Complexity for computers and the real world

[Kromm]

The definition of hardware Complexity suggests that it's a poor match to "raw power" but a decent fit to number of simultaneous sessions – that's what the game appears to mean by how much software a computer can run at once. Yes, that's bizarre for computers mostly being used by lone adventurers. It fits with a centralized client-server model that might've been common in the 80s and earlier, but not so much today. For that matter, so does the whole "buy the 'terminal' separately" approach in the rules.

The definition of software Complexity looks like it has something to do with demands on memory or processors. It can be forced to work with the previous definition, but that isn't always a straight-up match.

So . . . the "right" question for hardware isn't "How fast?" or "How much memory?", but "How capable of virtualization?" If a computer can support many users on virtual machines, it gets a high Complexity; that's the only way to explain why it should scale up with the size of a mainframe, given that bigger mainframes mostly just support more users.

There should probably be a completely different Speed Rating or whatever for execution time, and that's the thing that goes up sharply enough over time to merit a base-10 log scale. I mean, I just bought a computer that would strain to support multiple users running even simple software, but that can blitz through calculations – especially for rendering graphics – that would've flattened the old mainframes I use to code on that were quite able to support several hundred simultaneous sessions.

There's should also probably be some sort of Capacity Rating, though this may coincide with Complexity for single-user machines.

It looks to me as if Complexity was an oversimplification, and that computers and software should properly have 2-4 stats, not just a single, unified one. Then you could say things like old mainframes had {Users, Speed, Capacity per User} in the {7, 1, 1} range on some log scale, while modern PCs have {1, 7, 3} or whatever. Please don't take those numbers literally!

[johndallman]

Originally Posted by maximara

The M1 is a variant of the CPU found in iPhones and iPads ...

Not so. It uses the same instruction set as the processors in iPads and iPhones, but it's a very different implementation, because it's designed to be capable of using far more energy.

Intel sell processors that are slower and use less power than the ones they sell for desktops and servers. They just aren't very popular or widely known.

[AlexanderHowl]

Originally Posted by maximara

But this makes no sense even by the Complexity guidelines as via emulators I can run far more programs of 1981 Complexity 3 on a modern Complexity 3 computer than I can with regards to modern Complexity 3 programs.

More over per High-Tech, Electricity and Electronics a 1981 Desktop is TL7 not TL8 which kinds of points to a connection between TL and Complexity.

TL8 starts in 1980 according to High-Tech, so a 1981 Desktop is early TL8 while a 2021 Desktop would be late TL8. The fact that there is no mechanical difference between the two is an admitted weakness in the Complexity system, which cannot really account for the fact that a 1981 megacomputer is less capable than a 2021 desktop. However, I think that it would have been worse had they used discrete measures, as they would have likely been quite incorrect.

For example, take Transhuman Space. When it was published, TL9 technology by 2020 was a possibility, if somewhat optimistic. By 2021 though, the technological progression proved to be much too rapid, so the setting probably needs to be pushed forward 50 years, to 2150 AD, to get the same level of changes. After all, laser launch technology is still in its infancy, there is no AIDS vaccine, and teletourism on Mars has not left the drawing board.

[Anders]

Originally Posted by AlexanderHowl

The fact that there is no mechanical difference between the two is an admitted weakness in the Complexity system, which cannot really account for the fact that a 1981 megacomputer is less capable than a 2021 desktop.

But there is a difference. Look at the table on p. 37! 1980 you have early VLSI, which gives -2 to Complexity. 2010 you have advanced VLSI, which gives +2 Complexity. That's a difference of four magnitudes!

[Tyneras]

Originally Posted by Kromm

There's should also probably be some sort of Capacity Rating, though this may coincide with Complexity for single-user machines.

It looks to me as if Complexity was an oversimplification, and that computers and software should properly have 2-4 stats, not just a single, unified one. Then you could say things like old mainframes had {Users, Speed, Capacity per User} in the {7, 1, 1} range on some log scale, while modern PCs have {1, 7, 3} or whatever. Please don't take those numbers literally!

I rather like this approach. I think I'd call users "Functions" instead, I was tempted to call it cores but that gets used in real hardware and would likely cause confusion.

Making Cycle Speed logarithmic would be good, maybe with a bonus program CS based on unused Capacity. The only Complexity 2 program running on a Capacity 3 machine would be screaming fast compared to one with a full Capacity.

Capacity would be a good abstraction of memory, disk space, CPU speed, etc. and any mention of exact program size could be dropped.

The only thing I can think of beyond that that would matter to adventurers would be power requirements.

[whswhs]

When I started work on GURPS High-Tech: Electricity and Electronics, there had been a huge amount of discussion of Complexity scores for computers at various TLs, and what they meant, and how well or poorly they fit real world developments in computer technology. So I decided this was an opportunity to reduce the gap. This had to be an alternate system, as it had multiple boosts to complexity over the 40 years (at that point) of TL8, which would render the entire treatment in GURPS Ultra-Tech incompatible; in fact I explicitly discussed this in E&E, and suggested different ways to address it, from assuming that computer progress was about to stop to requiring a much higher Complexity to emulate a human nervous system.

