Defining Analog Computer Power

[Jaziel]

TL;DR

Seeking to define power output in analog computing.

According to Google AI, one way computer power is defined is by its ability to process data and perform tasks quickly and efficiently. Processing speed (CPU) is measured in clock speed (GHz) and the number of cores, indicating how many instructions a processor can execute per second.

The above quote applies to digital computers. So how is computer power defined in the case of analog computers? More to the point is there a convenient term that defines differences in power rating between two otherwise identical analog PCs? An example would be for real-time data processing.

Many thanks.

 

[Baluncore]

More to the point is there a convenient term that defines differences in power rating between two otherwise identical analog PCs?

The bandwidth of a computational feedback loop, and the settling time of a digital to analog converter are relevant concepts.

 

[FactChecker]

It is complicated. Just like digital computers, with several indicators of capability (RAM memory size, HDD size, MIPS, clock GHz, etc.), analog computers also have several. Its frequency capability, the size of an analog computer patch board and the number of op-amps are important.
I would say that its frequency capability is the closest thing to a digital computer's CPU clock GHz. But the other measures are important too.

 

[DaveE]

Back when digital computers were just a glint in the eye of Alan Turing and the like, Claude Shannon was measuring information as bits. This is a key part of the definition you seek. I would do like Shannon, ask what is the digital equivalent, and how powerful is that.

Analog systems have errors. Always. A digital computer can calculate 2⋅2=4, but an analog computer can't. It may do something like 2.010⋅2.003=4.042 when you ask for 2⋅2=4. So your first step is to characterize the variability of your machine in ENOB. Also define an acceptable error rate (accuracy).

Next you'll want to measure the speed, which may include bandwidth, sampling rates, anti-aliasing, windows, etc. This often relates to the ENOB step with things like settling times, aliasing, noise, etc.

Finally you need to define the operations being done. There are about a billion benchmarks for digital computer "power". You'll need one too. I would steal one of theirs.

I doubt that there is a single definition.

 

[DaveE]

Also, you might look into digital communication systems. Even though this text is sent, received, and is quintessentially digital. It was sent to you on analog cables/fibers/radios. It might be the simplest example of the digital-analog interface problem. How fast can you move a bit in an analog world?

 

[Rive]

between two otherwise identical analog PCs?

I'm pretty sure no such thing exists.

An example would be for real-time data processing.

For a specific example you can only set up a benchmark based on the requirements for the example itself.
But going general and comparing apples to oranges is hard: to dogs would be even harder.

 

[Jaziel]

Next you'll want to measure the speed, which may include bandwidth, sampling rates, anti-aliasing, windows, etc. This often relates to the ENOB step with things like settling times, aliasing, noise, etc.

So along with frequency capability, can ENOB (Effective Number of Bits), at least as it applies of analog-to-digital converters, be viewed as a reasonable metric to calibrate differences between two comparable analog computers, yes?

At risk of going slightly off-topic, would neuromorphic computing, were it ever to become practical, be regarded as analagous to analog computing in certain respects?

 

[FactChecker]

So along with frequency capability, can ENOB (Effective Number of Bits), at least as it applies of analog-to-digital converters, be viewed as a reasonable metric to calibrate differences between two comparable analog computers, yes?

I would say not. Analog computers may not need DACs or ADCs at all. And a complicating factor is that the conversion between analog and digital signals can depend on the voltage of the analog signal. There are DACs that generate different number of bits of accuracy and work with different analog voltages.

At risk of going slightly off-topic, would neuromorphic computing, were it ever to become practical, be regarded as analagous to analog computing in certain respects?

I am not an expert in this field, but I believe that neuromorphic computing often incorporates analog components. In fact, it has been mentioned as a reason that analog chips will be developed.

 

[FactChecker]

Even now, the field of AI is adding NPUs to the hardware of digital computers (even inexpensive System On a Chip) and NPU TOPs (Trillions of Operations per Second) to the associated metrics. This may become important in the comparison with analog systems.