t beat them except with absurd once in a million luck.

And once more, 2 dimensional problem vs many dimensional with a lot of extra complexity added, which means computer power required to even reach the current level of "incompetence" goes up literally astronomically higher.

There are 24,978 towns in Sweden. How long will your 10 year old take to provide the optimal solution for the traveling salesmen problem for this data set and the proof that it is the correct solution? It took less then 92 CPU (single core) years to solve this in 2004, which means with a single rack you could solve it in under 10 days today assuming no algorithmic improvements and that the 1 core = 1 core. Current trends seem to be using GPUs to attack this, which appear to be considerable faster, one paper suggests that a single GPU is equivalent to 256 CPU cores in this role.

It kind of seems unlikely that Honorverse computers are slower then modern computers and seems a lot more likely they are 2-5+ orders of magnitude faster. So the same scale system would solve this in between 2.4 hours and 1.4 minutes. This is producing a proven best solution, it takes significantly less time to get a close to optimal solution.

So no, you won't be doing hyperspace path determination without using a computer except to prove that it can be done in an emergency, and the path you would produce in reasonable period of time is almost certainly going to be suboptimal.

kzt wrote:It took less then 92 CPU (single core) years to solve this in 2004

Uh, hello? That one used preprogrammed paths. Talk about pointless comparison.

kzt wrote:Current trends seem to be using GPUs to attack this, which appear to be considerable faster, one paper suggests that a single GPU is equivalent to 256 CPU cores in this role.

No/yes/maybe. Generally that exact statement is extremely incorrect, but under specific conditions it may be true.

Depending on what you´re calculating, a "gpu" can be anywhere from a hundred million times faster than a cpu, to a good bit slower.

IF you take a SLIGHTLY more realistic way of looking at it and look at performance per watt or per transistor, you will get somewhat better comparisons.

Generally, if you want FPU calculations, a GPU is better, if you want ALU/Logic/Integer calculations, a cpu is better.

Though for some parts, that is changing thanks to AMD and NVidia recently releasing much improved programmability of GPUs, and especially AMDs GCN and GCN2 GPUs were engineered to hugely improve this capability.
With the addition of the brand new Mantle API, they can do some amazing stuff, but mostly in theory as programming is way behind what is now possible.

kzt wrote:It kind of seems unlikely that Honorverse computers are slower then modern computers and seems a lot more likely they are 2-5+ orders of magnitude faster. So the same scale system would solve this in between 2.4 hours and 1.4 minutes. This is producing a proven best solution, it takes significantly less time to get a close to optimal solution.

And you´re still specifically ignoring textevs about just how problematic it is to do this.

And my previous mention that we are talking 4-dimensional minimum here, with a bunch of complex additions, like variations in hyperspace, vs 2 dimensional simplistic.

You actually expect that transition to be without penalty?

Tenshinai wrote:And you´re still specifically ignoring textevs about just how problematic it is to do this.

And my previous mention that we are talking 4-dimensional minimum here, with a bunch of complex additions, like variations in hyperspace, vs 2 dimensional simplistic.

You actually expect that transition to be without penalty?

To me the text-ev seems more supportive that the defaults built into the computer aren't always what you want.

Shadow of Saganami wrote:Next she punched in a search order, directing the computer to overlay her rough course with the strongest h-space gravity waves and to isolate the wave patterns which would carry them towards Spindle. She also remembered to allow for velocity loss on downward hyper translations to follow a given grav wave. She'd forgotten to do that once in an Academy astrogation problem and wound up adding over sixty hours to the total voyage time she was calculating.
She felt a small trickle of satisfaction as she realized the same thing would have happened here, if she'd simply asked the computers to plot a course along the most powerful gravity waves, because one strong section of them never rose above the Gamma bands, which would have required at least three downward translations. That would not only have cost them over sixty percent of their base velocity at each downward translation, but Hexapuma's maximum apparent velocity would have been far lower in the lower bands, as well

Heck, Helen didn't even ask it to generate a best course, just for the local grav waves so she could manually tweak her initial route.

So sure if you give a computer a stupid command it'll follow it. Tell it to stick to the grav waves as much as possible and it'll do so; transit time be damned.

Or if you tell it to pick the least time course it'll do so, even if that means you're plowing through a rift dead parallel to a nearby 'wave the entire trip. (And just as long as it'd take 1 minute longer to detour over to the wave than you'd save from the higher acceleration it would be "correct" to do so. But 95% of the time that minute isn't one that counted and a normal navigator would at minimum offer the trade-off to their captain)


But neither of those commands are always wrong. A merchant ship more worried about fuel costs than transit time might well want to take longer routes if they minimize the time between waves; or even drop a level or two to keep in wave that's going your way. And a warship in a situation where literally ever second might count would skip the detour and the 'wave; spending the fuel to get there that miniscule bit sooner.