Why digital and not analog?

[rafe_28]

well... this is something i thought about the other day. why are PC's controlled by digital signals mainly. i am not talking about cables. like the pc components. MB, cpu, you name it. If you would run it analog you could have a wave length. lets say just 1-1000Hz. and you run those at 3000MHz. So you could build the instruction sets not with 1 or 0 but 1, 2, 3, 4... 999, 1000MHz... PER CLOCK. wouuldnt taht massively increase the performance and the ammount of numbers and calculations a PC can crunch? Like i couldnt imagine why not to do it except having to engineer more complicated instruction sets blah blah blah. but what would be the technical, the hardware, issues there to why it would not work? i have very little knowledge in curcuits

[SIERRA880]

I had a computer science class where the teacher said that analog computer existed but not for long. Unfortunately I can't remember exactly why :/

But my guess is because analog signal is not as "stable" as digital signal. That's why logic gate have threshold to avoid changing state if there's a perturbation on the wire.

[Hairless Monkey Boy]

The 0s and 1s in a computer are fundamentally analog signals already. Without going too much into it, the reason why this is not possible and/or is not done comes down to signal integrity and errors.

[Sauron]

8 minutes ago, rafe_28 said:

If you would run it analog you could have a wave length. lets say just 1-1000Hz. and you run those at 3000MHz. So you could build the instruction sets not with 1 or 0 but 1, 2, 3, 4... 999, 1000MHz... PER CLOCK.

And how do you propose we produce and, more importantly, control that oscillation?

 

Transistors can be made to represent way more values than just 1 or 0 (we differentiate based on a voltage threshold) but it gets significantly harder to accurately set them to each value the more intermediate steps you add. It's simpler and more efficient to just shrink and add more transistors, which doubles the amount of distinct values you can represent for each transistor you add.

[Eigenvektor]

23 minutes ago, rafe_28 said:

So you could build the instruction sets not with 1 or 0 but 1, 2, 3, 4... 999, 1000MHz... PER CLOCK.

Discrete steps like this are still digital, not analog. Analog computers have no discrete steps at all and no frequency you're running at either. See: Analog computer

 

The main reason is that representing 1 and 0 is easy. No (or very low) voltage is zero. A high enough voltage is one. The more different values you want to represent, the more difficult it gets to distinguish them, if you also want to keep voltages low.

 

Also, more complexity doesn't necessarily mean more speed. You might not be able to reach such high clock rates anymore. So a "bit" might now have 20 instead of 2 states, but this might possibly mean due to error correction etc, your clock is reduced by a factor of twenty as well.

 

Take SATA for example: Serial ATA, which replaced Parallel ATA. Due to the reduced complexity it can reach much higher speeds than parallel ATA, even though you'd think parallel should be faster, right?

[Moonzy]

imagine yourself yelling from one mountain to another person on the next mountain

it's easy to determine if he yelled or not, but not so easy to tell how loud did he yell due to fluctuations in air density and what not

so 1 and 0 is more stable than if you try to detect more levels

 

iirc G6x or G5x has 4 levels of voltage? so it can transmit 00 01 10 or 11 in one pulse

 

say if you want to transmit 8 levels of voltages, it can transmit 3 bits of data per pulse

only 1 bit of data more, but way more voltage levels, thus decreasing stability and lowering clock speeds

[asquirrel]

The reason why is best explained this way:

 

It is easy to build an analog computer that poos all over a digital one at a specific task. But for the set of all tasks, you don't need a computer that is tuned to do any one thing perfectly, you need a system that is designed to a specification and consistently hits that specification. In the case of digital vs analog, that spec is that 'data start' and 'data stop' are known with digital (rising and/or falling edges of the signal) but with analog you have a basically infinite number of ways that a signal may be interpreted.

 

When it comes to mass production, hitting a specification for an infinite possibility set is problematic at best. So you end up defining a specification that is a subset of infinity, but then your analog computer is just a more complex solution to the exact same problem digital can solve with a LOT less work. So digital ends up winning because it's easier to manufacture at scale, and easier to design for in general terms (IE: every design hits the same spec, so a design that works on one system will work on another).

 

You can build analog systems to a spec as well, but it's harder than building to a spec with digital, so why bother? And thus, digital took over the world. Not because it's 'better', but because it's easier.

[hftvhftv]

If you would like to go deep into this subject quantum computing is what you're looking for as far as why everything is digital. If you look into the history of computer science you'll see that almost all computing devices are "Turing Complete" or otherwise could have any task done on a Turing Machine, which is an abstract device that can compute using an infinitely long tape that is able to write only 1s and 0s. CPUs operate nowadays using transistors that turn on and off with various levels of memory all the way from registers on the CPU to hard drive storage. Eventually this will be replaced but I couldn't give you a timeline as to when.