The basic table of Complexity in E&E is, if I recall, roughly for mid-TL8, circa 1990. It says that a computer of this size is Complexity 1 and a computer of this size is Complexity 5 and so on. But attached to it is a table of modifiers for the processor technology, from mechanical through vacuum tube and transistor up to the extremely high-end VLSI we have now. You have to apply those modifiers to the values in the basic table. That accommodates, for example, the fact that the first high-end computer animation was done in the early 1980s on supercomputers, but by the turn of the century it was being done on much smaller workstations.

As for dimensions of performance, I suggest that for speed, you look at the processor technology; in fact, I gave speed values for processors from mechanical and electromechanical on up. This is in principle independent of Complexity, though it coordinates with size, and the sheer size of processors limits the Complexity you can get at a given weight. But Complexity as such isn't a measure of speed. It's a measure of how much information the computer can have in memory, which is a measure of both "how big a program" and "how many programs at once". (For classic von Neumann architecture machines, even though it appears that the computer is running several or many programs at once, in fact it's running one at a time, but hopping back and forth between them very fast. Things get more interesting with parallel and distributed systems.)

It looks as if computers haven't gained in complexity since, say, 2000. But that's if you look at consumer systems running Word or Excel or various browsers. Those really don't need any more complexity; they respond to it by getting overloaded with features. But where complexity still makes a difference is, on one hand, in a computer's ability to handle things like audio, radio, or video, and on the other, in its ability to run incredible simulations like a single cortical column in a mammalian brain. The latter are intrinsically more complex, whereas word processing just is intrinsically rather simple and doesn't actually draw on a current desktop's full capabilities.

[maximara]

Originally Posted by AlexanderHowl

TL8 starts in 1980 according to High-Tech, so a 1981 Desktop is early TL8 while a 2021 Desktop would be late TL8.

Per GURPS Low-Tech p 7 "Define a society’s TL by the tools and techniques in common use there – not by the calendar date."

Originally Posted by AlexanderHowl

The fact that there is no mechanical difference between the two is an admitted weakness in the Complexity system, which cannot really account for the fact that a 1981 megacomputer is less capable than a 2021 desktop. However, I think that it would have been worse had they used discrete measures, as they would have likely been quite incorrect.

Kromm's idea of a Capacity Rating works far better and is quite doable. Though SSD storage kicks the

Even with Computer-Design Options you can tweak things. Apple's M1 is Fast (+1) with Advanced VLSI (+1) and the developer prototype used an A12x from the iPad Pro (Complexity 2). Therefore the entry level M1 Macs are Complexity 4.

Originally Posted by AlexanderHowl

For example, take Transhuman Space. When it was published, TL9 technology by 2020 was a possibility, if somewhat optimistic. By 2021 though, the technological progression proved to be much too rapid, so the setting probably needs to be pushed forward 50 years, to 2150 AD, to get the same level of changes. After all, laser launch technology is still in its infancy, there is no AIDS vaccine, and teletourism on Mars has not left the drawing board.

Transhuman Space's main book is from Classic when start dates were taken a little more literally. GURPS Ultra-Tech p. 8 has a table for TL 9 and above with Accelerated, Fast, Medium, Slow, and Retarded as the categories.

Originally Posted by johndallman

Not so. It uses the same instruction set as the processors in iPads and iPhones, but it's a very different implementation, because it's designed to be capable of using far more energy.

Actually while the implementation is different it is not so it is capable of using far more energy because the development platform demoed in July was running on A12x. The A12x was straight from an iPad Pro and was two years old ie an old iPad CPU.

[johndallman]

Originally Posted by maximara

Actually while the implementation is different it is not so it is capable of using far more energy because the development platform demoed in July was running on A12x. The A12x was straight from an iPad Pro and was two years old ie an old iPad CPU.

The development platform is not the same thing as the M1 at all. It's a good machine, and I've got a lot of work done on my employer's one, but the M1 is much faster and uses much more power.

[maximara]

Originally Posted by johndallman

The development platform is not the same thing as the M1 at all. It's a good machine, and I've got a lot of work done on my employer's one, but the M1 is much faster and uses much more power.

An M1 is rated at 14 Watts while the Apple A12X Bionic is 15 W so the M1 uses less power then the A12X developer computer Apple showed off in July.

More over Apple actually down clocks the M1 to only 10 watts: "At just 10 watts (the thermal envelope of a MacBook Air), M1 delivers up to 2x the CPU performance of the PC chip."

So as you pointed out you have a chip down clocked to 10 watts out performing a 15 watt chip. Hardly "uses much more power" when in reality no matter how you slice it the M1 uses less.

[Anthony]

The basic use of complexity is for determining how many programs a computer can run. Now, newer computers can run more simultaneous applications, but not by nearly as much as the amount of memory and CPU power has increased, because newer versions of software are just plain bigger than older.

Consider text editors. The first versions of 'vi' (a text editor still used today) had difficulty running on a PDP 11/70, with 4 megabytes of ram, so it probably took up 2mb or so. A modern vi on a blank document is 11MB. Microsoft word, with a blank document, loads to a size of 64MB. Of course, a modern computer can run hundreds of instances, but not by as much as hardware has improved.

Some of this is expansion in capacity. Some of this is changes in programming style and tools; as computers become more powerful, it simply doesn't make economic sense to optimize them for the limitations of less developed computers (spending an engineer-week to save a megabyte on a program might have been worthwhile in 1979; in 2020, unless you're running thousands of instances, it isn't).