3 ms·

Using analog computers in today's largest computational challenges (2021) [pdf]

43 points by mindcrime 4y ago

dr_dshiv 4y ago
Cool! Was hoping for some discussion of coupled oscillators as a platform for computing, like suggested (and posthumously patented) by Von Neumann. But maybe here analog means “an analogous system”—like using actual liquids to compute PDE about liquids?
- ktpsns 4y ago
  Coupled oscillator computing is frequently mentioned in one sentence as "quantum computing inspired classical computing" and in fact a powerful method for many computational problems such as the Ising problem. Here is a recent review on that topic: https://www.mdpi.com/2072-666X/13/7/1016 https://www.mdpi.com/2072-666X/13/7/1016
opwieurposiu 4y ago
The authors of this paper are selling fancy German-made analog computers: https://analogparadigm.com/products.html https://analogparadigm.com/products.html
- rapjr9 4y ago
  They are selling a fun less expensive analog computer also: https://the-analog-thing.org/ https://the-analog-thing.org/ Looks like a great toy to integrate with a Eurorack modular music system. Analog computing can cost less energy than digital computing, look at the work of Kofi Odame at Dartmouth College: https://sites.dartmouth.edu/odame/ https://sites.dartmouth.edu/odame/
theamk 4y ago
> An analog computer is limited by its intrinsic accuracy with a typical error delta-f_analog ∼ 10E−(4±1) (averaging over the Analog Paradigm Model-1 and future analog computers on chip). Yes, solving FDEs is useful, but how often are you happy with results equivalent to 12-bit fixed-point integers? And with problems which needs minutes of human attention to change? Most analog computers, including Analog Paradigm Model 1, are programmed by humans dialing in the dozens of potentiometer dials and plugging in jumper wire. And sure, one could replace every potentiomenter with multiplier driven by DAC, but this will dramatically change every figure mentioned in the paper -- power consumption will go up, speed will go down, accuracy might get even lower... The case for analog computer will be even weaker than it is now.
- ktpsns 4y ago
  I think the authors adress all your points in the paper. They both say about needs for resolution in particular applications as well as their plans to replace jumper wires by something on a chip, and they say how going to chip will decrease (not increase) power consumption and increase (not decrease) speed. They don't tell about accuracy on chip, thought. I guess they assume a similar accuracy.

43 points by mindcrime 4y ago

Cool! Was hoping for some discussion of coupled oscillators as a platform for computing, like suggested (and posthumously patented) by Von Neumann. But maybe here analog means “an analogous system”—like using actual liquids to compute PDE about liquids?

ktpsns 4y ago
Coupled oscillator computing is frequently mentioned in one sentence as "quantum computing inspired classical computing" and in fact a powerful method for many computational problems such as the Ising problem. Here is a recent review on that topic: https://www.mdpi.com/2072-666X/13/7/1016 https://www.mdpi.com/2072-666X/13/7/1016

opwieurposiu 4y ago

The authors of this paper are selling fancy German-made analog computers: https://analogparadigm.com/products.html https://analogparadigm.com/products.html

rapjr9 4y ago
They are selling a fun less expensive analog computer also: https://the-analog-thing.org/ https://the-analog-thing.org/ Looks like a great toy to integrate with a Eurorack modular music system. Analog computing can cost less energy than digital computing, look at the work of Kofi Odame at Dartmouth College: https://sites.dartmouth.edu/odame/ https://sites.dartmouth.edu/odame/

theamk 4y ago

> An analog computer is limited by its intrinsic accuracy with a typical error delta-f_analog ∼ 10E−(4±1) (averaging over the Analog Paradigm Model-1 and future analog computers on chip). Yes, solving FDEs is useful, but how often are you happy with results equivalent to 12-bit fixed-point integers? And with problems which needs minutes of human attention to change? Most analog computers, including Analog Paradigm Model 1, are programmed by humans dialing in the dozens of potentiometer dials and plugging in jumper wire. And sure, one could replace every potentiomenter with multiplier driven by DAC, but this will dramatically change every figure mentioned in the paper -- power consumption will go up, speed will go down, accuracy might get even lower... The case for analog computer will be even weaker than it is now.

ktpsns 4y ago
I think the authors adress all your points in the paper. They both say about needs for resolution in particular applications as well as their plans to replace jumper wires by something on a chip, and they say how going to chip will decrease (not increase) power consumption and increase (not decrease) speed. They don't tell about accuracy on chip, thought. I guess they assume a similar accuracy.