[lordsmurf]

On 4/6/2021 at 12:42 PM, HairlessMonkeyBoy said:

The 0s and 1s in a computer are fundamentally analog signals already. Without going too much into it, the reason why this is not possible and/or is not done comes down to signal integrity and errors.

^ This.

 

Digital is precise. Very precise.

Analog is wibbly-wobbly, not precise, and devices must be very forgiving of the rampant errors present.

 

Analog VHS is a good real-world for you to see this yourself. Trying to capture VHS on a computer reveals all the warts of the format, and the computer will almost always reject the raw signal without some (timebase) correction.

[Mark Kaine]

Uh, some interesting answers that are probably not wrong but I always thought it's mostly because analog computers would be rather huge, like room sized or bigger with modern instruction sets and functions,  ie you can't make analog circuits small enough to be feasible,  and it would also be prohibitively expensive.  Same with radios,  stereos,  tvs,  etc.  Those could be easily made analog but making them mostly digital is just more efficient? 

 

 

PS: Im pretty sure those were analog? 

 

https://www.computerhistory.org/revolution/birth-of-the-computer/4/78

 

[Laborant]

The transistor alone is extremely cheap, fast, easy to understand and (with nowadays technology) easy to shrink to these arbitrary small nanometer-sizes. With a transistor, you have a basic "measuring device" that basically has an input and an output. 

This "measuring device" looks if the input voltage is above a certain threshold (defined by the chemistry of the junction chemistry and possibly geometry) and sets its output to "1".

Analog circuits need to distinguish not only "is the voltage there or not", but they have to decode the actual voltage level. The fastest analog couplers sip quite a large amount of current. Analog technology is kinda coming to the consumer electronics in the form of 10gig Ethernet and PCIe 6 - these use a "PAM4" modulation, that distinguishes four different voltage levels and therefore sends two bit per transfer. And boy, these Ethernet-Chips are getting HOT. 

For certain applications, an analog "computer" can be built to follow a certain model - for example to solve differential equations, that might end in a numeric catastrophe when solved numerically. To be clear, these computers are not certainly working with continuous signals (waveform-electricity) but form an "analogy" to a model. The idea of an analog computer is not based on "fixed structure and flexible code" but "flexible structure and no code". Meaning, the structure itself is the program. 
But these computers are by far not "multi purpose" but solve ONE TYPE of differential equations depending on their current configuration/structure - and no, Crysis is not a differential equation.

 

[Mark Kaine]

6 hours ago, Laborant said:

Crysis is not a differential equation.

Dreams shattered...

[David82]

On 4/6/2021 at 6:39 PM, SIERRA880 said:

I had a computer science class where the teacher said that analog computer existed but not for long. Unfortunately I can't remember exactly why

But my guess is because analog signal is not as "stable" as digital signal. That's why logic gate have threshold to avoid changing state if there's a perturbation on the wire.

There is an analogue computer (I believe one of the first) called "baby" on display at the museum of science and industry in manchester. 

 

It's pretty interesting, they had loads on info on in on their website.

[rafe_28]

On 4/9/2021 at 3:40 PM, Laborant said:

The transistor alone is extremely cheap, fast, easy to understand and (with nowadays technology) easy to shrink to these arbitrary small nanometer-sizes. With a transistor, you have a basic "measuring device" that basically has an input and an output. 

This "measuring device" looks if the input voltage is above a certain threshold (defined by the chemistry of the junction chemistry and possibly geometry) and sets its output to "1".

Analog circuits need to distinguish not only "is the voltage there or not", but they have to decode the actual voltage level. The fastest analog couplers sip quite a large amount of current. Analog technology is kinda coming to the consumer electronics in the form of 10gig Ethernet and PCIe 6 - these use a "PAM4" modulation, that distinguishes four different voltage levels and therefore sends two bit per transfer. And boy, these Ethernet-Chips are getting HOT. 

For certain applications, an analog "computer" can be built to follow a certain model - for example to solve differential equations, that might end in a numeric catastrophe when solved numerically. To be clear, these computers are not certainly working with continuous signals (waveform-electricity) but form an "analogy" to a model. The idea of an analog computer is not based on "fixed structure and flexible code" but "flexible structure and no code". Meaning, the structure itself is the program. 
But these computers are by far not "multi purpose" but solve ONE TYPE of differential equations depending on their current configuration/structure - and no, Crysis is not a differential equation.

 

ty this really cleared up things the most for me